Method and system for optimizing visual platform operation service combined with multi-modal large model

Abstract

The application provides a kind of visual platform operation service optimization method and system combined with multimodal large model, it is related to digital service technical field, by layout snapshot flow is divided into layout component instance, and is bound with interactive behavior flow, generates binding relationship mapping table;Based on the binding relationship mapping table, the interactive behavior sequence and the local performance index sequence of each layout component instance are extracted, combined into component interactive performance joint observation sequence, input into multimodal large model for cross-modal causal inference, generate causal influence relationship topological graph with layout component instance as node and causal conduction relationship as directed edge, identify root layout component instance according to the causal influence intensity distribution therein, generate service process reconstruction scheme along upstream and downstream path and trigger linkage update.The application realizes the automatic positioning of root cause and system-level automatic reconstruction of visual platform operation problem.

Description

Technical Field

[0001] This invention relates to the field of digital service technology, and more specifically, to a method and system for optimizing the operation services of a visualization platform that combines a multimodal large model. Background Technology

[0002] Visualization platforms, as core carriers of data-driven decision-making, have been widely deployed in smart cities, industrial monitoring, and business analytics. The operational quality of these platforms directly depends on the rationality of their interface layout and the stability of their backend service performance. Current technologies primarily rely on manual experience or passive response mechanisms based on single-dimensional monitoring metrics for operational optimization of visualization platforms. For example, some solutions collect performance metrics such as server response latency and throughput, triggering alarms when these metrics exceed preset thresholds, requiring maintenance personnel to manually troubleshoot problematic components and adjust the interface. Other solutions analyze high-frequency user operation areas based on user click heatmaps, using this information to statically optimize the interface layout.

[0003] However, the above solutions all have significant shortcomings: First, there is a lack of deep correlation between performance monitoring and interface layout, making it difficult for operations and maintenance personnel to pinpoint the true source of performance bottlenecks from a global perspective. They often can only perform localized repairs on surface-level components, leading to recurring problems. Second, user interaction data and performance metrics data are usually stored and analyzed in isolation, failing to establish a causal transmission link between the two. This makes it impossible to reveal how an interaction anomaly in a particular layout component gradually propagates through inter-component relationships and ultimately leads to overall service performance degradation. Furthermore, existing solutions do not introduce mechanisms for cross-modal joint reasoning on multi-source heterogeneous data, making it impossible to automatically identify root causes and generate systematic refactoring solutions from the integrated analysis of behavior, layout, and performance dimensions. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for optimizing the operation service of a visualization platform that incorporates a multimodal large model, the method comprising: During the operation of the visualization platform, user interaction behavior records, platform interface layout snapshots, and platform service performance monitoring metrics are collected simultaneously. The platform interface layout snapshot stream is processed by layout component instance segmentation to obtain layout component instance segmentation results. The user interaction behavior record stream is associated and bound according to the layout component instance segmentation results to generate a binding relationship mapping table. Based on the binding relationship mapping table, the interaction behavior sequence and the corresponding local performance index sequence of each layout component instance within a continuous time window are extracted and combined into a joint observation sequence of component interaction performance at the layout component granularity. The joint observation sequence of component interaction performance is input into a pre-built multimodal large model. Cross-modal causal reasoning is performed on the joint observation sequence of component interaction performance to generate a causal influence relationship topology graph between each layout component instance. The causal influence relationship topology graph is constructed with layout component instances as nodes and causal transmission relationships of interaction performance between components as directed edges. Based on the causal influence intensity distribution of each layout component instance in the causal influence topology diagram, the root layout component instance that causes the platform service performance degradation is identified. After generating a service process reconstruction scheme based on the upstream and downstream association paths of the root layout component instance in the causal influence topology diagram, the scheme is sent to the background configuration system of the visualization platform to trigger the linkage update operation of the interface layout and service configuration.

[0005] Furthermore, embodiments of the present invention also provide a visualization platform operation service optimization system that combines a multimodal large model, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned method for optimizing the operation service of a visualization platform incorporating a multimodal large model by executing the machine-executable instructions.

[0006] Based on the above, this invention synchronously collects user interaction behavior recording streams, platform interface layout snapshot streams, and platform service performance monitoring indicator streams during the operation of the visualization platform. It then performs layout component instance segmentation on the layout snapshot streams to obtain fine-grained component-level annotations. Furthermore, it associates and binds the interaction behavior recording streams with the segmentation results, constructing a precise mapping relationship between behavioral data and interface structure. On this basis, using a binding relationship mapping table as a link, this invention extracts the interaction behavior sequence and corresponding local performance indicator sequence of each layout component instance within a continuous time window and combines them into a joint observation sequence of component interaction performance. This achieves temporal alignment and joint representation of the behavioral and performance dimensions at the component granularity. Further, by inputting the aforementioned joint observation sequence into a pre-constructed multimodal large model and performing cross-modal causal inference processing, it can overcome the information limitations of single-modal analysis, automatically inferring the causal influence relationship between each layout component instance from the coupling mode of behavioral anomalies and performance fluctuations, and explicitly presenting the causal transmission path between components in the form of a topological graph. This transforms the implicit, cross-component performance decay transmission mechanism into interpretable and quantifiable structured knowledge. Ultimately, based on the distribution of causal influence intensity in the causal influence topology graph, the root cause layout component instance is identified, and a service process reconstruction scheme is generated along the upstream and downstream related paths. This enables the platform to accurately identify the root cause and update the interface layout and service configuration in the early stages of performance degradation, significantly improving the timeliness, accuracy and systematicness of operation optimization, and overcoming the inherent defects of existing technologies that rely on human experience and can only perform local passive repairs. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the visualization platform operation service optimization method combining multimodal large models provided in the embodiments of the present invention.

[0008] Figure 2 This invention demonstrates the service health monitoring overview interface of the visualization platform operation service optimization method that combines multimodal large models during operation.

[0009] Figure 3 The interactive interface of the causal relationship topology diagram during the operation of the present invention is shown.

[0010] Figure 4 The interface for generating bottleneck identification and service process refactoring solutions during the operation of this invention is shown.

[0011] Figure 5 The interface demonstrates the comparison of solutions and the reuse of historical experience during the operation of this invention. Detailed Implementation

[0012] Figure 1 This is a flowchart illustrating a visualization platform operation service optimization method combining a multimodal large model, provided by an embodiment of the present invention. This method is applied to the operation and maintenance of a data visualization analysis platform. During platform operation, users explore data, generate reports, and monitor metrics through interactive dashboards, and complex interrelationships exist between various visualization components. When platform service performance degrades, traditional operation and maintenance methods struggle to pinpoint which visualization component triggered the chain reaction of performance issues. This method synchronously collects user interaction behavior records, platform interface layout snapshots, and platform service performance monitoring metrics to construct a joint observation sequence of component interaction performance at the layout component granularity. It then utilizes a pre-built multimodal large model to perform cross-modal causal inference, generating a causal influence topology graph between instances of each layout component, identifying the root cause layout component instance, and generating a service process refactoring scheme. In this embodiment, all data collection is conducted with user authorization and consent, strictly adhering to data security and personal information protection regulations. The collected data is used solely for platform performance optimization purposes.

[0013] Step S110: During the operation of the visualization platform, simultaneously collect user interaction behavior record streams, platform interface layout snapshot streams, and platform service performance monitoring indicator streams.

[0014] The user interaction behavior log stream originates from a data acquisition probe on the front end of the visualization platform. This probe is attached to the interaction interfaces of various visualization components via event listening. Whenever a user performs an action such as clicking, dragging, zooming, filtering, switching tabs, or entering text, the data acquisition probe generates a user interaction event record. This record contains an interaction event identifier, an interaction timestamp, an interaction type identifier, interaction coordinates, and a set of key-value pairs for interaction parameters. User interaction event records are appended to the user interaction behavior log stream in order of their interaction timestamps. The user interaction behavior log stream is an ordered sequence of records, stored in the first topic partition of the message queue.

[0015] The platform interface layout snapshot stream originates from the layout snapshot timed collection mechanism of the visualization platform's front-end rendering engine. The rendering engine automatically triggers the generation of interface layout snapshots at a preset collection period. Each frame of the platform interface layout snapshot includes a snapshot timestamp and an interface rendering tree structure. The interface rendering tree structure records the hierarchical organization, component type identifiers, component attribute configurations, and component spatial bounding box coordinates of all visual components in the current interface. Platform interface layout snapshots are appended to the platform interface layout snapshot stream in the order of their snapshot timestamps, and the platform interface layout snapshot stream is stored in the second topic partition of the message queue.

[0016] The platform service performance monitoring metric stream originates from the performance monitoring component of the visualization platform's backend services. This component intercepts service call chains and collects server resource metrics, generating a platform service performance monitoring record for each monitoring period. This record includes a monitoring timestamp, service endpoint identifier, service response time, service request frequency, service resource utilization, and service error rate. These records are appended to the platform service performance monitoring metric stream in timestamp order, and the stream is stored in the third topic partition of the message queue. The timestamps of these three data streams are aligned using a unified clock synchronization mechanism.

[0017] Step S120: Perform layout component instance segmentation processing on the platform interface layout snapshot stream to obtain layout component instance segmentation results, and perform association binding processing on the user interaction behavior record stream according to the layout component instance segmentation results to generate a binding relationship mapping table.

[0018] Step S121: Extract the interface rendering tree structure of each frame of the platform interface layout snapshot in the platform interface layout snapshot stream, perform a depth-first traversal on the interface rendering tree structure, extract the node type identifier, node bounding box coordinates and node level depth value of each interface rendering tree node, and generate an interface rendering tree node attribute set.

[0019] The platform interface layout snapshot stream is read from the second topic partition of the message queue and processed frame by frame in snapshot timestamp order. For each frame of the platform interface layout snapshot, the root node of its interface rendering tree structure is read, and a stack-based depth-first traversal algorithm is used to traverse all nodes of the interface rendering tree structure. During the traversal, for each visited interface rendering tree node, the node type identifier is extracted from the attribute field of the interface rendering tree node. The node type identifier has the value of an enumeration string such as chart component, table component, filter component, text component, or container component. The node bounding box coordinates are extracted from the layout field of the interface rendering tree node. The node bounding box coordinates include the x-coordinate of the top left corner, the y-coordinate of the top left corner, the width value, and the height value. The node level depth value is extracted from the level field of the interface rendering tree node. The node level depth value is the number of edges traversed from the root node to the interface rendering tree node. The node type identifier, node bounding box coordinates, and node level depth value are combined into the attribute record of the interface rendering tree node. All the attribute records of the interface rendering tree nodes are collected into the interface rendering tree node attribute set of the platform interface layout snapshot frame.

[0020] Step S122: Input the set of attributes of the interface rendering tree node into the pre-trained layout component instance segmentation model, perform semantic grouping processing on the interface rendering tree node through the layout component instance segmentation model, aggregate multiple interface rendering tree nodes belonging to the same functional component into layout component instances, and generate the component instance identifier, component instance bounding box coordinate set and component instance type identifier for each layout component instance.

[0021] The pre-trained layout component instance segmentation model adopts a semantic segmentation architecture based on graph neural networks, including a node feature encoding layer and a node grouping prediction layer. The node feature encoding layer maps the node type identifier, node bounding box coordinates, and node level depth of each UI / UX node into node feature vectors through embedding and fully connected layers, respectively. The node grouping prediction layer constructs an adjacency graph on the UI / UX structure, using parent-child and sibling relationships as adjacency edges. It aggregates neighborhood node features through graph convolution operations, outputting the probability distribution of each UI / UX node belonging to various functional component clusters. The functional component cluster identifier corresponding to the maximum value of the probability distribution is selected, and UI / UX nodes with the same functional component cluster identifier are aggregated into a single layout component instance. A globally unique component instance identifier is assigned to this layout component instance. The union boundary of the node bounding box coordinates of all UI / UX nodes within this layout component instance is used as the component instance bounding box coordinate set. The node type identifier with the highest frequency among all UI / UX nodes within this layout component instance is taken as the component instance type identifier.

[0022] Step S123: Using the set of component instance bounding box coordinates of the layout component instance, mark the spatial location area of ​​each layout component instance in the platform interface layout snapshot stream, and generate a platform interface layout snapshot sequence with component instance spatial annotations.

[0023] For each frame of the platform interface layout snapshot, a layout component instance annotation layer is attached to the interface rendering tree structure of that frame. The layout component instance annotation layer contains the mapping relationship between the component instance identifiers and the set of component instance bounding box coordinates of all layout component instances in that frame. The platform interface layout snapshots with attached annotation layers are arranged in order of snapshot timestamp to generate a sequence of platform interface layout snapshots with component instance spatial annotations.

[0024] Step S124: Extract the interaction coordinates, interaction timestamp, and interaction type identifier of each user interaction event in the user interaction behavior record stream. Perform spatial coordinate matching processing between the interaction coordinates and the platform interface layout snapshots with the same timestamp in the platform interface layout snapshot sequence with component instance spatial annotations to determine the target layout component instance corresponding to each user interaction event.

[0025] Read the user interaction behavior record stream from the first topic partition of the message queue, and process each user interaction event record one by one in order of interaction timestamp. Extract the interaction coordinates and interaction timestamp of each user interaction event record. In the platform interface layout snapshot sequence with component instance space annotation, retrieve the platform interface layout snapshot with the interaction timestamp closest to the user interaction event and a time difference less than a preset time window threshold. In the layout component instance annotation layer of the platform interface layout snapshot, traverse the set of component instance bounding box coordinates of all layout component instances, and determine whether the interaction coordinates of the user interaction event fall within the rectangular area defined by the set of component instance bounding box coordinates of a certain layout component instance. If they do, then the layout component instance is determined as the target layout component instance corresponding to the user interaction event. If the interaction coordinates do not fall within the bounding box of any layout component instance, then the target layout component instance is set to null.

[0026] Step S125: Based on the target layout component instance corresponding to each user interaction event, establish an association binding relationship between the interaction event identifier and the layout component instance identifier, and combine the interaction event identifier, layout component instance identifier, interaction timestamp, and interaction type identifier into an association binding record.

[0027] For each user interaction event record with a identified target layout component instance, create an associated binding record. The associated binding record contains four fields: the interaction event identifier field (valued as the interaction event identifier of the user interaction event), the layout component instance identifier field (valued as the component instance identifier of the target layout component instance), the interaction timestamp field (valued as the interaction timestamp of the user interaction event), and the interaction type identifier field (valued as the interaction type identifier of the user interaction event). If the target layout component instance is null, skip the user interaction event record and do not create an associated binding record.

[0028] Step S126: Sort all associated binding records in ascending order of interaction timestamp to generate a binding relationship mapping table. Each record in the binding relationship mapping table uniquely identifies the correspondence between a user interaction event and the layout component instance it acts upon.

[0029] Sort all associated binding records by the interaction timestamp field value in ascending order. The sorted list of associated binding records is the binding relationship mapping table. The binding relationship mapping table is stored in a table structure, containing columns for interaction event identifier, layout component instance identifier, interaction timestamp, and interaction type identifier. Each row contains one associated binding record.

[0030] Step S130: Based on the binding relationship mapping table, extract the interaction behavior sequence and the corresponding local performance index sequence of each layout component instance within a continuous time window, and combine them into a joint observation sequence of component interaction performance at the layout component granularity.

[0031] Step S131: Based on the binding relationship mapping table, group and aggregate all associated binding records using the layout component instance identifier as the grouping primary key, and merge all associated binding records belonging to the same layout component instance identifier into a layout component instance interaction record group.

[0032] Iterate through all associated binding records in the binding relationship mapping table, using the layout component instance identifier field of the associated binding record as the grouping key, and perform grouping and aggregation using a hash map table. The key of the hash map table is the layout component instance identifier, and the value is the list of associated binding records corresponding to that layout component instance identifier. Append each associated binding record to the associated binding record list corresponding to the layout component instance identifier. After the iteration is complete, each entry in the hash map table corresponds to a group of layout component instance interaction records.

[0033] Step S132: Group the interaction records of each layout component instance, extract the interaction type identifier sequence and interaction time interval sequence corresponding to the layout component instance in ascending order of unified timestamp, and combine the interaction type identifier sequence and interaction time interval sequence into the interaction behavior sequence of the layout component instance.

[0034] For each layout component instance's interaction record group, sort the associated binding record list in ascending order by the interaction timestamp field. Extract the interaction type identifier field value for each associated binding record from the sorted list, forming an interaction type identifier sequence. Calculate the difference between the interaction timestamp field values ​​of two adjacent associated binding records in the sorted list, forming an interaction time interval sequence. The length of the interaction time interval sequence is equal to the length of the interaction type identifier sequence minus 1. Align the interaction type identifier sequence and the interaction time interval sequence by time index and concatenate them along the feature dimension to obtain the interaction behavior sequence of the layout component instance. Each time step in the interaction behavior sequence contains two feature values: the interaction type identifier and the time interval between the previous interaction.

[0035] Step S133: Based on the platform service performance monitoring indicator stream, extract the local service response time sequence, local service request frequency sequence, and local resource occupancy rate sequence corresponding to the spatial location region of each layout component instance, and combine the local service response time sequence, local service request frequency sequence, and local resource occupancy rate sequence into a local performance indicator sequence for the layout component instance.

[0036] Read the platform service performance monitoring metric stream from the third topic partition of the message queue. For each platform service performance monitoring record, extract its service endpoint identifier. Based on the preset mapping table between the service endpoint identifier and the layout component instance identifier, determine the layout component instance identifier associated with the platform service performance monitoring record. Append the service response time value from the platform service performance monitoring record to the local service response time sequence of the corresponding layout component instance, append the service request frequency value to the local service request frequency sequence, and append the service resource utilization value to the local resource utilization sequence. After processing, for each layout component instance, align the local service response time sequence, local service request frequency sequence, and local resource utilization sequence by time index and concatenate them along the feature dimension to obtain the local performance metric sequence. Each time step of the local performance metric sequence contains three feature values: service response time value, service request frequency value, and service resource utilization value.

[0037] Step S134: Align the interaction behavior sequence and local performance index sequence of each layout component instance with a sliding window in the time dimension. Divide the continuous time window according to the preset window duration parameter and window sliding step parameter, and extract the interaction behavior sequence segment and local performance index sequence segment in each continuous time window.

[0038] Using a unified starting time point as a baseline, time windows are continuously divided starting from that point, with the window duration parameter W as the window length and the sliding step parameter S as the distance the window slides each time. The starting time difference between two adjacent time windows is S. For the interaction behavior sequence of each layout component instance, a segment of the interaction behavior sequence at the corresponding time step is extracted within the time range of each consecutive time window. The same extraction operation is performed on the local performance indicator sequence. If the time sampling rates of the interaction behavior sequence and the local performance indicator sequence are different, a linear interpolation method is used to unify them to the same time resolution before extraction.

[0039] Step S135: Horizontally combine the interactive behavior sequence segments and local performance index sequence segments extracted within the same continuous time window to generate a joint observation segment of the component interaction performance of the layout component instance within the continuous time window. Arrange the joint observation segments of the component interaction performance of all continuous time windows in chronological order to form a joint observation sequence of component interaction performance at the layout component level.

[0040] For each consecutive time window of each layout component instance, the interaction behavior sequence fragments and local performance indicator sequence fragments within that consecutive time window are horizontally concatenated along the feature dimensions. The concatenated joint observation fragment of component interaction performance at each time step includes five feature dimensions: interaction type identifier, interaction time interval, service response time value, service request frequency value, and service resource utilization value. The joint observation fragments of component interaction performance for all consecutive time windows are arranged in ascending order of time window number to form the joint observation sequence of component interaction performance for that layout component instance. The joint observation sequences of component interaction performance for all layout component instances constitute the set of joint observation sequences of component interaction performance at the layout component granularity.

[0041] Step S140: Input the joint observation sequence of component interaction performance into the pre-built multimodal large model, and generate a causal influence relationship topology graph between each layout component instance by performing cross-modal causal reasoning processing on the joint observation sequence of component interaction performance. The causal influence relationship topology graph is constructed with layout component instances as nodes and causal transmission relationships of interaction performance between components as directed edges.

[0042] Step S141: Input the joint observation sequence of component interaction performance of all layout component instances into the multimodal feature encoding layer of the multimodal large model, perform temporal feature extraction processing on the interaction behavior sequence of each layout component instance to generate an interaction behavior semantic encoding vector, and perform performance feature extraction processing on the local performance index sequence of each layout component instance to generate a performance index encoding vector.

[0043] The pre-built multimodal large model comprises three core parts: a multimodal feature encoding layer, a cross-modal fusion layer, and a causal reasoning analysis module. The multimodal feature encoding layer includes an interaction behavior encoding sub-network and a performance metric encoding sub-network. The interaction behavior encoding sub-network employs a temporal feature extraction network based on a Transformer encoder architecture, containing multiple self-attention layers and feedforward layers. The portion of the joint observation sequence of component interaction performance belonging to the interaction behavior is input into the interaction behavior encoding sub-network. The self-attention layer calculates the attention weights between different time steps in the interaction behavior sequence, aggregates global temporal context information, and takes the mean-pooled vector of the output sequence of the last encoder layer as the interaction behavior semantic encoding vector. The performance metric encoding sub-network employs a one-dimensional temporal convolutional neural network architecture, containing multiple one-dimensional convolutional layers and pooling layers. The portion of the joint observation sequence of component interaction performance belonging to the local performance metric is input into the performance metric encoding sub-network. Multi-scale performance fluctuation features are extracted through layer-by-layer convolution and downsampling. The vector of the output of the last convolutional layer after global average pooling is taken as the performance metric encoding vector.

[0044] Step S142: Input the interaction behavior semantic encoding vector and performance index encoding vector into the cross-modal fusion layer of the multimodal large model, perform intermodal information interaction fusion processing on the interaction behavior semantic encoding vector and performance index encoding vector, and generate a cross-modal fusion representation vector for each layout component instance.

[0045] The cross-modal fusion layer employs a multimodal fusion architecture based on a cross-attention mechanism. The semantic encoding vector of the interaction behavior is used as the first modality query vector, and the performance indicator encoding vector is used as the first modality key and value vectors. Cross-attention is used to calculate the attention weight of the interaction behavior on the performance indicator, generating a behavior-to-performance fusion vector. The performance indicator encoding vector is used as the second modality query vector, and the semantic encoding vector of the interaction behavior is used as the second modality key and value vectors. Cross-attention is used to calculate the attention weight of the performance indicator on the interaction behavior, generating a performance-to-behavior fusion vector. The behavior-to-performance fusion vector and the performance-to-behavior fusion vector are concatenated along the feature dimension and compressed through a fully connected layer to generate the cross-modal fusion representation vector of the layout component instance.

[0046] Step S143: Input the cross-modal fusion representation vectors of all layout component instances into the causal reasoning analysis module of the multimodal large model, perform Granger causality test processing on the cross-modal fusion representation vectors between all layout component instances, and generate the causal association strength coefficients between the pairs of layout component instances.

[0047] The causal inference analysis module employs the Granger causality test algorithm based on a vector autoregression model. For each pair of layout component instances X and Y, a first regression model is constructed to predict the cross-modal fusion representation vector of Y using the cross-modal fusion representation vector of X as the independent variable, and a second regression model is constructed to predict the cross-modal fusion representation vector of X using the cross-modal fusion representation vector of Y as the independent variable. Simultaneously, a baseline regression model is constructed that predicts itself using only its own historical values. The causal association strength coefficient of X to Y is calculated by comparing the ratio of the sum of squared prediction residuals of the first regression model and the baseline regression model. The causal association strength coefficient of Y to X is calculated by comparing the ratio of the sum of squared prediction residuals of the second regression model and the baseline regression model. The causal association strength coefficient is a positive real number; a larger value indicates a stronger causal association.

[0048] Step S144: Perform direction determination processing on the timing precedence of the interaction behavior and the lag of the performance index response between each pair of layout component instances to determine the directional causal transmission direction.

[0049] For each pair of layout component instances, hysteresis cross-correlation analysis is used for direction determination. The cross-correlation function values ​​of the interaction behavior sequence of X and the performance index sequence of Y at different hysteresis orders are calculated, as well as the cross-correlation function values ​​of the interaction behavior sequence of Y and the performance index sequence of X at different hysteresis orders. The hysteresis order directions corresponding to the maximum peak values ​​of the two sets of cross-correlation functions are compared, and the hysteresis direction corresponding to the peak value is determined as the causal transmission direction. If the interaction behavior change of X precedes the performance index response of Y, then the causal transmission direction is from X to Y.

[0050] Step S145: Using each layout component instance as a node, the causal relationship between any two layout component instances with a causal relationship strength coefficient exceeding a preset causal threshold as directed edges, and the direction of directed causal propagation as the edge direction, generate a topology graph of causal influence relationships between each layout component instance.

[0051] Create a set of nodes for a causal relationship topology graph. Each node corresponds to a layout component instance, and the node identifier is the component instance identifier of that layout component instance. Iterate through all pairs of layout component instances. If the causal association strength coefficient between a pair of layout component instances is greater than a preset causal threshold, create a directed edge in the causal relationship topology graph. The starting node of the directed edge is the node corresponding to the starting layout component instance in the causal propagation direction, and the ending node is the node corresponding to the ending layout component instance in the causal propagation direction. The weight attribute of the directed edge is the causal association strength coefficient. The causal relationship topology graph is stored in the form of an adjacency list.

[0052] Step S150: Based on the causal influence intensity distribution of each layout component instance in the causal influence relationship topology diagram, identify the root layout component instance that causes the platform service performance degradation, and generate a service process reconstruction scheme based on the upstream and downstream association paths of the root layout component instance in the causal influence relationship topology diagram, and send it to the background configuration system of the visualization platform to trigger the linkage update operation of the interface layout and service configuration.

[0053] Traverse all nodes in the causal relationship topology graph. For each node, calculate the ratio of the number of times it appears as the starting point of all causal transmission paths containing that node to the number of times it appears as an intermediate node in those paths. If the ratio of a node exceeds a preset root cause determination threshold, and the node's out-degree is greater than its in-degree, then the layout component instance corresponding to that node is identified as the root layout component instance. The root layout component instance is the initiator of the causal transmission chain, and its influence on other layout component instances is propagated step by step through directed edges.

[0054] In the causal relationship topology graph, starting from the node corresponding to the root layout component instance, a depth-first traversal is performed along the directed edges to extract all downstream layout component instances reachable from the root layout component instance, forming the root influence range set. Based on the causal association strength coefficient of each downstream layout component instance in the root influence range set, the performance degradation weight of that downstream layout component instance is calculated. The service process refactoring scheme includes two parts: an interface layout adjustment scheme and a service configuration adjustment scheme. The interface layout adjustment scheme generates suggestions for spatial relocation and visual presentation attribute modification of the root layout component instance based on the component instance type identifier and the component instance bounding box coordinate set. The service configuration adjustment scheme generates resource quota adjustment parameters and rate limiting threshold adjustment parameters for the corresponding service endpoints based on the performance degradation weights of the downstream layout component instances.

[0055] The service process refactoring solution is serialized into a configuration update command data packet, which is then sent to the backend configuration system via the visualization platform's backend configuration system interface. After parsing the configuration update command data packet, the backend configuration system calls the interface layout rendering interface to update the position and style of the components displayed on the front end of the visualization platform, and calls the service configuration management interface to adjust the resource allocation and rate limiting parameters of the backend service, thus completing the linked update operation of the interface layout and service configuration.

[0056] Step S210: Extract the causal transmission path of each layout component instance in the causal influence relationship topology graph, trace the entire sequence of layout component instances from the root layout component instance to the end layout component instance along the directed edge direction of each causal transmission path, and generate a set of causal transmission paths.

[0057] In the causal relationship topology graph, the node corresponding to each root layout component instance identified in step S150 is used as the search starting point, and the node corresponding to the end layout component instance with an out-degree of 0 is used as the search ending point. A backtracking-based depth-first search algorithm is used to enumerate all complete directed paths from the starting point to the ending point. During the search process, the current path stack and the set of visited nodes are maintained. When the search reaches the end node, the sequence of nodes in the current path stack is added to the causal transmission path set as a causal transmission path. Each causal transmission path is an ordered list, where the elements are layout component instance identifiers, arranged in the causal transmission direction from the root to the end.

[0058] Step S220: For each causal transmission path in the set of causal transmission paths, calculate the cumulative causal influence strength value of the path based on the causal influence strength between adjacent layout component instances on the path, and calculate the path length value based on the number of layout component instances on the path.

[0059] For each causal propagation path, traverse adjacent pairs of layout component instances along the path and query the causal association strength coefficient of the directed edge between the pair of layout component instances from the causal influence topology graph. Multiply the causal association strength coefficients between all adjacent pairs of layout component instances along the path; the product is the cumulative causal influence strength value of the path. Subtract 1 from the total number of layout component instances along the path to obtain the path length value, which represents the number of hops traversed during causal propagation.

[0060] Step S230: Using a preset causal path clustering analysis algorithm, perform path structure similarity clustering on all causal transmission paths in the causal transmission path set, and aggregate causal transmission paths with similar causal transmission chain structures into the same causal transmission path cluster, generating multiple causal transmission path clusters.

[0061] For each pair of causal transmission paths, the component instance type identifiers of each layout component instance on both paths are extracted to form a component instance type identifier sequence. The length of the longest common subsequence of the two component instance type identifier sequences is calculated and divided by the average length of the two paths to obtain the path structure similarity. Connection edges are established between causal transmission path pairs whose path structure similarity exceeds a preset similarity threshold. On the path similarity graph constructed in this way, a connected component analysis algorithm is used to aggregate causal transmission paths with connectivity into the same causal transmission path cluster.

[0062] Step S240: For each causal transmission path cluster, extract the layout component instances that all causal transmission paths in the cluster pass through as the cluster common bottleneck node, and generate the bottleneck severity parameter of the cluster common bottleneck node based on the average causal influence intensity of the cluster common bottleneck node on each path in its respective causal transmission path cluster.

[0063] For each cluster of causal transmission paths, the sequence of layout component instances for all causal transmission paths in the cluster is converted into a set. The intersection of all path sets is taken as the set of cluster common bottleneck nodes. For each cluster common bottleneck node, the causal association strength coefficient of the directed edge between the node and its predecessor node is found on each causal transmission path of the cluster. The arithmetic mean of this coefficient over all paths is calculated. This average value is the bottleneck severity parameter of the cluster common bottleneck node.

[0064] Step S250: Perform comprehensive sorting of the cluster common bottleneck nodes and corresponding bottleneck severity parameters of all causal transmission path clusters to generate a priority sequence of bottleneck nodes across path clusters. The cluster common bottleneck node ranked first in the priority sequence of bottleneck nodes across path clusters is identified as the core bottleneck node of the cross-path cluster.

[0065] Collect the common bottleneck nodes and their bottleneck severity parameters for all causal propagation path clusters, and sort them in descending order of bottleneck severity parameter values. If the same layout component instance appears in multiple causal propagation path clusters, its bottleneck severity parameters in each cluster are summed before being included in the sorting. The sorting result is the priority sequence of bottleneck nodes across path clusters. The common bottleneck node at the top of this sequence, i.e., the one with the largest bottleneck severity parameter, is identified as the core bottleneck node across the path cluster.

[0066] Step S260: Based on the topological position of the cross-path cluster core bottleneck node in the causal influence relationship topology graph, extract all incoming edge connection layout component instances and all outgoing edge connection layout component instances of the cross-path cluster core bottleneck node, and generate a set of upstream and downstream related layout components for the cross-path cluster core bottleneck node.

[0067] In the causal relationship topology graph, taking the node corresponding to the core bottleneck node of the cross-path cluster as the center, collect all layout component instances corresponding to the directly upstream adjacent nodes along the inbound edge direction, forming the inbound edge connected layout component instance set. Collect all layout component instances corresponding to the directly downstream adjacent nodes along the outbound edge direction, forming the outbound edge connected layout component instance set. Perform a union operation on the above two sets, and the union result is the upstream and downstream related layout component set.

[0068] Step S270: Perform interface layout reconstruction graph generation processing on the upstream and downstream related layout component set of the cross-path cluster core bottleneck node, calculate the optimal spatial arrangement coordinates and optimal visual presentation attributes of each layout component instance in the upstream and downstream related layout component set, and generate a local reconstruction scheme for the cross-path cluster interface layout.

[0069] Centered on the layout component instance corresponding to the core bottleneck node of the cross-path cluster, the current spatial arrangement coordinates and causal correlation strength coefficients of all layout component instances in its upstream and downstream related layout component sets are input into the force-oriented layout algorithm. The force-oriented layout algorithm treats the layout component instance as a mass point, the causal correlation strength coefficient as a spring stiffness coefficient, and the component instance bounding box coordinate set as the collision detection boundary, iteratively solving for the coordinates of each mass point in the force equilibrium state as the optimal spatial arrangement coordinates. Based on the performance degradation weight of each layout component instance, the component size scaling ratio in the optimal visual presentation attribute is calculated; a larger performance degradation weight results in a smaller scaling ratio to reduce visual prominence. Simultaneously, the component color transparency value in the optimal visual presentation attribute is calculated; a larger performance degradation weight results in higher transparency. The above optimal spatial arrangement coordinates and optimal visual presentation attributes are combined into a local reconstruction scheme for the cross-path cluster interface layout.

[0070] Step S280: Merge the cross-path cluster interface layout partial reconstruction scheme with the service process reconstruction scheme to generate a collaborative service process reconstruction scheme that integrates multi-causal path cluster bottleneck optimization, and send the collaborative service process reconstruction scheme that integrates multi-causal path cluster bottleneck optimization to the background configuration system of the visualization platform to trigger the interface layout update operation of multi-causal path cluster linkage.

[0071] The optimal spatial layout coordinates and optimal visual presentation attributes from the cross-path cluster interface layout partial reconstruction scheme are added to the interface layout adjustment scheme of the service process reconstruction scheme as supplementary adjustment items. The newly introduced backend service configuration adjustment items from the cross-path cluster interface layout partial reconstruction scheme are added to the service configuration adjustment scheme of the service process reconstruction scheme. The merged scheme is the collaborative service process reconstruction scheme that integrates multi-causal path cluster bottleneck optimization. This scheme is serialized into a configuration update command data packet and sent to the backend configuration system interface of the visualization platform. After parsing, the backend configuration system calls the interface layout rendering interface and the service configuration management interface to perform a linked update.

[0072] Step S310: Extract the causal propagation delay parameter of each layout component instance in the causal influence relationship topology graph. The causal propagation delay parameter represents the time delay length experienced from the change in the interaction behavior of the upstream layout component instance to the change in the performance index of the downstream layout component instance.

[0073] For each directed edge in the causal relationship topology graph, the time interval in which the interaction frequency changes in the interaction behavior sequence of its upstream layout component instance is extracted, and the start time of the significant response of the corresponding indicator in the performance indicator sequence of its downstream layout component instance is extracted. The time difference between the two is the causal propagation delay parameter of that directed edge. The causal propagation delay parameter is measured in time units and is accurate to the same sampling period as the joint observation sequence of component interaction performance.

[0074] Step S320: Based on the causal propagation delay parameter and the directed edge structure of the causal influence relationship topology graph, map the causal influence relationship topology graph into a causal temporal propagation network graph, wherein each directed edge in the causal temporal propagation network graph is marked with a corresponding causal propagation delay parameter.

[0075] Create a causal temporal transmission network graph object. Its node set is directly copied from the node set of the causal influence relationship topology graph, and each node retains the corresponding layout component instance identifier. Its directed edge set is directly copied from the directed edge set of the causal influence relationship topology graph, and a delay parameter field is added to the edge attribute of each directed edge. The value of the delay parameter field is assigned to the causal transmission delay parameter corresponding to the directed edge calculated in step S310.

[0076] Step S330: Perform oscillation mode analysis on the causal time-series transmission network diagram to identify a set of oscillating layout component instances in the causal time-series transmission network diagram that exhibit periodic fluctuations in causal transmission delay. The fluctuation amplitude of the causal transmission delay parameter of each oscillating layout component instance in the set of oscillating layout component instances exceeds a preset fluctuation threshold within adjacent time windows.

[0077] Causal temporal transmission network graphs are generated within multiple consecutive time windows, yielding a causal transmission delay parameter sequence for each directed edge across these windows. The difference between the maximum and minimum values ​​of this delay parameter sequence is calculated and divided by the mean of the sequence to obtain the fluctuation amplitude ratio. If this fluctuation amplitude ratio exceeds a preset fluctuation threshold, both ends of the directed edge's layout component instances are marked as oscillating layout component instances. The deduplicated set of all marked layout component instances constitutes the oscillating layout component instance set.

[0078] Step S340: Extract the causal propagation delay fluctuation sequence of each oscillating layout component instance in the set of oscillating layout component instances, perform fluctuation period analysis on the causal propagation delay fluctuation sequence, and extract the fluctuation period length and fluctuation phase offset parameters of the oscillating layout component instance.

[0079] For each oscillating layout component instance, the mean sequence of its causal propagation delay parameters across multiple consecutive time windows on all relevant directed edges is collected as the causal propagation delay fluctuation sequence of that oscillating layout component instance. A Fast Fourier Transform is performed on this causal propagation delay fluctuation sequence to obtain its spectrum. The period corresponding to the frequency component with the largest amplitude in the spectrum is taken as the fluctuation period length. The phase angle of this frequency component in the spectrum is taken as the fluctuation phase offset parameter.

[0080] Step S350: Aggregate the oscillation layout component instances with the same fluctuation period length and a difference in fluctuation phase offset parameter less than a preset phase difference threshold into a co-frequency resonance component group, generating multiple co-frequency resonance component groups.

[0081] A disjoint-set data structure is used to cluster the set of oscillating layout component instances. Each oscillating layout component instance is initialized as an independent disjoint-set. All pairs of oscillating layout component instances are traversed; if the difference in their oscillation period lengths is less than a preset period tolerance and the difference in their oscillation phase offset parameters is less than a preset phase difference threshold, then the disjoint-sets containing both instances are merged. After the traversal is complete, each disjoint-set corresponds to a group of resonant components operating at the same frequency.

[0082] Step S360: For each resonant component group, extract the directed edge connection relationship of all oscillating layout component instances in the causal time-series transmission network diagram, and identify the causal transmission oscillation amplification loop inside the resonant component group. The causal transmission oscillation amplification loop is a closed transmission path that enhances oscillation by periodically superimposing the causal transmission delay between oscillating layout component instances.

[0083] In the subgraph of the causal time-series propagation network corresponding to the resonant component group, Johnson's algorithm is used to enumerate all basic loops. For each basic loop, the product of the causal correlation strength coefficients of all directed edges on the loop is calculated as the loop gain. If the loop gain is greater than 1, the basic loop is identified as a causal propagation oscillation amplification loop. A causal propagation oscillation amplification loop indicates that the causal influence intensity is amplified with each propagation cycle in the loop, forming a positive feedback oscillation.

[0084] Step S370: Perform oscillation breakpoint analysis on each causal conduction oscillation amplification loop, select the layout component instance with the largest fluctuation amplitude of the causal conduction delay parameter in the causal conduction oscillation amplification loop as the oscillation breakpoint, and generate a time-series oscillation suppression strategy for the oscillation breakpoint.

[0085] For each layout component instance on each causal propagation oscillation amplification loop, calculate the ratio of the oscillation amplitude of its causal propagation delay oscillation sequence. The layout component instance with the largest oscillation amplitude ratio is taken as the oscillation breakpoint of that loop. The timing oscillation suppression strategy consists of two parts: adding a delay compensation buffer to the interaction response of the layout component instance corresponding to the oscillation breakpoint, with the buffer duration equal to the average causal propagation delay parameter of that layout component instance; and adding isolation protection to the performance resource allocation of the layout component instance corresponding to the oscillation breakpoint, lowering its service resource utilization cap to a preset percentage of its historical average service resource utilization.

[0086] Step S380: According to the time-series oscillation suppression strategy, perform time-series decoupling processing on the interaction behavior sequence and local performance index sequence of the layout component instance corresponding to the oscillation breakpoint, generate the decoupled interaction behavior sequence and decoupled local performance index sequence of the layout component instance corresponding to the oscillation breakpoint, and re-input the decoupled interaction behavior sequence and decoupled local performance index sequence into the causal reasoning analysis module of the multimodal large model for secondary causal reasoning processing, and generate the causal influence relationship topology diagram after oscillation suppression.

[0087] In the interaction behavior sequence of the layout component instance corresponding to the oscillation breakpoint, the time interval between adjacent interaction events is compared with the buffer duration of the delay compensation buffer. If the time interval is less than the buffer duration, the timestamp of the subsequent interaction event is shifted backward by the buffer duration minus the original time interval, thus generating a decoupled interaction behavior sequence. In the local performance indicator sequence, the portion of the service resource occupancy value exceeding the isolation protection limit is truncated to the limit value, generating a decoupled local performance indicator sequence. The two decoupled sequences replace the original sequences, are combined with the original sequences of other layout component instances, and are re-input into the multimodal large model to perform the cross-modal causal reasoning processing in step S140, generating a new causal influence relationship topology graph, which is the causal influence relationship topology graph after oscillation suppression.

[0088] Step S410: Extract the causal influence radiation range parameter of each layout component instance in the causal influence relationship topology graph. The causal influence radiation range parameter is determined based on the number of all downstream layout component instances reachable from the layout component instance along the directed edge and the cumulative causal influence intensity of the layout component instance on these downstream layout component instances.

[0089] For each layout component instance, a breadth-first traversal is performed along the directed edges in the causal influence topology graph, starting from the node corresponding to that layout component instance, to collect all reachable downstream layout component instance nodes. The total number of downstream layout component instances is denoted as the number of radiating nodes. For each path from the given layout component instance to each reachable downstream layout component instance, the product of the causal association strength coefficients of all directed edges on that path is calculated, and the sum of these products over all paths yields the cumulative causal influence strength. The causal influence radiation range parameter is equal to the number of radiating nodes multiplied by the normalization constant plus the cumulative causal influence strength.

[0090] Step S420: Based on the causal influence radiation range parameter of each layout component instance, identify layout component instances whose causal influence radiation range parameter exceeds a preset radiation threshold in the causal influence relationship topology graph as causal influence radiation source components, and generate a set of causal influence radiation source components.

[0091] The causal influence radiation range parameter of all layout component instances is compared with the preset radiation threshold one by one. Layout component instances with causal influence radiation range parameters greater than the preset radiation threshold are selected. These layout component instances constitute the causal influence radiation source component set.

[0092] Step S430: For each causal influence radiation source component in the set of causal influence radiation source components, extract all downstream radiation paths of the causal influence radiation source component in the causal influence relationship topology graph. The downstream radiation path is a complete path that starts from the causal influence radiation source component and propagates along the directed edge to the radiation boundary layout component instance.

[0093] Starting from the node corresponding to the causal radiation source component, and using nodes with an out-degree of 0 or edges with a causal correlation strength coefficient lower than a preset radiation attenuation cutoff threshold as path endpoints, a depth-first search algorithm is used to enumerate all complete paths from the starting point to nodes that satisfy the endpoint conditions. These paths are the downstream radiation paths.

[0094] Step S440: Calculate the radiation attenuation gradient of each downstream radiation path based on the causal influence intensity of each layout component instance on each downstream radiation path. The radiation attenuation gradient represents the rate of change of the causal influence intensity attenuating step by step along the downstream radiation path.

[0095] For each downstream radiation path, starting from the starting point, sequentially take adjacent layout component instance pairs, calculate the difference in the causal correlation strength coefficient between adjacent layout component instance pairs, divide it by the path distance between them (i.e., the hop count 1), and obtain the radiation attenuation rate between adjacent nodes. The arithmetic mean of the radiation attenuation rates between all adjacent nodes on the path is taken as the radiation attenuation gradient of the downstream radiation path.

[0096] Step S450: Sort all downstream radiation paths of the causal radiation source component in order of radiation attenuation gradient from steep to gentle, and identify the downstream radiation path with the steepest radiation attenuation gradient as the key intervention radiation path.

[0097] All downstream radiation paths affecting the causal radiation source component are sorted in descending order of radiation attenuation gradient value. A larger radiation attenuation gradient value indicates a faster attenuation of the causal influence along that path, suggesting a more dramatic change in the downstream impact of causal transmission along that path. The path with the largest radiation attenuation gradient value in the sorted results is identified as the key intervention radiation path.

[0098] Step S460: Based on the path structure and radiation attenuation gradient distribution of the key intervention radiation path, select the layout component instance with the largest radiation attenuation gradient change rate on the key intervention radiation path as the radiation blocking node, and generate a radiation blocking strategy for the radiation blocking node. The radiation blocking strategy includes the interactive function flow limiting parameters and performance resource isolation parameters of the radiation blocking node.

[0099] For each layout component instance along the key intervention radiation path, excluding the start and end points, the absolute value of the difference between the radiation attenuation rate between adjacent nodes at the location of that layout component instance and the radiation attenuation rate between the previous adjacent node is calculated as the radiation attenuation gradient change rate of that layout component instance. The layout component instance with the largest radiation attenuation gradient change rate is selected as the radiation blocking node. The interaction function rate limiting parameter of the radiation blocking strategy is set according to the incoming edge causal correlation strength coefficient of the radiation blocking node, and the rate limiting threshold is the normalized value of the average interaction frequency of that layout component instance in the joint observation sequence of component interaction performance divided by the incoming edge causal correlation strength coefficient. The performance resource isolation parameter is set according to the peak value of the local resource occupancy rate sequence of the radiation blocking node, and the isolation upper limit is a preset proportion of the peak value.

[0100] Step S470: Integrate the radiation blocking strategy into the service process reconstruction scheme to generate a service process reconstruction scheme with integrated radiation blocking. Send the service process reconstruction scheme with integrated radiation blocking to the background configuration system of the visualization platform to trigger the coordinated update operation of interface layout, service configuration and causal radiation blocking.

[0101] A new "Radiation Blocking Configuration Section" is added to the service configuration adjustment scheme of the service process refactoring solution. This section includes radiation blocking node identifiers, interactive function rate limiting parameters, and performance resource isolation parameters. The updated service process refactoring solution is the integrated radiation blocking service process refactoring solution. This is sent and triggered via the backend configuration system interface.

[0102] Step S510: Extract all layout component instance pairs with bidirectional causal transmission relationships in the causal influence relationship topology graph. The bidirectional causal transmission relationship refers to the presence of mutually pointing directed edges between two layout component instances. Layout component instance pairs with bidirectional causal transmission relationships are identified as causal coupling component pairs.

[0103] Traverse all directed edges in the causal relationship topology graph. For each directed edge from node A to node B, check if there is a directed edge from node B to node A in the edge set. If so, pair the layout component instance corresponding to node A and the layout component instance corresponding to node B into a causal coupling component pair and store them in unordered form.

[0104] Step S520: For each causal coupling component pair, calculate the causal coupling tightness parameter between the two layout component instances in the causal coupling component pair. The causal coupling tightness parameter is determined based on the sum of the causal influence strengths on the directed edges of the bidirectional causal transmission relationship.

[0105] For each causal coupling component pair, the causal association strength coefficient F12 of the directed edge from the first layout component instance to the second layout component instance, and the causal association strength coefficient F21 of the directed edge from the second layout component instance to the first layout component instance are read from the causal influence relationship topology graph. The causal coupling tightness parameter is the sum of F12 and F21.

[0106] Step S530: Aggregate causal coupling component pairs whose causal coupling tightness parameter exceeds the preset coupling threshold into the same causal coupling component group. Use the connected component analysis algorithm to merge causal coupling component pairs with shared layout component instances into causal coupling component groups, generating multiple causal coupling component groups.

[0107] An undirected coupling graph is constructed using all causal coupling component pairs whose causal coupling tightness parameters exceed a preset coupling threshold as input. Nodes in the undirected coupling graph represent layout component instances, and edges represent causal coupling component pairs. Breadth-first search or depth-first search is used to perform connected component analysis on the undirected coupling graph; all layout component instances contained in each connected component constitute a causal coupling component group.

[0108] Step S540: For each causal coupling component group, extract the interaction behavior sequence and local performance index sequence of all layout component instances in the component interaction performance joint observation sequence, and input the interaction behavior sequence and local performance index sequence of all layout component instances in the causal coupling component group into the pre-built coupling component collaborative optimization model.

[0109] The pre-built coupled component collaborative optimization model comprises two sub-networks: a coupling effect analysis layer and a coupling / decoupling strategy generation layer. The coupling effect analysis layer employs a multi-head graph attention network architecture, constructing a fully connected graph for each layout component instance within the causal coupled component group. It then calculates the coupling effect weights between each layout component instance and other layout component instances using a graph attention mechanism. The coupling / decoupling strategy generation layer utilizes a reinforcement learning network based on an actor-critic framework. The actor network outputs the probability distribution of decoupling actions, while the critic network outputs the expected cumulative reward for each decoupling action.

[0110] Step S550: The coupling effect analysis layer in the coupled component collaborative optimization model performs coupling effect analysis on the interaction behavior collaboration mode and performance index linkage mode between the layout component instances in the causal coupled component group, and generates coupling effect analysis results.

[0111] The interaction behavior sequences of all layout component instances within the causal coupling component group are aligned along time steps and input as the interaction behavior matrix into the coupling effect analysis layer. The first attention layer of the coupling effect analysis layer calculates the attention weight matrix of interaction behaviors between layout component instances, and the second attention layer calculates the performance linkage attention weight matrix by combining the performance index sequence. The two attention weight matrices are concatenated and dimensionality reduced through a fully connected layer to output the coupling effect feature vector for each layout component instance. The coupling effect analysis results include the coupling effect feature vector of each layout component instance and an evaluation value of the overall coupling tightness of the group calculated based on the mean of the attention weight matrices.

[0112] Step S560: Based on the coupling effect analysis results, a coupling decoupling strategy is generated for the causal coupling component group through the coupling decoupling strategy generation layer in the coupling component collaborative optimization model. The coupling decoupling strategy includes an interface layout splitting scheme that splits the causal coupling component group into multiple independent layout component instances and a service call chain splitting scheme that decouples the background service call chain of the causal coupling component group.

[0113] The coupling effect analysis results are used as the input state vector to the actor network of the coupling-decoupling strategy generation layer. The actor network outputs a multi-dimensional action vector, where each dimension of the action vector corresponds to a layout component instance in the causal coupling component group, and the action value is either maintaining coupling or performing decoupling. The critic network evaluates the value of this state-action pair. The actor network updates its parameters based on the critic network's evaluation gradient, and outputs the optimal decoupling action combination after multiple iterations. Based on the layout component instances marked as performing decoupling in the optimal decoupling action combination, a UI layout splitting scheme is generated: the decoupled layout component instances are separated from each other in the UI space, and the spatial separation distance is proportional to the Euclidean distance of their coupling effect feature vectors. A service call chain splitting scheme is generated: the background service call chains of the decoupled layout component instances are separated from the shared call chain, and an independent service resource quota is allocated to each independent service call chain, with the quota amount proportional to the average service resource utilization rate in the local performance index sequence of that layout component instance.

[0114] Step S570: Integrate the coupling and decoupling strategies of each causal coupling component group with the service process reconstruction scheme to generate a fusion coupling and decoupling collaborative service process reconstruction scheme. Send the fusion coupling and decoupling collaborative service process reconstruction scheme to the background configuration system of the visualization platform to trigger the linkage update operation of interface layout, service configuration and causal coupling and decoupling.

[0115] Iterate through the coupling and decoupling strategies of all causally coupled component groups, extracting the UI layout splitting scheme and service call chain splitting scheme from each strategy. Merge the spatial separation distance and visual independence parameters from the UI layout splitting scheme into the UI layout adjustment scheme of the service process refactoring scheme. If the same layout component instance appears in multiple coupling and decoupling strategies, take the maximum spatial separation distance. Merge the resource quota allocation from the service call chain splitting scheme into the service configuration adjustment scheme of the service process refactoring scheme. The merged scheme is the collaborative service process refactoring scheme that integrates coupling and decoupling. Send this scheme to the backend configuration system. The backend configuration system calls the UI layout rendering interface to update the spatial position of the layout component instance, calls the service configuration management interface to split the service call chain and reallocate resource quotas, and calls the causal relationship graph update interface to remove the bidirectional causal transmission edges between decoupled component pairs in the causal influence relationship topology graph.

[0116] Step S910: Extract the temporal variation features of the causal influence intensity of each layout component instance in the causal influence relationship topology graph, compare the causal influence intensity of the same layout component instance in the causal influence relationship topology graph generated by different continuous time windows in a temporal sequence, and generate the causal influence intensity evolution sequence of each layout component instance.

[0117] In step S140, a causal influence relationship topology graph is generated for each consecutive time window. For each layout component instance, the sum of the causal association strength coefficients of all outgoing edges in the causal influence relationship topology graph corresponding to each consecutive time window is extracted as the causal influence strength value of that time window. The causal influence strength values ​​of all consecutive time windows are arranged in ascending order according to the time window number to form the causal influence strength evolution sequence of that layout component instance, and the sequence length is the total number of consecutive time windows.

[0118] Step S920: Perform trend inflection point detection processing on the causal influence intensity evolution sequence of each layout component instance, identify the trend inflection point in the causal influence intensity evolution sequence where the direction of change of causal influence intensity changes, and mark the causal influence intensity evolution sequence segments within a preset time range before and after the trend inflection point as causal mutation segments.

[0119] The Petit mutation detection algorithm is used to detect trend inflection points in the evolution sequence of causal influence intensity. This algorithm calculates the cumulative rank sum statistic of the two subsequences preceding and following each position in the sequence, identifying the position where the statistic reaches an extreme value as the trend inflection point. A predetermined number of time windows are then extended forward and backward from the trend inflection point; the causal influence intensity evolution sequence segments within this range are considered causal mutation segments.

[0120] Step S930: Align the causal mutation fragments of all layout component instances according to the trend inflection point, identify layout component instances that have causal mutations within the same trend inflection point or adjacent trend inflection point range, and aggregate these layout component instances into a causal mutation synchronization component set.

[0121] All causal mutation fragments of all layout component instances are grouped according to trend inflection points. Instances with an absolute difference between their trend inflection points less than a preset adjacent threshold are grouped together. For each group of trend inflection points, all layout component instances within that group are extracted to form a causal mutation synchronization component set.

[0122] Step S940: For the layout component instances in the causal mutation synchronization component set, backtrack the interaction behavior sequence and local performance index sequence at the corresponding trend turning point in the joint observation sequence of component interaction performance, and extract the abnormal interaction pattern features in the interaction behavior sequence and the abnormal performance fluctuation features in the local performance index sequence.

[0123] For each layout component instance in the causal mutation synchronization component set, a continuous time window corresponding to the trend inflection point is located in the joint observation sequence of component interaction performance. Interaction behavior sequence segments and local performance indicator sequence segments are extracted within this continuous time window. In the interaction behavior sequence segments, the deviation factor of the interaction frequency from the historical average interaction frequency of the layout component instance is calculated as an abnormal interaction mode feature, and the difference between the Shannon entropy of the interaction type distribution and the Shannon entropy of the historical interaction type distribution is calculated as an interaction mode variation feature. In the local performance indicator sequence segments, the difference between the percentile of the service response time value and the percentile of the historical service response time value is calculated as an abnormal performance fluctuation feature, and the difference between the percentile of the service resource utilization rate value and the percentile of the historical service resource utilization rate value is calculated as an abnormal resource utilization feature.

[0124] Step S950: Input the abnormal interaction pattern features and abnormal performance fluctuation features into the pre-constructed causal mutation tracing analysis model, and perform correlation tracing reasoning processing on the abnormal interaction pattern features and abnormal performance fluctuation features through the causal mutation tracing analysis model to identify the initial triggering layout component instance and the initial triggering interaction event that triggered the causal mutation.

[0125] The pre-constructed causal mutation tracing analysis model adopts a structural causal model based on a causal discovery algorithm. The structural causal model uses layout component instances as nodes, abnormal interaction mode features and abnormal performance fluctuation features as observed variables, and causal mutation triggering relationships as latent variables. Independent component analysis (ICA) is used to separate non-Gaussian distributed independent disturbance sources from the observed variables. The causal direction between these independent disturbance sources is inferred using the LiNGAM algorithm, identifying the layout component instance corresponding to the upstream independent disturbance source in the causal chain as the initial triggering layout component instance. From the abnormal interaction mode features of this initial triggering layout component instance, the interaction event type with the largest deviation factor is extracted as the initial triggering interaction event type.

[0126] Step S960: Based on the initial trigger layout component instance and the initial trigger interaction event, add an interaction event pre-interception strategy for the initial trigger layout component instance to the service process reconstruction scheme. The interaction event pre-interception strategy includes identification rules for the initial trigger interaction event and interception response operations.

[0127] The identification rules for the pre-interaction event interception strategy are set based on the interaction type identifier of the initial triggered interaction event type and the interaction frequency deviation multiple in the abnormal interaction mode characteristics. Interception is triggered when the number of requests for this type of interaction event to the initial triggered layout component instance within a unit time exceeds the interception threshold calculated based on the historical average interaction frequency and the deviation multiple. The interception response operation includes request merging for this type of interaction event to reduce the number of duplicate requests, and asynchronous processing of the merged requests to reduce instantaneous load.

[0128] Step S970: Send the service process reconstruction plan with the added interactive event pre-interception strategy to the background configuration system of the visualization platform. The background configuration system will execute the pre-interception response operation when it detects the initial triggering interactive event to block the further propagation of causal mutation.

[0129] The service process refactoring scheme, which includes a pre-interception strategy for interactive events, is serialized into a configuration update instruction data packet and sent to the backend configuration system. The backend configuration system deploys identification rules in the data acquisition probe at the front end of the visualization platform. When the number of requests for this type of interactive event to the initial trigger layout component instance exceeds the interception threshold, it automatically triggers request merging and asynchronous processing operations to prevent the performance fluctuations caused by this interactive event from propagating again through the causal transmission path.

[0130] For example, step S1010: extract the causal influence transmission topology features of each layout component instance in the causal influence relationship topology graph, input the causal influence transmission topology features into a pre-built topology similarity matching model, and perform topology similarity matching processing with the known fault causal topology patterns in the preset known fault causal topology pattern library.

[0131] For the causal influence topology graph, taking the node corresponding to each layout component instance as the center, its local neighborhood subgraph is extracted as the topological feature of the causal influence transmission of that layout component instance. The local neighborhood subgraph includes the layout component instance, all its incoming and outgoing neighbor nodes, and the directed edges between these nodes. The local neighborhood subgraph is represented in the form of a graph adjacency matrix and a node feature matrix. The pre-built topology similarity matching model adopts a graph similarity calculation architecture based on graph isomorphic networks, which encodes the local neighborhood subgraph into a fixed-dimensional graph embedding vector. In the multilayer perceptron update function of the graph isomorphic network, the node feature vector of the (k+1)th layer is jointly determined by the node's k-th layer feature vector, the aggregated value of the k-th layer feature vectors of the neighboring nodes, and the node degree normalization factor. The cosine similarity of the graph embedding vector of the local neighborhood subgraph is calculated one by one with the graph embedding vector of each known fault causal topology pattern in the known fault causal topology pattern library.

[0132] The pre-defined known fault causal topology pattern library stores causal topology patterns extracted from historical service failure events, along with their corresponding known fault type identifiers and known repair strategies. Each record in the known fault causal topology pattern library contains a graph embedding vector of the known fault causal topology pattern, a known fault type identifier, and a known repair strategy.

[0133] Step S1020: When there is a subgraph in the causal influence relationship topology graph that has a topological similarity to the known fault causal topology pattern that exceeds a preset matching threshold, extract the known fault type identifier and the known repair strategy corresponding to the subgraph, use the known fault type identifier as an auxiliary diagnostic label for the current platform service performance degradation, and use the known repair strategy as a candidate repair strategy.

[0134] For each layout component instance's local neighborhood subgraph, if its topological similarity to a known fault causal topological pattern exceeds a preset matching threshold, the matching relationship is recorded. The known fault type identifier and known repair strategy corresponding to the known fault causal topological pattern are extracted. The known fault type identifier is used as an auxiliary diagnostic label on the layout component instance. The known repair strategy is used as a candidate repair strategy, including verified repair operation steps and configuration parameters from historical faults.

[0135] Step S1030: Compare and analyze the consistency of the candidate repair strategy with the service process reconstruction scheme generated by the multimodal large model causal reasoning analysis module. When the candidate repair strategy includes supplementary repair operations not covered in the service process reconstruction scheme, merge the supplementary repair operations into the service process reconstruction scheme to generate an enhanced service process reconstruction scheme that integrates historical fault experience knowledge.

[0136] Each repair step in the candidate repair strategy is semantically matched with the adjustment items in the service process refactoring plan. Semantic matching is achieved by calculating the cosine similarity of the semantic vectors of the repair step description text and the adjustment item description text. If the semantic similarity between a repair operation in the candidate repair strategy and all adjustment items in the service process refactoring plan is lower than a preset semantic matching threshold, then the repair operation is determined to be a supplementary repair operation not covered in the service process refactoring plan. The supplementary repair operation is added to the adjustment item list of the service process refactoring plan to generate an enhanced service process refactoring plan.

[0137] Step S1040: When the topological similarity between the causal influence topology graph and all known fault causal topology patterns in the preset known fault causal topology pattern library is lower than the preset matching threshold, the causal influence topology graph is stored as a new unknown fault causal topology pattern in the known fault causal topology pattern library, and the currently generated service process reconstruction scheme and its effect feedback results after execution are associated and stored with the new unknown fault causal topology pattern.

[0138] If the topological similarity between the local neighborhood subgraphs of all layout component instances in the causal relationship topology graph and the topological patterns of all known fault causal topology patterns in the known fault causal topology pattern library is lower than a preset matching threshold, then the graph embedding vector of the entire causal relationship topology graph is calculated through a graph isomorphic network. This graph embedding vector and the adjacency matrix of the causal relationship topology graph are then stored as a new unknown fault causal topology pattern in the known fault causal topology pattern library, and a new fault pattern identifier is assigned. The currently generated service process refactoring scheme is associated and stored with this new fault pattern identifier. After the service process refactoring scheme is executed, the platform service performance monitoring indicator stream is collected, and the change ratio of the service response time value and service error rate value relative to before execution is calculated. If the change ratio exceeds a preset recovery threshold, the effect feedback result is marked as valid; otherwise, it is marked as invalid. The effect feedback result is synchronously associated and stored.

[0139] Step S1110: Extract the number of incoming edges and the number of outgoing edges of each layout component instance in the causal influence relationship topology graph. Calculate the causal transmission hub degree parameter of the layout component instance based on the number of incoming edges and the number of outgoing edges. The causal transmission hub degree parameter indicates the criticality of the layout component instance as a causal transmission intermediary node in the causal influence relationship topology graph.

[0140] Traverse the topology graph of causal relationships, and for each layout component instance corresponding to a node, count the number of incoming edges (Nin) and the number of outgoing edges (Nout) of that node. The number of incoming edges is the total number of directed edges that the node is the endpoint of directed edges, and the number of outgoing edges is the total number of directed edges that the node is the starting point of directed edges. The causal transmission pivotality parameter H is calculated as H = Nin × Nout + Nin + Nout. This calculation method allows nodes with larger numbers of both incoming and outgoing edges to obtain higher pivotality parameter values, while nodes with larger single-dimensional incoming or outgoing edges also obtain appropriate pivotality parameter values.

[0141] Step S1120: Identify the layout component instances whose causal transmission hub degree parameter exceeds the preset hub threshold as causal transmission hub nodes, extract all incoming edge source layout component instances and all outgoing edge destination layout component instances of the causal transmission hub nodes, and construct a star-shaped causal transmission subgraph centered on the causal transmission hub nodes.

[0142] The causal transmission hub degree parameter of all layout component instances is compared with a preset hub threshold. Layout component instances with a causal transmission hub degree parameter greater than the preset hub threshold are selected and marked as causal transmission hub nodes. For each causal transmission hub node, the starting nodes corresponding to all directed edges ending at this node are extracted from the causal influence relationship topology graph. The layout component instances corresponding to these starting nodes constitute the set of incoming edge source layout component instances. The ending nodes corresponding to all directed edges starting at this node are extracted. The layout component instances corresponding to these ending nodes constitute the set of outgoing edge destination layout component instances. With the causal transmission hub node as the center node, the incoming edge source layout component instances and the outgoing edge destination layout component instances as outer nodes, and the directed edges between the center node and the outer nodes as connecting edges, a star-shaped causal transmission subgraph is constructed.

[0143] Step S1130: Perform behavior pattern consistency analysis on the interaction behavior sequence of the incoming edge source layout component instances in the star-shaped causal transmission subgraph, extract the co-occurrence interaction type and co-occurrence interaction timing pattern of the interaction behavior sequence between the incoming edge source layout component instances, and generate common features of incoming edge source behavior.

[0144] Interaction behavior sequences of all incoming edge source layout component instances are extracted from the joint observation sequence of component interaction performance. For each pair of incoming edge source layout component instance interaction behavior sequences, the overlap ratio of their interaction type identifiers within the same time window is calculated as the interaction type co-occurrence. Interaction type identifiers whose pairwise interaction type co-occurrence exceeds a preset co-occurrence threshold are collected into a co-occurrence interaction type set. After temporal alignment of the interaction behavior sequences of incoming edge source layout component instances, a dynamic time warping algorithm is used to calculate the temporal similarity between each sequence, and the common temporal pattern template of each sequence is extracted as the co-occurrence interaction temporal pattern. The co-occurrence interaction type set and the co-occurrence interaction temporal pattern together constitute the common features of incoming edge source behavior.

[0145] Step S1140: Perform performance response consistency analysis on the local performance index sequence of the outgoing edge layout component instances in the star-shaped causal transmission subgraph, extract the co-occurrence performance fluctuation pattern and co-occurrence performance decay characteristics of the local performance index sequences among the outgoing edge layout component instances, and generate common outgoing edge performance characteristics.

[0146] Local performance index sequences of all outgoing layout component instances in the star-shaped causal propagation subgraph are extracted from the joint observation sequence of component interaction performance. For every two outgoing layout component instances, the proportion of their service response time values ​​changing in the same direction within the same time window is calculated as the performance fluctuation co-occurrence degree. Fluctuation segments where the performance fluctuation co-occurrence degree between all outgoing layout component instances exceeds a preset co-occurrence threshold are aggregated and their common fluctuation patterns are extracted as co-occurring performance fluctuation patterns. Trend analysis is performed on the service response time values ​​and resource utilization values ​​in the local performance index sequences of outgoing layout component instances to extract the performance degradation characteristics that all outgoing layout component instances commonly exhibit after causal propagation occurs, including the degradation start delay time and degradation recovery period. The co-occurring performance fluctuation patterns and co-occurring performance degradation characteristics together constitute the common characteristics of outgoing performance.

[0147] Step S1150: Based on the common characteristics of incoming edge source behavior and the common characteristics of outgoing edge destination performance, the causal transmission hub node is split into multiple virtual causal transmission sub-nodes. Each virtual causal transmission sub-node corresponds to a combination of incoming edge source layout component instances and outgoing edge destination layout component instances with the same co-occurrence interaction type and the same co-occurrence performance fluctuation mode.

[0148] Based on the set of co-occurrence interaction types in the common characteristics of incoming edge source behavior, incoming edge source layout component instances are grouped, with instances having the same co-occurrence interaction type grouped into the same incoming edge source group. Based on the co-occurrence performance fluctuation pattern in the common characteristics of outgoing edge destination performance, outgoing edge destination layout component instances are grouped, with instances having the same co-occurrence performance fluctuation pattern grouped into the same outgoing edge destination group. The incoming edge source group and the outgoing edge destination group are combined using a Cartesian product, with each combination corresponding to a virtual causal transmission child node. The virtual causal transmission child node inherits the component instance type identifier and spatial location region information of the original causal transmission hub node, but the associated incoming edge source layout component instances and outgoing edge destination layout component instances are limited to those within the same combination.

[0149] Step S1160: Perform independent interface layout partitioning on the incoming edge source layout component instance and outgoing edge destination layout component instance corresponding to each virtual causal transmission child node, and generate an independent interface layout partitioning scheme for the virtual causal transmission child node. The independent interface layout partitioning scheme aggregates the corresponding layout component instances into an independent interface display partition.

[0150] For each virtual causal transmission child node, collect its corresponding set of incoming edge source layout component instances and set of outgoing edge destination layout component instances. Create an independent interface display partition on the visualization platform interface. The spatial boundary coordinates of this partition are determined by expanding the default margins of the union of the bounding box coordinates of the aggregated layout component instances. Translate the spatial coordinates of all aggregated layout component instances into this independent interface display partition, maintaining the relative positional relationships between the layout component instances after translation. The independent interface layout partition scheme includes the partition identifier, partition spatial boundary coordinates, and a list of layout component instances within the partition.

[0151] Step S1170: Combine the independent interface layout partitioning schemes of all virtual causal transmission child nodes into a hub node partitioning decoupling layout scheme. Modify the interface layout reconstruction part of the service process reconstruction scheme involving the causal transmission hub node according to the hub node partitioning decoupling layout scheme to generate a hub partitioning decoupling service process reconstruction scheme.

[0152] The independent interface layout partitioning schemes of all virtual causal transmission child nodes are aggregated into a hub node partitioning decoupling layout scheme. In the interface layout adjustment scheme of the service flow refactoring scheme, adjustment items involving the layout component instances corresponding to the original causal transmission hub nodes are retrieved, and these adjustment items are replaced with the independent interface display partitioning settings instructions for each virtual causal transmission child node in the hub node partitioning decoupling layout scheme. The modified service flow refactoring scheme is the hub partitioning decoupling service flow refactoring scheme.

[0153] Step S1180: Send the service process reconstruction scheme of the hub partition decoupling to the background configuration system of the visualization platform. The background configuration system redeploys the layout component instances corresponding to the causal transmission hub nodes to different interface display partitions according to the independent interface layout partition scheme, and configures an independent service resource allocation strategy for each interface display partition.

[0154] The service process refactoring scheme, which decouples the hub partitions, is serialized into configuration update command data packets and sent to the backend configuration system of the visualization platform. After parsing, the backend configuration system creates independent interface display partitions in the interface rendering engine. The layout component instances corresponding to the original causal transmission hub nodes are copied and redeployed according to the division of virtual causal transmission child nodes, with each copy placed within its corresponding independent interface display partition. For each independent interface display partition, an independent service resource pool is allocated in the service configuration management module. The resource quota of the service resource pool is allocated based on the proportion of the average service resource utilization rate in the local performance indicator sequence of the layout component instances within that partition.

[0155] Step S1210: Extract the service performance fluctuation pattern of the platform service performance monitoring index stream in the multi-source heterogeneous operation data stream during different service operation periods, and divide the platform service performance monitoring index stream into peak operation period performance index subsequence and off-peak operation period performance index subsequence according to the preset operation period division strategy.

[0156] The system extracts 24 / 7 service performance monitoring records from the platform's service performance monitoring metric stream and segments them into time periods according to a pre-defined operational time period division strategy. This strategy defines the start and end times of peak and off-peak operational periods. Platform service performance monitoring records whose timestamps fall within the peak operational period's start and end times are categorized into the peak operational period performance metric subsequence, while those whose timestamps fall within the off-peak operational period's start and end times are categorized into the off-peak operational period performance metric subsequence.

[0157] Step S1220: Extract the first causal influence relationship topology map corresponding to the performance index subsequence during peak operation period and the second causal influence relationship topology map corresponding to the performance index subsequence during off-peak operation period. Compare the causal influence intensity of the same layout component instance in the first causal influence relationship topology map and the second causal influence relationship topology map to generate the peak-valley causal difference parameter for each layout component instance.

[0158] The performance index subsequences during peak operating periods are combined with the corresponding user interaction behavior record streams and platform interface layout snapshot streams, and a first causal influence relationship topology graph is generated according to the processing flow from steps S110 to S140. The performance index subsequences during off-peak operating periods are combined with the corresponding time period data, and a second causal influence relationship topology graph is generated according to the same process. For each layout component instance, the causal influence intensity value Ipeak is extracted from the first causal influence relationship topology graph, and the causal influence intensity value Ioff is extracted from the second causal influence relationship topology graph. The peak-valley causal difference parameter Dpv = |Ipeak - Ioff| / (Ipeak + Ioff + ε), where ε is a very small positive number.

[0159] Step S1230: Identify the layout component instances whose peak-valley causal difference parameters exceed a preset difference threshold as peak-valley sensitive layout component instances, and extract the first causal transmission path set of the peak-valley sensitive layout component instances during peak operating hours and the second causal transmission path set during off-peak operating hours.

[0160] The peak-valley causal difference parameter of all layout component instances is compared with a preset difference threshold. Layout component instances with a peak-valley causal difference parameter greater than the preset difference threshold are selected and marked as peak-valley sensitive layout component instances. From the first causal influence topology graph, starting from the node corresponding to the peak-valley sensitive layout component instance, trace back along the incoming edge direction to the root layout component instance, and extract all complete causal transmission paths to form the first causal transmission path set. The second causal transmission path set is extracted from the second causal influence topology graph in the same way.

[0161] Step S1240: Perform path difference comparison processing on the first causal transmission path set and the second causal transmission path set, identify peak-valley unique causal transmission paths that only appear during peak operation periods and not during off-peak operation periods, and mark the intermediate layout component instances on the peak-valley unique causal transmission paths as peak-valley triggered layout component instances.

[0162] Each path in the first causal transmission path set is matched with all paths in the second causal transmission path set by comparing the edit distance of the layout component instance sequences on the path. If the edit distance of a path in the first causal transmission path set exceeds a preset similarity threshold with all paths in the second causal transmission path set, then the path is determined to be a peak-valley unique causal transmission path. Intermediate layout component instances, excluding the root and terminal nodes, are extracted from the peak-valley unique causal transmission path and identified as peak-valley triggered layout component instances.

[0163] Step S1250: Based on the interaction behavior sequence of the peak-valley triggered layout component instance in the joint observation sequence of component interaction performance, extract the interaction frequency change characteristics and interaction type change characteristics of the peak-valley triggered layout component instance between peak operation period and off-peak operation period, and generate a peak-valley triggered behavior feature description.

[0164] The interaction behavior sequence fragments of peak-valley triggered layout component instances during peak and off-peak operating periods are extracted from the joint observation sequence of component interaction performance. The interaction frequency variation feature is the ratio of interaction frequency during peak to that during off-peak operating periods. The interaction type variation feature is the KL divergence value between the interaction type distributions during peak and off-peak operating periods. The interaction frequency variation feature and the interaction type variation feature are combined to form a peak-valley triggered behavior feature description.

[0165] Step S1260: Based on the peak-valley triggering behavior feature description, generate a peak-valley adaptive interface layout strategy for the peak-valley triggering layout component instance. The peak-valley adaptive interface layout strategy includes an interface layout pre-adjustment operation performed on the peak-valley triggering layout component instance when triggered during peak operation periods and an interface layout restoration operation performed on the peak-valley triggering layout component instance when triggered during off-peak operation periods.

[0166] The interface layout pre-adjustment operation is determined based on the interaction frequency change characteristics. If the interaction frequency change characteristic is greater than 1, the scaling ratio of the component size of the peak-valley triggered layout component instance is increased, with the increase proportional to the interaction frequency change characteristic. If the KL divergence value in the interaction type change characteristic exceeds the preset divergence threshold, the component color transparency value of the peak-valley triggered layout component instance is reduced to improve its visual prominence. The interface layout restoration operation restores the component size scaling ratio and component color transparency value to the default configuration values ​​during off-peak operation periods.

[0167] Step S1270: According to the peak-valley adaptive interface layout strategy, embed the operation period condition judgment logic into the service process reconstruction scheme. When the operation period is detected to switch from off-peak operation period to peak operation period, the interface layout pre-adjustment operation is automatically executed. When the operation period is detected to switch from peak operation period to off-peak operation period, the interface layout restoration operation is automatically executed, thereby generating a peak-valley adaptive dynamic service process reconstruction scheme.

[0168] The service process refactoring solution adds an operational time period condition judgment logic module, which includes a time period switch detector and a strategy executor. The time period switch detector continuously monitors the matching relationship between the current system time and the preset operational time period division strategy. When it detects a switch from an off-peak operational time period to a peak operational time period, it triggers the strategy executor to perform a pre-adjustment operation on the interface layout of peak-valley triggered layout component instances. When it detects a switch from a peak operational time period to an off-peak operational time period, it triggers the strategy executor to perform an interface layout restoration operation. The service process refactoring solution with the added operational time period condition judgment logic is a peak-valley adaptive dynamic service process refactoring solution.

[0169] Step S1310: Extract the causal influence intensity value of each layout component instance in the causal influence relationship topology graph generated in multiple consecutive time windows, perform differential calculation on the causal influence intensity values ​​of the same layout component instance in adjacent time windows, and generate the causal influence intensity change gradient sequence of the layout component instance.

[0170] The causal influence intensity value of each layout component instance is extracted from the causal influence relationship topology graph corresponding to each consecutive time window and sorted in ascending order by time window number. The difference between the two causal influence intensity values ​​at time window numbers k and k+1 is calculated; the difference is the (k+1)th causal influence intensity value minus the kth causal influence intensity value. The difference values ​​of all adjacent time windows are sequentially arranged to form a gradient sequence of causal influence intensity changes.

[0171] Step S1320: Perform trend extrapolation prediction processing on the gradient sequence of causal influence intensity change, use a preset trend extrapolation prediction algorithm to calculate the predicted causal influence intensity value of the layout component instance in the future time window, and mark the layout component instance whose predicted causal influence intensity value exceeds the preset warning threshold as a causal risk precursor component.

[0172] The Holt exponential smoothing algorithm is used to extrapolate and predict the trend of the gradient sequence of causal influence intensity changes. The Holt exponential smoothing algorithm consists of two recursive equations: a level estimation equation and a trend estimation equation. The level estimate is Lk+1 = α × Ik + (1-α) × (Lk + Tk), and the trend estimate is Tk+1 = β × (Lk+1 - Lk) + (1-β) × Tk, where Ik is the causal influence intensity value in the k-th time window, and α and β are smoothing parameters between 0 and 1. The predicted causal influence intensity value in the (k+m)-th time window is Lk+m × Tk. Layout component instances whose predicted causal influence intensity values ​​exceed a preset warning threshold are marked as causal risk precursor components.

[0173] Step S1330: Extract all downstream layout component instances of the causal risk precursor component in the causal influence relationship topology graph, and calculate the future prediction performance degradation magnitude and prediction performance degradation time point of each downstream layout component instance based on the causal influence strength and causal propagation delay parameter between the causal risk precursor component and each downstream layout component instance.

[0174] Starting from the node corresponding to the causal risk precursor component in the causal influence topology graph, a breadth-first traversal is performed along the directed edges to collect all reachable downstream layout component instances. For each downstream layout component instance, the product of the causal association strength coefficients of all directed edges on the path from the causal risk precursor component to that downstream layout component instance is calculated as the propagation attenuation coefficient. The future performance degradation magnitude is equal to the predicted causal influence strength value multiplied by the propagation attenuation coefficient. The predicted performance degradation time point is equal to the current time plus the sum of the causal propagation delay parameters of all directed edges on the path.

[0175] Step S1340: Identify downstream layout component instances whose future predicted performance degradation exceeds a preset degradation tolerance threshold as causal risk affected components, and sort all causal risk affected components by time according to the predicted performance degradation time point to generate a time sorting list of causal risk affected components.

[0176] The projected future performance degradation of all downstream layout component instances is compared with a preset degradation tolerance threshold. Downstream layout component instances whose predicted future performance degradation exceeds the preset threshold are identified as components affected by causal risk. All components affected by causal risk are then sorted in ascending order of their predicted performance degradation time points, generating a time-sorted list of components affected by causal risk.

[0177] Step S1350: Based on the time-sorted list of components affected by causal risk, generate a pre-intervention protection strategy with time priority. The pre-intervention protection strategy assigns a pre-intervention protection operation time window and a pre-intervention protection operation type to each component affected by causal risk according to the order of predicted performance degradation time points.

[0178] For each causal risk-affected component in the time-sorted list, the start time of the pre-intervention protection operation time window is shifted forward by a preset lead time to its predicted performance degradation time point, and the end time of the pre-intervention protection operation time window is shifted backward by a preset delay time to its predicted performance degradation time point. If the number of incoming edges in the causal influence relationship topology graph of the causal risk-affected component is greater than the number of outgoing edges, the pre-intervention protection operation type is assigned as service resource pre-allocation. If the number of outgoing edges is greater than the number of incoming edges, the pre-intervention protection operation type is assigned as interface layout pre-adjustment. If both are equal, both operation types are assigned simultaneously.

[0179] Step S1360: For causal risk affected components whose pre-intervention protection operation type is interface layout pre-adjustment, generate an interface layout pre-adjustment operation instruction, wherein the interface layout pre-adjustment operation instruction includes the pre-migration target coordinates and pre-adjustment visual presentation attributes of the layout component instance; for causal risk affected components whose pre-intervention protection operation type is service resource pre-allocation, generate a service resource pre-allocation operation instruction, wherein the service resource pre-allocation operation instruction includes the pre-allocation service resource quota parameters of the corresponding backend service node of the causal risk affected component.

[0180] The pre-migration target coordinates of the layout component instance in the interface layout pre-adjustment operation instruction offset the layout component instance away from components with causal risk indicators, with the offset distance proportional to the future predicted performance degradation. The pre-adjustment visual presentation attribute increases the component color transparency value of the layout component instance, with the increase proportional to the urgency of the predicted performance degradation time point. The pre-allocated service resource quota parameter in the service resource pre-allocation operation instruction includes an additional allocated service resource quota, with the quota proportional to the future predicted performance degradation of the component affected by the causal risk.

[0181] Step S1370: Assemble the interface layout pre-adjustment operation instructions and service resource pre-allocation operation instructions into a pre-intervention protection instruction sequence according to the time sequence of the pre-intervention protection operation time window, and integrate the pre-intervention protection instruction sequence into the service process reconstruction scheme to generate a preventive service process reconstruction scheme with causal risk pre-intervention capability.

[0182] All interface layout pre-adjustment operation instructions and service resource pre-allocation operation instructions are arranged in ascending order according to the start time of their pre-intervention protection operation time window, forming a pre-intervention protection instruction sequence. A pre-intervention protection configuration section is added to the service process refactoring scheme. This section contains the pre-intervention protection instruction sequence and the corresponding timed triggering mechanism. The service process refactoring scheme with the added pre-intervention protection configuration section is a preventative service process refactoring scheme with causal risk pre-intervention capabilities.

[0183] Figure 2This document showcases the "Service Health Monitoring Overview" interface of the visualization platform operation service optimization method combining a multimodal large model, as presented in this invention. This interface serves as the visual entry point for the synchronous collection of multi-source heterogeneous operational data streams in step S110. Key indicator cards at the top of the interface, such as "Service Health Score," "Number of Active Components," and "Pending Alarms," ​​intuitively reflect the platform's current overall operational status. The function entries below, such as "Causal Topology Diagram" and "Bottleneck Location," correspond to the core functional modules for generating the causal influence relationship topology diagram in subsequent step S140 and identifying root cause layout component instances in step S150. The "Real-time Alarms" list at the bottom of the interface directly displays abnormal events detected in the platform service performance monitoring indicator stream, such as "increased response latency in the retrieval component" and "coupling oscillations in the data details component." These alarm events are crucial signal sources triggering subsequent causal inference and root cause analysis.

[0184] Figure 3 This document demonstrates the interactive interface of the "Causal Influence Relationship Topology Diagram" during the execution of steps S140 and S150 of this invention. The interface graphically presents the causal influence relationship topology diagram generated in step S140, where layout component instances such as the "Data Retrieval Bar" and "Results List Component" are abstracted as topology diagram nodes. The directed connection edges between nodes and the response delays of the annotations intuitively reflect the causal transmission relationship between components. The "Results List Component" is highlighted as a high-risk node, corresponding to the root cause layout component instance identified in step S150. The dashed boxes for "Tightly Coupled Regions" and the "Oscillation Path" markers in the interface correspond to the causal coupling component groups identified in steps S520 to S530 and the co-frequency resonance component groups and causal transmission oscillation amplification loops identified in steps S330 to S360, respectively. The "Node Details" pop-up at the bottom displays the "Number of Downstream Influencing Components" and "Radiation Source" attributes of the component, corresponding to the calculation results of the causal influence radiation range parameters in step S410, providing users with targeted root cause location information.

[0185] Figure 4This document demonstrates the interface for generating the "bottleneck location" and "service process refactoring scheme" during the execution of steps S150 and S1010 to S1040 of this invention. The top of the interface clearly identifies the "result list component" as the root bottleneck component and quantifies its "bottleneck severity" and "number of downstream components affected," corresponding to the core logic of identifying the root component based on the causal influence intensity distribution in step S150. The "bottleneck influence analysis" area in the middle displays multiple causal transmission paths originating from the root component, marking the path length and priority, corresponding to the causal path extraction and cumulative causal influence intensity calculation in steps S210 to S220. The "cross-path cluster common bottleneck" area below displays the cluster common bottleneck nodes identified in steps S240 to S250 and their priority ranking. The "Service Process Restructuring Plan" at the bottom of the interface lists specific optimization measures, such as "adjusting the display position", "implementing service rate limiting" and "optimizing the preloading strategy". These plans are the specific reconstruction instructions generated by step S150 based on the upstream and downstream related paths, and are finally sent to the backend configuration system through the "trigger linkage update" button at the bottom.

[0186] Figure 5 This document demonstrates the "Solution Comparison" and "Historical Experience Reuse" interfaces during the execution of steps S1010 to S1040 of this invention. This interface is a concrete application of topology similarity matching and historical fault experience reuse in step S1010. The "Topology Pattern Matching Result" area at the top of the interface compares the current causal relationship topology map with historical similar fault patterns and provides a quantitative result of "97% similarity," corresponding to the process in step S1020 where known repair strategies are extracted when the topology similarity exceeds a preset matching threshold. The "Strategy Consistency Comparison" table in the middle clearly lists the differences between the currently generated service process reconstruction scheme and historical successful schemes. For example, "Adding debouncing processing to the filtering panel" and "Adding queue buffering to the exported service" are marked as "Not Covered," which completely corresponds to the strategy consistency comparison analysis between candidate repair strategies and service process reconstruction schemes in step S1030. The "Enhanced Service Process Restructuring Solution" area at the bottom of the interface displays the complete solution that integrates "causal reasoning generation" and "historical experience supplementation". The historical experience supplementation operation provides optimization points that are not covered by the current solution, ultimately forming an enhanced solution with 100% completeness. The "Execute Enhanced Service Process Restructuring Solution" button triggers a linked update operation.

[0187] Based on the same inventive concept, the visualization platform operation service optimization system combining multimodal large models provided in this application includes a central processing unit, a system memory including random access memory and read-only memory, and a system bus connecting the system memory and the central processing unit. The visualization platform operation service optimization system combining multimodal large models also includes a basic input / output system to facilitate information transmission between various devices within the computer, and a large-capacity storage device for storing the operating system, applications, and other program modules.

[0188] A basic input / output system includes a display for showing information and input devices such as a mouse and keyboard for user input. Both the display and the input devices are connected to the central processing unit via an input / output controller connected to the system bus. The basic input / output system may also include an input / output controller for receiving and processing input from multiple other devices such as a keyboard, mouse, or electronic stylus. Similarly, the input / output controller also provides output to a display screen, printer, or other types of output devices.

[0189] A mass storage device is connected to the central processing unit via a mass storage controller connected to the system bus. The mass storage device and its associated computer-readable medium provide non-volatile storage for the visualization platform operation service optimization system incorporating a multimodal mass model. Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. According to various embodiments of this application, the visualization platform operation service optimization system incorporating a multimodal mass model can also be connected to a remote computer on a network, such as the Internet. That is, the visualization platform operation service optimization system incorporating a multimodal mass model can be connected to a network via a network interface unit connected to the system bus, or it can use a network interface unit to connect to other types of networks or remote computer systems.

[0190] In addition, in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0191] The above are merely exemplary embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application shall be included within the protection scope of this application.

Claims

1. A method for optimizing the operation and service of a visualization platform that combines a multimodal large model, characterized in that, The method includes: During the operation of the visualization platform, user interaction behavior records, platform interface layout snapshots, and platform service performance monitoring metrics are collected simultaneously. The platform interface layout snapshot stream is processed by layout component instance segmentation to obtain layout component instance segmentation results. The user interaction behavior record stream is associated and bound according to the layout component instance segmentation results to generate a binding relationship mapping table. Based on the binding relationship mapping table, the interaction behavior sequence and the corresponding local performance index sequence of each layout component instance within a continuous time window are extracted and combined into a joint observation sequence of component interaction performance at the layout component granularity. The joint observation sequence of component interaction performance is input into a pre-built multimodal large model. Cross-modal causal reasoning is performed on the joint observation sequence of component interaction performance to generate a causal influence relationship topology graph between each layout component instance. The causal influence relationship topology graph is constructed with layout component instances as nodes and causal transmission relationships of interaction performance between components as directed edges. Based on the causal influence intensity distribution of each layout component instance in the causal influence topology diagram, the root layout component instance that causes the platform service performance degradation is identified. After generating a service process reconstruction scheme based on the upstream and downstream association paths of the root layout component instance in the causal influence topology diagram, the scheme is sent to the background configuration system of the visualization platform to trigger the linkage update operation of the interface layout and service configuration.

2. The method for optimizing the operation service of a visualization platform combining a multimodal large model as described in claim 1, characterized in that, The process of segmenting the platform interface layout snapshot stream into layout component instances to obtain layout component instance segmentation results, and then associating and binding the user interaction behavior record stream according to the layout component instance segmentation results to generate a binding relationship mapping table, includes: Extract the interface rendering tree structure of each frame of the platform interface layout snapshot in the platform interface layout snapshot stream, perform a depth-first traversal on the interface rendering tree structure, extract the node type identifier, node bounding box coordinates and node level depth value of each interface rendering tree node, and generate an interface rendering tree node attribute set. The set of attributes of the interface rendering tree node is input into the pre-trained layout component instance segmentation model. The interface rendering tree node is semantically grouped by the layout component instance segmentation model. Multiple interface rendering tree nodes belonging to the same functional component are aggregated into layout component instances, generating the component instance identifier, component instance bounding box coordinate set and component instance type identifier for each layout component instance. Using the set of component instance bounding box coordinates of the layout component instance, mark the spatial location region of each layout component instance in the platform interface layout snapshot stream, and generate a platform interface layout snapshot sequence with component instance spatial annotations; Extract the interaction coordinates, interaction timestamp, and interaction type identifier of each user interaction event in the user interaction behavior record stream. Perform spatial coordinate matching processing between the interaction coordinates and the platform interface layout snapshots with the same timestamp in the platform interface layout snapshot sequence with component instance spatial annotation to determine the target layout component instance corresponding to each user interaction event. Based on the target layout component instance corresponding to each user interaction event, establish an association binding relationship between the interaction event identifier and the layout component instance identifier, and combine the interaction event identifier, layout component instance identifier, interaction timestamp, and interaction type identifier into an association binding record; All associated binding records are sorted in ascending order by interaction timestamp to generate a binding relationship mapping table. Each record in the binding relationship mapping table uniquely identifies the correspondence between a user interaction event and the layout component instance it affects.

3. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, Based on the binding relationship mapping table, the interaction behavior sequence and corresponding local performance index sequence of each layout component instance within a continuous time window are extracted and combined into a joint observation sequence of component interaction performance at the layout component granularity, including: Based on the binding relationship mapping table, all associated binding records are grouped and aggregated using the layout component instance identifier as the grouping primary key, and all associated binding records belonging to the same layout component instance identifier are merged into a layout component instance interaction record group. For each layout component instance, the interaction records are grouped, and the interaction type identifier sequence and interaction time interval sequence corresponding to the layout component instance are extracted in ascending order along a unified timestamp. The interaction type identifier sequence and interaction time interval sequence are combined into the interaction behavior sequence of the layout component instance. Based on the platform service performance monitoring indicator stream, extract the local service response time sequence, local service request frequency sequence, and local resource occupancy rate sequence corresponding to the spatial location region of each layout component instance, and combine the local service response time sequence, local service request frequency sequence, and local resource occupancy rate sequence into a local performance indicator sequence for that layout component instance. The interaction behavior sequence and local performance index sequence of each layout component instance are aligned with a sliding window in the time dimension. The continuous time window is divided according to the preset window duration parameter and window sliding step parameter. Interaction behavior sequence fragments and local performance index sequence fragments are extracted in each continuous time window. The interaction behavior sequence segments and local performance index sequence segments extracted within the same continuous time window are horizontally spliced ​​together to generate a joint observation segment of the component interaction performance of the layout component instance within the continuous time window. The joint observation segments of the component interaction performance of all continuous time windows are arranged in chronological order to form a joint observation sequence of component interaction performance at the layout component level.

4. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, The step involves inputting the joint observation sequence of component interaction performance into a pre-constructed multimodal large model, and generating a topological graph of causal influence relationships between various layout component instances by performing cross-modal causal inference processing on the joint observation sequence of component interaction performance. This includes: The joint observation sequence of component interaction performance of all layout component instances is input into the multimodal feature encoding layer of the multimodal large model. Temporal feature extraction processing is performed on the interaction behavior sequence of each layout component instance to generate an interaction behavior semantic encoding vector. Performance feature extraction processing is performed on the local performance index sequence of each layout component instance to generate a performance index encoding vector. The interaction behavior semantic encoding vector and performance index encoding vector are input into the cross-modal fusion layer of the multimodal large model. The interaction behavior semantic encoding vector and performance index encoding vector are subjected to intermodal information interaction fusion processing to generate a cross-modal fusion representation vector for each layout component instance. Input the cross-modal fusion representation vectors of all layout component instances into the causal reasoning analysis module of the multimodal large model, perform Granger causality test on the cross-modal fusion representation vectors between all layout component instances, and generate the causal association strength coefficients between the pairs of layout component instances. The direction of directional causal propagation is determined by processing the timing-ahead behavior and performance metric response lag of the interaction behavior between each pair of layout component instances. Using each layout component instance as a node, the causal relationship between any two layout component instances with a causal relationship strength coefficient exceeding a preset causal threshold is used as a directed edge, and the direction of directed causal propagation is used as the edge direction, a topology graph of causal influence relationships between each layout component instance is generated.

5. The method for optimizing the operation service of a visualization platform combining a multimodal large model as described in claim 1, characterized in that, The method further includes: Extract the causal transmission path of each layout component instance in the causal influence relationship topology graph, trace the entire sequence of layout component instances from the root layout component instance to the end layout component instance along the directed edge direction of each causal transmission path, and generate a set of causal transmission paths. For each causal transmission path in the set of causal transmission paths, the cumulative causal influence strength value of the path is calculated based on the causal influence strength between adjacent layout component instances on the path, and the path length value is calculated based on the number of layout component instances on the path. A preset causal path clustering analysis algorithm is used to perform path structure similarity clustering on all causal transmission paths in the causal transmission path set, and causal transmission paths with similar causal transmission chain structures are aggregated into the same causal transmission path cluster, generating multiple causal transmission path clusters; For each causal transmission path cluster, extract the layout component instances that all causal transmission paths in the cluster pass through as the cluster common bottleneck node, and generate the bottleneck severity parameter of the cluster common bottleneck node based on the average causal influence intensity of the cluster common bottleneck node on each path in its respective causal transmission path cluster. The cluster common bottleneck nodes and corresponding bottleneck severity parameters of all causal transmission path clusters are comprehensively sorted to generate a bottleneck node priority sequence across path clusters. The cluster common bottleneck node ranked first in the bottleneck node priority sequence of cross path clusters is identified as the core bottleneck node of the cross path cluster. Based on the topological position of the cross-path cluster core bottleneck node in the causal influence relationship topology graph, extract all incoming edge connection layout component instances and all outgoing edge connection layout component instances of the cross-path cluster core bottleneck node to generate a set of upstream and downstream related layout components for the cross-path cluster core bottleneck node. The interface layout reconstruction diagram of the upstream and downstream related layout component set of the core bottleneck node of the cross-path cluster is generated. The optimal spatial arrangement coordinates and optimal visual presentation attributes of each layout component instance in the upstream and downstream related layout component set are calculated to generate a local reconstruction scheme of the interface layout of the cross-path cluster. The cross-path cluster interface layout partial reconstruction scheme and the service process reconstruction scheme are merged to generate a collaborative service process reconstruction scheme that integrates multi-causal path cluster bottleneck optimization. The collaborative service process reconstruction scheme that integrates multi-causal path cluster bottleneck optimization is then sent to the background configuration system of the visualization platform to trigger the interface layout update operation of multi-causal path cluster linkage.

6. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, The method further includes: Extract the causal propagation delay parameter of each layout component instance in the causal influence relationship topology graph. The causal propagation delay parameter represents the time delay length experienced from the interaction behavior change of the upstream layout component instance to the performance index change of the downstream layout component instance. Based on the causal transmission delay parameter and the directed edge structure of the causal influence relationship topology graph, the causal influence relationship topology graph is mapped into a causal temporal transmission network graph, and each directed edge in the causal temporal transmission network graph is marked with a corresponding causal transmission delay parameter; Oscillation mode analysis is performed on the causal temporal transmission network diagram to identify a set of oscillating layout component instances in the causal temporal transmission network diagram that exhibit periodic fluctuations in causal transmission delay. The fluctuation amplitude of the causal transmission delay parameter of each oscillating layout component instance in the set of oscillating layout component instances exceeds a preset fluctuation threshold within adjacent time windows. Extract the causal propagation delay fluctuation sequence of each oscillating layout component instance in the set of oscillating layout component instances, perform fluctuation period analysis on the causal propagation delay fluctuation sequence, and extract the fluctuation period length and fluctuation phase offset parameters of the oscillating layout component instance. The oscillation layout component instances with the same fluctuation period length and a difference in fluctuation phase offset parameter less than a preset phase difference threshold are aggregated into a same-frequency resonance component group to generate multiple same-frequency resonance component groups. For each resonant component group, the directed edge connection relationship of all oscillating layout component instances in the causal time-series transmission network diagram is extracted, and the causal transmission oscillation amplification loop inside the resonant component group is identified. The causal transmission oscillation amplification loop is a closed transmission path that forms oscillation enhancement by the periodic superposition of causal transmission delays between oscillating layout component instances. For each causal propagation oscillation amplification loop, an oscillation breakpoint analysis is performed. The layout component instance with the largest fluctuation amplitude of the causal propagation delay parameter in the causal propagation oscillation amplification loop is selected as the oscillation breakpoint, and a time-series oscillation suppression strategy is generated for the oscillation breakpoint. According to the temporal oscillation suppression strategy, the interaction behavior sequence and local performance index sequence of the layout component instance corresponding to the oscillation breakpoint are temporally decoupled to generate the decoupled interaction behavior sequence and decoupled local performance index sequence of the layout component instance corresponding to the oscillation breakpoint. The decoupled interaction behavior sequence and decoupled local performance index sequence are then re-input into the causal reasoning analysis module of the multimodal large model for secondary causal reasoning processing to generate the causal influence relationship topology graph after oscillation suppression.

7. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, The method further includes: Extract the causal influence radiation range parameter of each layout component instance in the causal influence relationship topology graph. The causal influence radiation range parameter is determined based on the number of all downstream layout component instances reachable from the layout component instance along the directed edge and the cumulative causal influence intensity of the layout component instance on these downstream layout component instances. Based on the causal influence radiation range parameter of each layout component instance, layout component instances whose causal influence radiation range parameter exceeds a preset radiation threshold are identified in the causal influence relationship topology graph as causal influence radiation source components, and a set of causal influence radiation source components is generated. For each causal influence radiation source component in the set of causal influence radiation source components, extract all downstream radiation paths of the causal influence radiation source component in the causal influence relationship topology graph. The downstream radiation path is the complete path from the causal influence radiation source component along the directed edge to the radiation boundary layout component instance. Based on the causal influence intensity of each layout component instance on each downstream radiation path, the radiation attenuation gradient of that downstream radiation path is calculated. The radiation attenuation gradient represents the rate of change of the causal influence intensity attenuating step by step along the downstream radiation path. All downstream radiation paths of the causal radiation source component are sorted according to the order of radiation attenuation gradient from steep to gentle, and the downstream radiation path with the steepest radiation attenuation gradient is identified as the key intervention radiation path. Based on the path structure and radiation attenuation gradient distribution of the key intervention radiation path, the layout component instance with the largest radiation attenuation gradient change rate is selected as the radiation blocking node on the key intervention radiation path, and a radiation blocking strategy is generated for the radiation blocking node. The radiation blocking strategy includes the interactive function rate limiting parameters and performance resource isolation parameters of the radiation blocking node. The radiation blocking strategy is integrated into the service process reconstruction scheme to generate a service process reconstruction scheme that integrates radiation blocking. This scheme is then sent to the background configuration system of the visualization platform to trigger a coordinated update of the interface layout, service configuration, and causal radiation blocking.

8. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, The method further includes: Extract all layout component instance pairs with bidirectional causal transmission relationships in the causal influence relationship topology graph. The bidirectional causal transmission relationship refers to the presence of mutually pointing directed edges between two layout component instances. Layout component instance pairs with bidirectional causal transmission relationships are identified as causal coupling component pairs. For each causal coupling component pair, calculate the causal coupling tightness parameter between the two layout component instances in the causal coupling component pair. The causal coupling tightness parameter is determined based on the sum of the causal influence strengths on the directed edges of the bidirectional causal transmission relationship. Causal coupling component pairs whose causal coupling tightness parameter exceeds the preset coupling threshold are aggregated into the same causal coupling component group. The causal coupling component pairs with shared layout component instances are merged into causal coupling component groups through the connected component analysis algorithm, generating multiple causal coupling component groups. For each causal coupled component group, extract the interaction behavior sequence and local performance index sequence of all layout component instances in the causal coupled component group in the joint observation sequence of component interaction performance, and input the interaction behavior sequence and local performance index sequence of all layout component instances in the causal coupled component group into the pre-built coupled component collaborative optimization model; The coupling effect analysis layer in the coupled component collaborative optimization model performs coupling effect analysis on the interaction behavior collaboration mode and performance index linkage mode between the layout component instances in the causal coupled component group, and generates coupling effect analysis results. Based on the coupling effect analysis results, a coupling decoupling strategy is generated for the causal coupling component group through the coupling decoupling strategy generation layer in the coupling component collaborative optimization model. The coupling decoupling strategy includes an interface layout splitting scheme that splits the causal coupling component group into multiple independent layout component instances and a service call chain splitting scheme that decouples the background service call chain of the causal coupling component group. The coupling and decoupling strategies of each causal coupling component group are integrated with the service process reconstruction scheme to generate a collaborative service process reconstruction scheme that integrates coupling and decoupling. The collaborative service process reconstruction scheme that integrates coupling and decoupling is sent to the background configuration system of the visualization platform to trigger the linkage update operation of interface layout, service configuration and causal coupling and decoupling.

9. The method for optimizing the operation service of a visualization platform combining a multimodal large model according to claim 1, characterized in that, The method further includes: Extract the temporal variation characteristics of the causal influence intensity of each layout component instance in the causal influence relationship topology graph, compare the causal influence intensity of the same layout component instance in the causal influence relationship topology graph generated by different continuous time windows in a temporal sequence, and generate the causal influence intensity evolution sequence of each layout component instance. For each layout component instance, the causal influence intensity evolution sequence is processed to detect trend turning points, identify the trend turning points in the causal influence intensity evolution sequence where the direction of change of causal influence intensity changes, and mark the causal influence intensity evolution sequence segments within a preset time range before and after the trend turning points as causal mutation segments. Align the causal mutation fragments of all layout component instances according to the trend turning point, identify layout component instances that have causal mutations within the same trend turning point or adjacent trend turning point range, and aggregate these layout component instances into a causal mutation synchronization component set. For the layout component instances in the causal mutation synchronization component set, backtrack the interaction behavior sequence and local performance index sequence at the corresponding trend turning point in the joint observation sequence of component interaction performance, and extract the abnormal interaction pattern features in the interaction behavior sequence and the abnormal performance fluctuation features in the local performance index sequence. The abnormal interaction pattern features and abnormal performance fluctuation features are input into the pre-constructed causal mutation tracing analysis model. The causal mutation tracing analysis model is used to perform correlation tracing reasoning on the abnormal interaction pattern features and abnormal performance fluctuation features to identify the initial triggering layout component instance and initial triggering interaction event that triggers the causal mutation. Based on the initial trigger layout component instance and the initial trigger interaction event, a pre-interception strategy for the interaction event of the initial trigger layout component instance is added to the service process reconstruction scheme. The pre-interception strategy for the interaction event includes the identification rules for the initial trigger interaction event and the interception response operation. The service process refactoring scheme, which adds a pre-interception strategy for interactive events, is sent to the backend configuration system of the visualization platform. The backend configuration system will then execute a pre-interception response operation when it detects the initial triggering interactive event to prevent the further propagation of causal mutations.

10. A visualization platform operation service optimization system combining multimodal large models, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the visualization platform operation service optimization method combining a multimodal large model as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images