Business data processing method and system based on visual lane modeling

Abstract

The invention relates to the technical field of visual lane modeling, and discloses a business data processing method and system based on visual lane modeling, and a multi-task prediction model based on multi-source historical business chain data training graph neural network and Transform fusion. A plurality of groups of lane layout and event configuration schemes are generated in combination with user real-time operation captured by a design interface, an optimal scheme is screened through simulation evaluation and then is visually displayed, finally, visual lane modeling is executed according to user feedback, and through automatic data processing, high-precision model prediction, real-time interaction optimization and visual modeling, the optimal lane layout and event configuration scheme is obtained. According to the method, the service chain design efficiency and quality are improved, meanwhile, artificial experience deviation is avoided through model prediction based on historical data, the comprehensive requirements of complex equipment development for real-time performance, reliability and resource efficiency are met, and efficient technical support is provided for landing of model-based system engineering in the field of complex equipment.

Description

Technical Field

[0001] This invention relates to the field of visual swimlane modeling technology, and in particular to a business data processing method and system based on visual swimlane modeling. Background Technology

[0002] In the field of complex equipment development, Model-Based Systems Engineering (MBSE) has become the mainstream paradigm for business chain design. Its core requirement is to achieve a balance of multi-dimensional performance indicators (latency, resources, reliability) and rapid adaptation to dynamic business requirements. Currently, mainstream business chain design tools still rely on the traditional drag-and-drop and manual configuration mode (manually building and dragging swimlanes and operation units within swimlanes, and manually configuring event types between operation units to build a visual business carrier canvas for different service requirements). Although this can meet basic visualization needs, it has significant limitations in large-scale, highly complex business scenarios.

[0003] Existing business chain design technologies are mainly divided into two categories: one is based on static visualization tools (such as IBM Rhapsody), which express business logic through UML sequence diagrams and flowcharts. However, these tools can only describe simple call relationships and cannot distinguish event types (such as periodic events / conditional events). Moreover, the swimlane positions are fixed, making it difficult to dynamically reconstruct the process. The other category is a customized design method based on human experience. It relies on designers' understanding of the business to configure swimlane layout and event relationships. However, this method has three major drawbacks: First, design decisions are highly dependent on experience and lack historical data support, resulting in poor consistency of solutions produced by different designers. Second, as the complexity of the business chain increases (such as the number of swimlanes > 10 and event relationships > 20), the efficiency of manual adjustment decreases exponentially and it is difficult to guarantee global optimization. Third, it lacks adaptive learning capabilities and cannot build predictive models based on historical design data and performance logs. Swimlane design directly affects the underlying system's response time, resource utilization, and other performance indicators by determining data transmission paths, resource scheduling logic, and event interaction complexity. If problems can only be discovered through testing after the design is completed, the cost of later iterations accounts for a high proportion of the total design cost.

[0004] While some current improvement solutions attempt to introduce automated configuration functions, they still fail to address issues such as insufficient feature compactness, high real-time interaction latency, and poor adaptability to multiple scenarios. For example, existing solutions cannot capture user design operation features in real time and quickly generate optimization suggestions, and lack an online model update mechanism. This results in the need for manual rule readjustment in different scenarios such as avionics systems (high real-time requirements) and industrial monitoring (high reliability), leading to long adaptation cycles. Therefore, the industry urgently needs a business data processing solution based on visual swimlane modeling to improve the automation and accuracy of business chain design while ensuring low latency and low memory consumption. Summary of the Invention

[0005] This invention provides a business data processing method and system based on visual swimlane modeling, aiming to solve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides a business data processing method based on visual swimlane modeling, the method comprising the following steps: S1: Collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and construct a training sample set; S2: Based on the fusion architecture of graph neural network and Transformer, a multi-task prediction model is constructed. By combining the loss function and the constructed training sample set to train the model parameters, a multi-task prediction model with the ability to predict swimlane layout and event configuration is obtained. S3: Capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate standardized input vectors; S4: Input the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes, and select the optimal scheme set through simulation evaluation; S5: Display the optimal solution group in a visual overlay on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data.

[0007] Optionally, step S1: Collect historical business chain design data from the multi-source system, perform standardization and time-series alignment, and construct a training sample set, specifically including: S11: Parse the version history of the business chain design file from the Git version control system, extract swimlane layout data L, and organize it into CSV format data according to version number and timestamp; wherein, the swimlane layout data includes swimlane ID, service name, service type, center coordinates, width and hierarchical relationship; S12: Parse the event definition section in the design file, construct the event graph, take service operations as nodes and event relationships as directed edges, generate event graph data G=(V,E), and transform the event graph data G into an adjacency matrix A and an edge feature matrix E_feat; S13: Collect performance indicator data P from the system test log database; wherein, the performance indicator data P includes event response time, packet loss rate, CPU utilization, and memory usage; S14: Based on the timestamp of the design version, the swimlane layout data L, event graph data G, and performance index data P are time-aligned to form a single sample. The single sample is then standardized using Z-score to obtain the training sample set.

[0008] Optionally, in step S2, a multi-task prediction model is constructed based on the fusion architecture of graph neural networks and Transformers, specifically including: S21: Construct a two-branch neural network architecture that includes a local prediction branch and an event configuration branch; The local prediction branch employs a graph attention network (GAT). The input layer receives the swimlane feature matrix L_feat extracted from the swimlane layout data L and the spatial adjacency matrix S. It learns the swimlane spatial dependencies through a two-layer graph attention GAT. The GAT has four attention heads and a hidden layer dimension of 64. The expression for the attention weights is as follows: ;

[0009] In the formula, Let represent the initial eigenvector of lane u. Represents a linear transformation matrix. Represents the attention weight vector. Denotes the neighborhood set of lane u. This represents a vector concatenation operation. Let represent the initial eigenvector of lane v. Describe the neighborhood set of lane u Any lane in the middle, Indicates swimming lane The initial feature vector; The event configuration branch uses a Transformer decoder. The input layer receives the node feature matrix V_feat and edge feature matrix E_feat extracted from the event graph data G. The event association pattern is learned through a 3-layer encoder. The output layer outputs the event type probability distribution through the Softmax activation function.

[0010] Optionally, in step S2, the model parameters are trained by combining the loss function and the constructed training sample set to obtain a multi-task prediction model with swimlane layout prediction and event configuration capabilities, specifically including: S22: Define the combination loss function, expressed as: ; ; ; ; In the formula, This represents the total loss value. Indicates the loss weight. Indicates the predicted loss from the layout. Indicates event classification loss. Indicates learning loss, Indicates the number of lanes. Represents the actual coordinates of lane i. This represents the predicted coordinates of lane i. Indicates the number of events. The label representing the true type of event j. This represents the predicted probability that event j belongs to type k. Indicates batch size, This represents the characteristic representation of event b. This represents a positive sample of event b. Indicates a negative sample. Represents cosine similarity. Indicates temperature parameter; S23: The Adam optimizer is used to train the model under the preset conditions of learning rate, weight decay, batch size and training epochs. After training, a multi-task prediction model with swimlane layout prediction and event configuration capabilities is obtained.

[0011] Optionally, step S3: Capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate a standardized input vector, specifically including: S31: In the design interface, a centralized event listener is set up to capture three types of operation events; wherein, the operation events include mouse drag events, service placement events, or connection operation events; S32: Call the service registry through the backend API to extract the metadata of the currently operating service and convert it into a service metadata feature vector; wherein, the metadata includes service ID, service type, maximum processing latency, CPU resource requirements, memory requirements, and interface protocol; S33: Traverse the current canvas lanes, extract spatial features and density features, and concatenate the spatial features, density features and service element feature vectors to generate a canvas state feature vector as a standardized input vector.

[0012] Optionally, step S4: Input the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes, and select the optimal scheme group through simulation evaluation, specifically including: S41: Input the standardized input vector into the layout prediction branch, and output the three lane position adjustment vectors with the highest confidence obtained by the layout prediction branch as layout schemes. Input the standardized input vector into the event configuration branch, and output the three configuration schemes with the highest probability obtained by the event configuration branch. S42: Cross-combination layout and configuration schemes form 9 initial candidate schemes, which are simulated in the SimPy simulation environment. Several core performance indicators of each scheme are output. The performance score is obtained by weighted summation of several core performance indicators. In the design constraint rule base, the design specification score is obtained by summing the mapping scores of several preset design specification standards and the configuration parameters of each scheme. S43: For the 9 initial candidate schemes, calculate the comprehensive performance score based on the performance score and the design specification score, and select the 5 schemes with the highest scores as the optimal scheme group; wherein, the expression for the comprehensive performance score is: ； In the formula, S represents the overall performance score. Indicates the performance score. This indicates the score for design specifications.

[0013] Optionally, S5: Display the optimal solution group visually overlaid on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data, specifically including: S51: Based on the overall performance score, each solution in the optimal solution group is superimposed on the canvas with a contour of preset transparency. S52: Mark the differences between the current design and the optimal design with the target color, and generate a detailed display panel on one side of the canvas to show the performance score and design specification score of the current optimal design; S53: Capture and record user feedback, and select the optimal solution to perform visual swimlane modeling based on the operation instructions corresponding to the user feedback data; wherein, the operation instructions include solution application instructions, solution switching instructions, or solution modification instructions corresponding to keyboard shortcut feedback data or page menu operation feedback data.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a business data processing system based on visual swimlane modeling, comprising: The module is used to collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and build a training sample set. The training module is used to build a multi-task prediction model based on the fusion architecture of graph neural network and Transformer. By combining the loss function and the constructed training sample set, the model parameters are trained to obtain a multi-task prediction model with the ability to predict swimlane layout and event configuration. The generation module is used to capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate standardized input vectors. The filtering module is used to input real-time standardized input vectors into the trained multi-task prediction model, generate multiple sets of swimlane layout and event configuration optimization schemes, and filter the optimal scheme group through simulation evaluation. The execution module is used to display the optimal solution group in a visual overlay on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data.

[0015] The beneficial effects of this invention are as follows: It proposes a business data processing method and system based on visualized swimlane modeling. A multi-task prediction model fused with graph neural networks and Transformer, trained based on multi-source historical business chain data, is generated by combining real-time user operations captured through the design interface to create multiple swimlane layout and event configuration schemes. After simulation evaluation and selection of the optimal scheme, the scheme is visualized. Finally, visualized swimlane modeling is executed based on user feedback. Through automated data processing, high-precision model prediction, real-time interactive optimization, and visualized modeling, the efficiency and quality of business chain design are improved simultaneously. Furthermore, model prediction based on historical data avoids human experience bias, meeting the comprehensive requirements of real-time performance, reliability, and resource efficiency in the development of complex equipment. This provides efficient technical support for the implementation of Model-Based Systems Engineering (MBSE) in the field of complex equipment. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the business data processing method based on visual swimlane modeling according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] This invention provides a business data processing method based on visual swimlane modeling, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the business data processing method based on visual swimlane modeling according to an embodiment of the present invention.

[0019] In this embodiment, a business data processing method based on visual swimlane modeling is provided, the method comprising the following steps: S1: Collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and construct a training sample set; S2: Based on the fusion architecture of graph neural network and Transformer, a multi-task prediction model is constructed. By combining the loss function and the constructed training sample set to train the model parameters, a multi-task prediction model with the ability to predict swimlane layout and event configuration is obtained. S3: Capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate standardized input vectors; S4: Input the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes, and select the optimal scheme set through simulation evaluation; S5: Display the optimal solution group in a visual overlay on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data.

[0020] It should be noted that existing business chain design technologies are mainly divided into two categories: one is based on static visualization tools (such as IBM Rhapsody), which express business logic through UML sequence diagrams and flowcharts. However, such tools can only describe simple call relationships, cannot distinguish event types (such as periodic events / conditional events), and the swimlane positions are fixed, making it difficult to dynamically reconstruct the process. Another type is the customized design method based on human experience, which relies on designers' understanding of the business to configure swimlane layout and event relationships. However, this method has three major drawbacks: First, design decisions are highly dependent on experience and lack historical data support, resulting in poor consistency of solutions produced by different designers. Second, as the complexity of the business chain increases (e.g., the number of swimlanes > 10, event relationships > 20), the efficiency of manual adjustment decreases exponentially and it is difficult to guarantee global optimization. Third, it lacks adaptive learning capabilities and cannot build predictive models based on historical design data and performance logs. Swimlane design directly affects the response time, resource utilization, and other performance indicators of the underlying system by determining data transmission paths, resource scheduling logic, and event interaction complexity. If problems can only be discovered through testing after the design is completed, the cost of later iterations accounts for a high proportion of the total design cost.

[0021] While some current improvement solutions attempt to introduce automated configuration functions, they still fail to address issues such as insufficient feature compactness, high real-time interaction latency, and poor adaptability to multiple scenarios. For example, existing solutions cannot capture user design operation features in real time and quickly generate optimization suggestions, and lack an online model update mechanism. This results in the need for manual rule readjustment in different scenarios such as avionics systems (high real-time requirements) and industrial monitoring (high reliability), leading to long adaptation cycles. Therefore, the industry urgently needs a business data processing solution based on visual swimlane modeling to improve the automation and accuracy of business chain design while ensuring low latency and low memory consumption.

[0022] To address the aforementioned issues, this embodiment employs a multi-task prediction model that integrates graph neural networks and Transformer, trained using multi-source historical business chain data. It generates multiple swimlane layout and event configuration schemes by combining real-time user operations captured through the design interface. After simulation evaluation and selection of the optimal scheme, the model is visualized. Finally, based on user feedback, visualized swimlane modeling is executed. Through automated data processing, high-precision model prediction, real-time interactive optimization, and visualized modeling, both the efficiency and quality of business chain design are improved. Furthermore, model prediction based on historical data avoids biases arising from human experience, meeting the comprehensive requirements of real-time performance, reliability, and resource efficiency in the development of complex equipment. This provides efficient technical support for the implementation of Model-Based Systems Engineering (MBSE) in the field of complex equipment.

[0023] In a preferred embodiment, step S1: Collect historical business chain design data from the multi-source system, perform standardization and time-series alignment, and construct a training sample set, specifically including: S11: Parse the version history of the business chain design file from the Git version control system, extract swimlane layout data L, and organize it into CSV format data according to version number and timestamp; wherein, the swimlane layout data includes swimlane ID, service name, service type, center coordinates, width and hierarchical relationship; S12: Parse the event definition section in the design file, construct the event graph, take service operations as nodes and event relationships as directed edges, generate event graph data G=(V,E), and transform the event graph data G into an adjacency matrix A and an edge feature matrix E_feat; S13: Collect performance indicator data P from the system test log database; wherein, the performance indicator data P includes event response time, packet loss rate, CPU utilization, and memory usage; S14: Based on the timestamp of the design version, the swimlane layout data L, event graph data G, and performance index data P are time-aligned to form a single sample. The single sample is then standardized using Z-score to obtain the training sample set.

[0024] In practical applications, the GitPython library is first used to parse the version history of the business chain design file in the Git version control system, extracting the swimlane layout data for each version, including swimlane ID, service name, service type, center coordinates, width, and hierarchical relationship. This data is then organized into CSV format according to the version number, timestamp, and swimlane attributes to ensure structured storage of swimlane information. Next, the event definition section in the design file is parsed, with service operations as nodes (containing attributes such as operation ID and swimlane ID) and event relationships as directed edges (containing attributes such as event type, period, or timeout parameters). An event graph is constructed and transformed into an adjacency matrix and an edge feature matrix to achieve a mathematical expression of event relationships. Following this, performance metrics data for the corresponding design version are collected from system test log databases (such as Elasticsearch), covering event response time, packet loss rate, CPU utilization, and memory usage. Extreme outliers are removed, and missing values ​​are filled with the average of metrics for similar events to ensure the integrity of the performance data. Finally, based on the submission timestamp of the design version, swimlane layout data, event graph data, and processed performance metrics data are time-series aligned to form a single sample containing all three. Z-score standardization is used to eliminate the influence of dimensions for continuous features in the sample, and the training set, validation set, and test set are divided in a 7:2:1 ratio to provide a standardized and highly correlated data source for subsequent model training.

[0025] In this embodiment, swimlane layout, event graph, and performance index data are collected from multiple sources including Git, design documents, and test logs. After time alignment and Z-score standardization, a structured training sample set is obtained. This successfully solves the problems of heterogeneous multi-source business chain data, low sample quality, and lack of historical data support in the design in the background technology. By unifying the text data from Git, the numerical data from the logs, and the structured data from the graph into CSV or matrix format, the format differences between data from different sources are eliminated. The time alignment process ensures the correlation between swimlane, event, and performance data, while the standardization process avoids the interference of units on model training, which greatly improves the signal-to-noise ratio of the training samples. This effectively provides a high-quality data foundation for the multi-task prediction model, avoids prediction bias caused by chaotic or incomplete input data, and lays a data foundation for the subsequent accurate prediction of swimlane layout and event configuration.

[0026] In a preferred embodiment, step S2 involves constructing a multi-task prediction model based on a graph neural network and Transformer fusion architecture, specifically including: S21: Construct a two-branch neural network architecture that includes a local prediction branch and an event configuration branch; The local prediction branch employs a graph attention network (GAT). The input layer receives the swimlane feature matrix L_feat extracted from the swimlane layout data L and the spatial adjacency matrix S. It learns the swimlane spatial dependencies through a two-layer graph attention GAT. The GAT has four attention heads and a hidden layer dimension of 64. The expression for the attention weights is as follows: ; In the formula, Let represent the initial eigenvector of lane u. Represents a linear transformation matrix. Represents the attention weight vector. Denotes the neighborhood set of lane u. This represents a vector concatenation operation. Let represent the initial eigenvector of lane v. Describe the neighborhood set of lane u Any lane in the middle, Indicates swimming lane The initial feature vector; The event configuration branch uses a Transformer decoder. The input layer receives the node feature matrix V_feat and edge feature matrix E_feat extracted from the event graph data G. The event association pattern is learned through a 3-layer encoder. The output layer outputs the event type probability distribution through the Softmax activation function.

[0027] In practical applications, this embodiment achieves synchronous prediction of swimlane layout and event configuration through a dual-branch design: the layout prediction branch employs a graph attention network (GAT). The input layer receives the swimlane feature matrix and spatial adjacency matrix extracted from the swimlane layout data. The swimlane feature matrix includes features such as the service type code, level, and normalized width of the swimlane. The spatial adjacency matrix is ​​calculated using the reciprocal of the Euclidean distance between swimlanes to represent the spatial similarity between them. Inside the branch, two layers of GAT (each with four attention heads and a hidden layer dimension of 64) learn the spatial dependencies between swimlanes and dynamically allocate the importance of different swimlanes using an attention weight formula. Finally, the predicted position coordinates of each swimlane are output; events... The configuration branch uses a Transformer decoder. The input layer receives the node feature matrix and edge feature matrix extracted from the event graph data. The node feature matrix contains attributes such as the ID encoding of the operation node and the swimlane ID to which it belongs. The edge feature matrix records the event type parameters. Inside the branch, a 3-layer Transformer encoder (8 attention heads per layer, 128 hidden layer dimensions, and 512 FeedForward network dimensions) learns the event association rules between operation nodes and uses a self-attention mechanism to capture the dependencies in the event sequence. The output layer uses a Softmax activation function to transform the network output into an event type probability distribution to achieve event type prediction.

[0028] Therefore, this embodiment constructs a layout prediction branch based on a Graph Attention Network (GAT) and an event configuration branch based on a Transformer decoder, forming a dual-branch neural network architecture. These branches learn swimlane spatial dependencies and event association patterns respectively, effectively solving the problems of insufficient model feature compactness, the inability of single-task models to simultaneously handle swimlane layout and event configuration prediction, and poor adaptability to multiple scenarios in the background technology. Specifically, GAT focuses on the spatial dependencies of key swimlanes through an attention mechanism, while the Transformer accurately captures the association patterns of event sequences, making the model's expression of swimlane spatial features and event logical features more compact. Simultaneously, the dual-branch design can optimize swimlane layout and event configuration tasks separately, avoiding the single model's neglect of one aspect, significantly improving event classification accuracy. Furthermore, the architecture supports adjusting the branch loss weights according to different business scenarios (e.g., increasing the event configuration branch weight in avionics scenarios), adapting to different scenario requirements without model reconstruction, and meeting the multi-scenario design needs of complex equipment development.

[0029] In a preferred embodiment, step S2 involves training model parameters by combining a loss function and a constructed training sample set to obtain a multi-task prediction model capable of swimlane layout prediction and event configuration. Specifically, this includes: S22: Define the combination loss function, expressed as: ; ; ;

[0030] ;

[0031] In the formula, This represents the total loss value. Indicates the loss weight. Indicates the predicted loss from the layout. Indicates event classification loss. Indicates learning loss, Indicates the number of lanes. Represents the actual coordinates of lane i. This represents the predicted coordinates of lane i. Indicates the number of events. The label representing the true type of event j. This represents the predicted probability that event j belongs to type k. Indicates batch size, This represents the characteristic representation of event b. This represents a positive sample of event b. Indicates a negative sample. Represents cosine similarity. Indicates temperature parameter; S23: The Adam optimizer is used to train the model under the preset conditions of learning rate, weight decay, batch size and training epochs. After training, a multi-task prediction model with swimlane layout prediction and event configuration capabilities is obtained.

[0032] In practical applications, this embodiment achieves efficient model training through loss function design and optimization strategies: First, a combined loss function is defined, where the total loss is obtained by weighted summation of layout prediction loss, event classification loss, and representation learning loss. The layout prediction loss uses mean squared error (MSE), which measures the deviation of swimlane layout prediction by calculating the mean of the squared differences between the true and predicted coordinates of the swimlane. For event classification loss, cross-entropy loss is used, which measures the accuracy of event type prediction by calculating the logarithmic loss between the true event type label (one-hot encoding) and the predicted probability. For representation learning loss, InfoNCE loss is used, which enhances the similarity of similar event features and the difference between dissimilar event features by comparing the cosine similarity between positive samples (same type events) and negative samples (different type events) of event features. The loss weights can be adjusted according to the business scenario (e.g., the event classification loss has a higher weight in avionics scenarios).

[0033] Following this, the Adam optimizer was used for model training, with an initial learning rate of 1e-4, weight decay of 1e-5, batch size of 32, and training epochs of 100. Gradient clipping (clip_norm=1.0) was used during training to prevent gradient explosion, and the learning rate was dynamically adjusted using a cosine annealing learning rate scheduling strategy to ensure stable model convergence. At the same time, an early stopping strategy was introduced, where the model performance (layout prediction MAE, event classification Acc) was evaluated using a validation set after each training epoch. If the validation loss did not decrease for 5 consecutive epochs, training was stopped to avoid model overfitting. Finally, the model parameters with the minimum validation loss were saved to ensure that the model has good generalization ability.

[0034] In this embodiment, a multi-task model capable of predicting swimlane layout and event configuration is obtained by defining a combined loss function including layout prediction loss, event classification loss, and representation learning loss, and training the model with the Adam optimizer under preset parameters. This effectively solves the problems of overfitting during model training, imbalance of multi-task losses, and low training efficiency in the background technology.

[0035] In a preferred embodiment, step S3: Capture real-time user actions through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user actions, and generate a standardized input vector, specifically including: S31: In the design interface, a centralized event listener is set up to capture three types of operation events; wherein, the operation events include mouse drag events, service placement events, or connection operation events; S32: Call the service registry through the backend API to extract the metadata of the currently operating service and convert it into a service metadata feature vector; wherein, the metadata includes service ID, service type, maximum processing latency, CPU resource requirements, memory requirements, and interface protocol; S33: Traverse the current canvas lanes, extract spatial features and density features, and concatenate the spatial features, density features and service element feature vectors to generate a canvas state feature vector as a standardized input vector.

[0036] In this embodiment, a JavaScript event listener is first integrated into the design interface developed in Vue to capture three core user operations in real time: mouse drag events (recording drag trajectory coordinates, sampling frequency 50Hz), service placement events (recording initial service placement coordinates and timestamp), and connection operation events (recording source operation and target operation IDs). All operation data is stored in the front-end localStorage in real time in JSON format (including operation type, service ID, trajectory information, and canvas state) to ensure that operation data is not lost and can be quickly retrieved.

[0037] Next, the MySQL service registry is called via the backend API to extract metadata of the current user's service, including service ID, service type, maximum processing latency, CPU resource requirements, memory requirements, and interface protocol. Categorical metadata (such as service type and interface protocol) is converted into numerical features using one-hot encoding, and continuous metadata (such as maximum processing latency and CPU resource requirements) is normalized to the [0,1] interval using min-max normalization, forming a 1×12 service metadata feature vector. Then, all swimlanes in the current design canvas are traversed to extract the spatial features of the swimlanes. The spatial features (current coordinates, width, and spacing between adjacent lanes) and density features (event connection density = number of event edges / number of lanes, layout uniformity = lane spacing variance) are concatenated with the service element feature vector according to their dimensions to form a canvas state feature vector (dimension is 1×(12+2N+2), where N is the current number of canvas lanes). Finally, the feature vector is checked for completeness and rationality. Missing values ​​are filled with the feature mean of the same type of lane, and features that exceed the reasonable range are truncated according to the boundary value. The checked feature vector is transmitted to the backend model service in real time via the WebSocket protocol to meet the real-time interaction requirements.

[0038] Therefore, this embodiment captures user operations through the event listener of the design interface, calls the service registry to extract and encode service metadata, and combines the spatial and density features of the canvas swimlanes to generate standardized input vectors, effectively solving the problems of high real-time interaction latency, incomplete feature extraction, and inconsistent feature formats in the background technology.

[0039] In a preferred embodiment, step S4: inputting the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes, and selecting the optimal scheme group through simulation evaluation, specifically including: S41: Input the standardized input vector into the layout prediction branch, and output the three lane position adjustment vectors with the highest confidence obtained by the layout prediction branch as layout schemes. Input the standardized input vector into the event configuration branch, and output the three configuration schemes with the highest probability obtained by the event configuration branch. S42: Cross-combination layout and configuration schemes form 9 initial candidate schemes, which are simulated in the SimPy simulation environment. Several core performance indicators of each scheme are output. The performance score is obtained by weighted summation of several core performance indicators. In the design constraint rule base, the design specification score is obtained by summing the mapping scores of several preset design specification standards and the configuration parameters of each scheme. S43: For the 9 initial candidate schemes, calculate the comprehensive performance score based on the performance score and the design specification score, and select the 5 schemes with the highest scores as the optimal scheme group; wherein, the expression for the comprehensive performance score is: ； In the formula, S represents the overall performance score. Indicates the performance score. This indicates the score for design specifications.

[0040] In this embodiment, the standardized input vectors obtained from real-time feature extraction are first input into the two branches of the multi-task prediction model. The layout prediction branch outputs the three swimlane position adjustment vectors with the highest confidence, forming three layout schemes (adjusting completely according to the predicted position, retaining the core swimlane position and only adjusting the non-core swimlane, and fine-tuning based on the predicted position). During the adjustment process, it is ensured that the swimlanes do not overlap. The event configuration branch outputs the three event configuration schemes with the highest event type probability distribution, including event type (such as periodic event, response event) and corresponding parameters (such as period value, timeout time).

[0041] Next, the three layout schemes were cross-combined with the three event schemes to form nine initial candidate schemes. Each scheme was simulated for one hour in the SimPy simulation environment, and four core performance indicators were output: event response time, CPU utilization, memory utilization, and logical clarity. After standardization, the indicators were weighted and summed according to preset weights (event response time 0.4, CPU utilization 0.25, memory utilization 0.2, and logical clarity 0.15) to obtain the performance score. At the same time, each scheme was checked against the design constraint rule base (including five specifications: service level compliance, swimlane layout uniformity, event type matching degree, ICD protocol version, and core swimlane position). Each specification has a maximum score of 20 points. The total score was calculated and normalized to obtain the design specification score. Finally, the comprehensive score of each scheme was calculated according to the comprehensive score formula, and the schemes were sorted from high to low. Redundant schemes with layout differences of less than 5 pixels and identical event configurations were eliminated, and the top five schemes were retained as the optimal scheme group to ensure that the schemes have both high performance and compliance.

[0042] Therefore, this embodiment generates multiple layout and event schemes by inputting standardized input vectors into the model, cross-combines them, evaluates performance scores through SimPy simulation, calculates specification scores based on the design constraint rule library, and selects the optimal scheme group according to the comprehensive score. This effectively solves the problems of low efficiency, difficulty in balancing scheme performance and compliance, and insufficient scheme diversity in the background technology. It automates the combination, evaluation, and selection of schemes, providing high-quality scheme selection for subsequent visualization modeling.

[0043] In a preferred embodiment, S5: Display the optimal solution group in a visual form on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data, specifically including: S51: Display each solution in the optimal solution group in order of its comprehensive performance score, and then overlay the current design solution on the canvas with a preset transparency outline. S52: Mark the differences between the current design and the optimal design with the target color, and generate a detailed display panel on one side of the canvas to show the performance score and design specification score of the current optimal design; S53: Capture and record user feedback, and select the optimal solution to perform visual swimlane modeling based on the operation instructions corresponding to the user feedback data; wherein, the operation instructions include solution application instructions, solution switching instructions, or solution modification instructions corresponding to keyboard shortcut feedback data or page menu operation feedback data.

[0044] In this embodiment, the selected optimal solution group is first sorted from high to low according to the comprehensive score. The swimlanes of each solution are then displayed on the design canvas with blue outlines of 60% transparency. The core swimlanes (such as "flight control calculation" and "sensor fusion") use a 2px wide outline, while the non-core swimlanes use a 1px wide outline, which makes it easy for users to quickly identify the priority of the solutions.

[0045] Next, the differences between the optimal solution and the current design are marked with red dashed boxes, including the area where the swimlane needs to be moved and the added or modified event edges. When the user hovers the mouse over the difference area, a tooltip pops up to show the difference (e.g., "Moving the swimlane 150 pixels to the right can reduce the event response time by 12ms"). At the same time, a details panel pops up on the right side of the canvas to show the solution's performance score, design specification score, core parameters (e.g., event cycle, ICD protocol version) and reasons for recommendation (e.g., "Based on 50 historical avionics project data, this layout reduces CPU utilization by 10%"). Users can click the "Split-screen Comparison" button to display the current design on the left and the optimal solution on the right, allowing them to intuitively view the differences between the two.

[0046] Finally, the system listens for user feedback, including keyboard shortcuts (Alt+Enter to apply all adjustments, Alt+L to apply only layout adjustments, and Alt+E to apply only event configuration adjustments) and right-click menu operations (Apply current scheme, Ignore current scheme, Customize adjustment parameters). After receiving user feedback instructions, the system automatically updates the swimlane positions and event configurations in the canvas. During the application process, a progress bar is displayed to indicate the processing progress (e.g., "In application: Swimlane 3 / 5 adjusted"). After the adjustment is completed, a prompt box "Scheme applied successfully, modeling completed" pops up. At the same time, the user feedback data (Accept = 1, Modify = 0.5, Reject = 0) is stored in the MongoDB database in the structure of "User ID-Scheme ID-Feedback Type-Operation Timestamp" to provide user preference data for subsequent online model learning.

[0047] Therefore, this embodiment effectively solves the problems of poor user interaction experience, modeling results that do not meet user needs, and unusable feedback data in the background technology by overlaying the optimal solution group with semi-transparent outlines according to the comprehensive score, marking the differences between the solutions and generating a details panel to display the scores, capturing user feedback (shortcut key / menu operation), and performing visual swimlane modeling based on the feedback instructions.

[0048] In the description of embodiments of the present invention, specific features, structures, materials or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0049] In the description of the embodiments of the present invention, it should be understood that "-" and "~" represent a range between two numerical values, and this range includes the endpoints. For example, "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0050] In the description of embodiments of the present invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A business data processing method based on visual swimlane modeling, characterized in that, The method includes the following steps: S1: Collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and construct a training sample set; S2: Based on the fusion architecture of graph neural network and Transformer, a multi-task prediction model is constructed. By combining the loss function and the constructed training sample set to train the model parameters, a multi-task prediction model with the ability to predict swimlane layout and event configuration is obtained. S3: Capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate standardized input vectors; S4: Input the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes, and select the optimal scheme set through simulation evaluation; S5: Display the optimal solution group in a visual overlay on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data.

2. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, Step S1: Collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and construct a training sample set, specifically including: S11: Parse the version history of the business chain design file from the Git version control system, extract swimlane layout data L, and organize it into CSV format data according to version number and timestamp; wherein, the swimlane layout data includes swimlane ID, service name, service type, center coordinates, width and hierarchical relationship; S12: Parse the event definition section in the design file, construct the event graph, take service operations as nodes and event relationships as directed edges, generate event graph data G=(V,E), and transform the event graph data G into an adjacency matrix A and an edge feature matrix E_feat; S13: Collect performance indicator data P from the system test log database; wherein, the performance indicator data P includes event response time, packet loss rate, CPU utilization, and memory usage; S14: Based on the timestamp of the design version, the swimlane layout data L, event graph data G, and performance index data P are time-aligned to form a single sample. The single sample is then standardized using Z-score to obtain the training sample set.

3. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, In step S2, a multi-task prediction model is constructed based on the fusion architecture of graph neural networks and Transformers, specifically including: S21: Construct a two-branch neural network architecture that includes a local prediction branch and an event configuration branch; The local prediction branch employs a graph attention network (GAT). The input layer receives the swimlane feature matrix L_feat extracted from the swimlane layout data L and the spatial adjacency matrix S. It learns the swimlane spatial dependencies through a two-layer graph attention GAT. The GAT has four attention heads and a hidden layer dimension of 64. The expression for the attention weights is as follows: ； In the formula, Let represent the initial eigenvector of lane u. Represents a linear transformation matrix. Represents the attention weight vector. Denotes the neighborhood set of lane u. This represents a vector concatenation operation. Let represent the initial eigenvector of lane v. Describe the neighborhood set of lane u Any lane in the middle, Indicates swimming lane The initial feature vector; The event configuration branch uses a Transformer decoder. The input layer receives the node feature matrix V_feat and edge feature matrix E_feat extracted from the event graph data G. The event association pattern is learned through a 3-layer encoder. The output layer outputs the event type probability distribution through the Softmax activation function.

4. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, In step S2, the model parameters are trained by combining the loss function and the constructed training sample set to obtain a multi-task prediction model with the ability to predict swimlane layout and event configuration. Specifically, this includes: S22: Define the combination loss function, expressed as: ； ； ； ； In the formula, This represents the total loss value. Indicates the loss weight. Indicates the predicted loss from the layout. Indicates event classification loss. Indicates learning loss, Indicates the number of lanes. Represents the actual coordinates of lane i. This represents the predicted coordinates of lane i. Indicates the number of events. The label representing the true type of event j. This represents the predicted probability that event j belongs to type k. Indicates batch size, This represents the characteristic representation of event b. This represents a positive sample of event b. Indicates a negative sample. Represents cosine similarity. Indicates temperature parameter; S23: The Adam optimizer is used to train the model under the preset conditions of learning rate, weight decay, batch size and training epochs. After training, a multi-task prediction model with swimlane layout prediction and event configuration capabilities is obtained.

5. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, Step S3: Capture real-time user actions through the event listener module of the design interface, extract service metadata and canvas state features from the real-time user actions, and generate a standardized input vector, specifically including: S31: In the design interface, a centralized event listener is set up to capture three types of operation events; wherein, the operation events include mouse drag events, service placement events, or connection operation events; S32: Call the service registry through the backend API to extract the metadata of the currently operating service and convert it into a service metadata feature vector; wherein, the metadata includes service ID, service type, maximum processing latency, CPU resource requirements, memory requirements, and interface protocol; S33: Traverse the current canvas lanes, extract spatial features and density features, and concatenate the spatial features, density features and service element feature vectors to generate a canvas state feature vector as a standardized input vector.

6. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, Step S4: Input the real-time standardized input vector into the trained multi-task prediction model to generate multiple sets of swimlane layout and event configuration optimization schemes. The optimal scheme set is then selected through simulation evaluation. Specifically, this includes: S41: Input the standardized input vector into the layout prediction branch, and output the three lane position adjustment vectors with the highest confidence obtained by the layout prediction branch as layout schemes. Input the standardized input vector into the event configuration branch, and output the three configuration schemes with the highest probability obtained by the event configuration branch. S42: Cross-combination layout and configuration schemes form 9 initial candidate schemes, which are simulated in the SimPy simulation environment. Several core performance indicators of each scheme are output. The performance score is obtained by weighted summation of several core performance indicators. In the design constraint rule base, the design specification score is obtained by summing the mapping scores of several preset design specification standards and the configuration parameters of each scheme. S43: For the 9 initial candidate schemes, calculate the comprehensive performance score based on the performance score and the design specification score, and select the 5 schemes with the highest scores as the optimal scheme group; wherein, the expression for the comprehensive performance score is: ； In the formula, S represents the overall performance score. Indicates the performance score. This indicates the score for design specifications.

7. The business data processing method based on visual swimlane modeling as described in claim 1, characterized in that, S5: Display the optimal solution group visually overlaid on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data, specifically including: S51: Based on the overall performance score, each solution in the optimal solution group is sequentially overlaid onto the current design solution on the canvas with a pre-set transparency outline. S52: Mark the differences between the current design and the optimal design with the target color, and generate a detailed display panel on one side of the canvas to show the performance score and design specification score of the current optimal design; S53: Capture and record user feedback, and select the optimal solution to perform visual swimlane modeling based on the operation instructions corresponding to the user feedback data; wherein, the operation instructions include solution application instructions, solution switching instructions, or solution modification instructions corresponding to keyboard shortcut feedback data or page menu operation feedback data.

8. A business data processing system based on visual swimlane modeling, characterized in that, include: The module is used to collect historical business chain design data from multi-source systems, perform standardization and time-series alignment, and build a training sample set. The training module is used to build a multi-task prediction model based on the fusion architecture of graph neural network and Transformer. By combining the loss function and the constructed training sample set, the model parameters are trained to obtain a multi-task prediction model with the ability to predict swimlane layout and event configuration. The generation module is used to capture real-time user operations through the event listening module of the design interface, extract service metadata and canvas state features from the real-time user operations, and generate standardized input vectors. The filtering module is used to input real-time standardized input vectors into the trained multi-task prediction model, generate multiple sets of swimlane layout and event configuration optimization schemes, and filter the optimal scheme group through simulation evaluation. The execution module is used to display the optimal solution group in a visual overlay on the design canvas, record user feedback data, and perform visual swimlane modeling based on the user feedback data.

Images