Intelligent error handling method and system for cross-platform RPA robots

Abstract

This application provides an intelligent error handling method and system for cross-platform RPA robots. The method includes: creating error detection factories for runtime data of various operating systems, capturing error data streams through the error detection factories; classifying and rating the error data streams to obtain error classification and rating results; jointly optimizing the error classification and rating results based on a preset rule engine and a pre-built reinforcement learning model to obtain a dynamic error recovery strategy; decomposing the dynamic error recovery strategy into multiple atomic operations, mapping the atomic operations to target recovery operations according to the target operating system; and coordinating the target operating system to execute the target recovery operations through a distributed transaction manager.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an intelligent error handling method and system for a cross-platform RPA robot. Background Art

[0002] Currently, in modern enterprises, cross-platform RPA (Robotic Process Automation) robots play an important role in improving work efficiency and reducing human errors. However, in cross-platform environments, such as Windows operating system, macOS operating system, and Linux operating system, the intelligent error handling of RPA robots faces many challenges and technical conflicts. Specifically, when desktop automation robots run between different operating systems, the error handling mechanism may be inconsistent due to differences in operating systems, making it difficult to implement a unified error management strategy. Summary of the invention

[0003] The present application provides a cross-platform RPA robot intelligent error handling method and system for improving the accuracy of error recognition and error handling of intelligent robots on multiple platforms.

[0004] In a first aspect, an embodiment of the present application provides an intelligent error handling method for a cross-platform RPA robot, the method comprising:

[0005] Creating error detection factories for operation data of various operating systems respectively, and capturing error data streams through the error detection factories;

[0006] Classify and rate the erroneous data stream to obtain an error classification and rating result;

[0007] According to a preset rule engine and a preset reinforcement learning model, the error classification and rating results are jointly optimized to obtain a dynamic error recovery strategy;

[0008] Decomposing the dynamic error recovery strategy into a plurality of atomic operations, and mapping the atomic operations into target recovery operations according to a target operating system;

[0009] The target operating system is coordinated by a distributed transaction manager to perform the target recovery operation.

[0010] In a second aspect, an embodiment of the present application provides an intelligent error handling system for a cross-platform RPA robot, the system comprising:

[0011] An error capture module, used to create error detection factories for the operation data of various operating systems respectively, and to capture error data streams through the error detection factories;

[0012] A classification and rating module, used for classifying and rating the erroneous data stream to obtain an error classification and rating result;

[0013] An error recovery module, used to jointly optimize the error classification and rating results according to a preset rule engine and a preset reinforcement learning model to obtain a dynamic error recovery strategy;

[0014] An operation mapping module, used for decomposing the dynamic error recovery strategy into multiple atomic operations, and mapping the atomic operations into target recovery operations according to a target operating system;

[0015] The execution coordination module is used to coordinate the target operating system to execute the target recovery operation through a distributed transaction manager.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor;

[0017] The memory is used to store computer programs;

[0018] The processor is used to execute the computer program and implement the intelligent error handling method of the cross-platform RPA robot as described in any one of the embodiments of the present application when executing the computer program.

[0019] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor enables the processor to implement an intelligent error handling method for a cross-platform RPA robot as described in any one of the embodiments of the present application.

[0020] An embodiment of the present application provides an intelligent error handling method for a cross-platform RPA robot, characterized in that the method includes: creating error detection factories for the operating data of multiple operating systems respectively, and capturing error data streams through the error detection factories; classifying and rating the error data streams to obtain error classification and rating results; jointly optimizing the error classification and rating results according to a preset rule engine and a preset reinforcement learning model to obtain a dynamic error recovery strategy; decomposing the dynamic error recovery strategy into multiple atomic operations, and mapping the atomic operations to target recovery operations according to the target operating system; and coordinating the target operating system to perform the target recovery operation through a distributed transaction manager. In the above method, the classification and rating process of the error data stream helps to quickly identify and prioritize key errors. The joint optimization mechanism of the preset rule engine and the reinforcement learning model can dynamically generate highly adaptable error recovery strategies, thereby improving the system's adaptability. The recovery strategy is decomposed into atomic operations and mapped to the target operating system, ensuring cross-platform compatibility and execution accuracy. The consistency problem of multi-system collaborative execution is solved through the distributed transaction manager, ensuring the reliability of complex recovery operations. This multi-level, adaptive error handling mechanism significantly improves the stability and reliability of RPA robots in complex heterogeneous environments, reduces the need for manual intervention, and improves the accuracy and efficiency of automated processes. At the same time, the cross-platform characteristics of this method expand the application scope of RPA robots, enabling them to switch seamlessly between different operating systems, thereby improving the accuracy of error recognition and error handling of multi-platform intelligent robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0022] Figure 1 A schematic flowchart of an intelligent error handling method for a cross-platform RPA robot provided in an embodiment of the present application;

[0023] Figure 2 A schematic block diagram of an intelligent error handling system for a cross-platform RPA robot provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] RPA, or Robotic Process Automation, is often referred to as an "intelligent assistant". RPA robots are software technologies that automatically perform repetitive tasks that follow fixed rules, rather than physical robots in the traditional sense. The core function of RPA robots is to simulate human business process operations on different digital platforms; they interact with existing information systems according to established procedures to complete their set tasks.

[0029] See also Figure 1 , Figure 1 The embodiment of the present application also provides a schematic flow chart of an intelligent error handling method for a cross-platform RPA robot. Figure 1 As shown, the specific steps of the intelligent error handling method of the cross-platform RPA robot include: S101-S105.

[0030] S101 . Create error detection factories for operation data of various operating systems respectively, and capture error data flows through the error detection factories.

[0031] For example, a cross-platform compatible error detection and capture module is designed. This module adopts the abstract factory design pattern to create a specific error detection factory class for each supported operating system (Windows, macOS, and Linux). Each factory class is responsible for implementing the underlying error detection logic of a specific operating system, including system call monitoring, window event capture, and process status tracking. To ensure cross-platform compatibility, a unified error interface is defined, which contains standardized fields such as error type, error code, and error description. At runtime, the RPA robot dynamically selects the corresponding error detection factory according to the current operating system and instantiates error detection objects. These objects use operating system-specific APIs and system calls to monitor abnormal conditions during the execution of RPA tasks. For example, on Windows, Windows API functions such as GetLastError() and FormatMessage() are used to obtain detailed error information; on macOS, the error handling mechanism provided by the NSError object and the Cocoa framework is used; on Linux, / var / log / syslog and application-specific log files are parsed. In this way, a flexible and extensible error detection system is implemented that can adapt to the characteristics of different operating systems while providing a unified error information format, laying the foundation for subsequent error handling and analysis.

[0032] S102: Classify and rate the erroneous data stream to obtain an error classification and rating result.

[0033] For example, an error feature vector is established, which includes dimensions such as error type, frequency of occurrence, scope of impact, and mission criticality. Then, the model is trained using historical error data, which comes from the error logs in a unified format in the previous step. During the model training process, cross-validation techniques are used to optimize model parameters and improve classification accuracy. In order to handle the unique error types that may occur in different operating systems, operating system identifiers are introduced as additional features during the feature engineering phase, so that the model can learn operating system-specific error patterns. In addition, an online learning mechanism is implemented to allow the model to be continuously updated and optimized based on newly collected error data. This intelligent classification system can quickly identify the nature and severity of errors, such as classifying errors as "system level", "application level" or "data level", and assigning priorities such as "urgent", "high", "medium" or "low". This classification and priority assessment provides an important basis for the subsequent selection of error handling strategies, ensuring that RPA robots can reasonably allocate resources and prioritize the most critical errors.

[0034] S103: According to a preset rule engine and a preset reinforcement learning model, the error classification and rating results are jointly optimized to obtain a dynamic error recovery strategy.

[0035] For example, a dynamic error recovery strategy generator is designed, which uses a combination of a preset rule engine and reinforcement learning to develop optimal recovery strategies for errors of different types and priorities. The rule engine contains a series of predefined error handling rules, which are written based on domain expert knowledge and best practices and cover common error scenarios. For example, for the "file does not exist" error, the rules may include operations such as retrying access, checking the file path, and creating a new file. The reinforcement learning model learns the most effective error recovery strategy by interacting with the environment. Ultimately, the dynamic error recovery strategy generator can generate targeted, multi-step error recovery solutions in real time, such as "restarting the service-clearing the cache-re-executing the task", which greatly improves the error handling capability and task completion rate of the RPA robot.

[0036] S104: Decompose the dynamic error recovery strategy into multiple atomic operations, and map the atomic operations into target recovery operations according to a target operating system.

[0037] Exemplarily, the dynamic error recovery strategy is semantically analyzed to construct an abstract syntax tree of the strategy. This step helps to understand the structure and logical relationship of the strategy. Next, based on the abstract syntax tree, the strategy is decomposed into a series of basic operation units, namely atomic operations. It should be noted that atomic operation refers to an operation that will not be interrupted by the thread scheduling mechanism in multi-threaded programming; that is, once the operation is started, it will run until the end without any switching or interruption in the middle. In other words, in a multi-processor or multi-core environment, an atomic operation is either fully executed or not executed at all, and there is no situation where part of it is executed. These atomic operations are cross-platform and universal and do not depend on a specific operating system. The dependencies between these atomic operations are analyzed to construct an operation dependency graph. The dependency graph can help determine the execution order and parallel possibility of the operation. Based on the dependency graph, task scheduling optimization can be performed to improve execution efficiency. A mapping library is established for the target operating system. This library contains the mapping relationship between atomic operations and specific operating system API calls. For each atomic operation, the corresponding target system call or command is found by querying the mapping library. In the mapping process, the characteristics and limitations of the target system also need to be considered. Some atomic operations may need to be combined or split to suit the characteristics of the target system. In addition, differences between different operating systems must be handled, such as file path formats, system command syntax, etc. The mapped operation sequence is organized into target recovery operations. This may involve generating scripts, configuration files, or direct API call sequences, depending on the interface method of the target operating system.

[0038] Through this extended process, it can be ensured that the dynamic error recovery strategy can be effectively converted into specific operations that can be executed on a specific target operating system, thereby achieving cross-platform error recovery capabilities.

[0039] S105: Coordinate the target operating system to execute the target recovery operation through a distributed transaction manager.

[0040] For example, the distributed transaction manager needs to perform task decomposition on the target recovery operation. It splits the overall recovery operation into multiple subtasks, each of which corresponds to one or more specific system calls or operations. This decomposition helps to achieve fine-grained control and management. Based on the decomposed subtasks, a directed acyclic graph (DAG) is constructed to represent the dependencies and execution order between tasks. DAG can help optimize task scheduling and identify tasks that can be executed in parallel, thereby improving overall execution efficiency. The distributed transaction manager needs to implement global transaction coordination. It uses protocols such as two-phase commit (2PC) or three-phase commit (3PC) to ensure that all participating system components maintain consistency when performing recovery operations. This includes the preparation phase, the commit phase, and possible rollback mechanisms. During the execution process, the distributed transaction manager also needs to handle potential failures and abnormal situations. This includes network partitions, node failures, timeouts, and other issues. The manager needs to implement fault detection and recovery mechanisms, such as heartbeat detection, timeout retry, and state recovery.

[0041] To improve performance and reliability, a distributed transaction manager may adopt an asynchronous execution mode. It can use a message queue or event bus to manage the asynchronous execution of tasks while ensuring execution order and data consistency. The distributed transaction manager needs to provide monitoring and logging capabilities. It tracks the execution status of each subtask in real time, and records key events and error information for subsequent analysis and auditing. At the same time, it should also provide a rollback mechanism to restore the system to a consistent state in the event of an unrecoverable error.

[0042] Through this complex coordination mechanism, the distributed transaction manager ensures that target recovery operations across multiple operating systems and components can be performed in a consistent and reliable manner, maintaining the overall consistency and reliability of the system even in the face of various potential failures and abnormal situations.

[0043] An embodiment of the present application provides an intelligent error handling method for a cross-platform RPA robot, characterized in that the method includes: creating error detection factories for the operating data of multiple operating systems respectively, and capturing error data streams through the error detection factories; classifying and rating the error data streams to obtain error classification and rating results; jointly optimizing the error classification and rating results according to a preset rule engine and a preset reinforcement learning model to obtain a dynamic error recovery strategy; decomposing the dynamic error recovery strategy into multiple atomic operations, and mapping the atomic operations to target recovery operations according to the target operating system; and coordinating the target operating system to perform the target recovery operation through a distributed transaction manager. In the above method, the classification and rating process of the error data stream helps to quickly identify and prioritize key errors. The joint optimization mechanism of the preset rule engine and the reinforcement learning model can dynamically generate highly adaptable error recovery strategies, thereby improving the system's adaptability. The recovery strategy is decomposed into atomic operations and mapped to the target operating system, ensuring cross-platform compatibility and execution accuracy. The consistency problem of multi-system collaborative execution is solved through the distributed transaction manager, ensuring the reliability of complex recovery operations. This multi-level, adaptive error handling mechanism significantly improves the stability and reliability of RPA robots in complex heterogeneous environments, reduces the need for manual intervention, and improves the accuracy and efficiency of automated processes. At the same time, the cross-platform characteristics of this method expand the application scope of RPA robots, enabling them to switch seamlessly between different operating systems, thereby improving the accuracy of error recognition and error handling of multi-platform intelligent robots.

[0044] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0045] In some embodiments, error detection factories are created for the operation data of various operating systems respectively, and the specific steps include: S1011-S1017.

[0046] S1011. Input the operating data into the first convolution sublayer, the second convolution sublayer, and the third convolution sublayer respectively to obtain the first time series feature, the second time series feature, and the third time series feature.

[0047] S1012: Fusing the first time series feature, the second time series feature, and the third time series feature to obtain a fused time series feature vector.

[0048] Exemplarily, a short-time convolution sublayer is applied to the error data stream, and 3 layers of convolution operations are performed to obtain short-time features. A medium-time convolution sublayer is applied to the error data stream, and 5 layers of convolution operations are performed to obtain medium-time features. A long-time convolution sublayer is applied to the error data stream, and 7 layers of convolution operations are performed to obtain long-time features. The short-time features, medium-time features, and long-time features are fused to obtain a multi-scale time feature vector.

[0049] S1013. Perform a linear transformation on the fused time series feature vector to obtain a query matrix, a key matrix and a value matrix, calculate the attention weights based on the query matrix, the key matrix and the value matrix to obtain weighted features, perform multi-head splicing and linear transformation on the weighted features to obtain attention weighted features.

[0050] Exemplarily, the multi-scale time series feature vectors are used as the input of the query matrix, the key matrix and the value matrix respectively. The query matrix, the key matrix and the value matrix are linearly transformed to obtain multiple attention heads. The attention weight is calculated for each attention head, and the weighted sum is performed to obtain the attention weighted feature.

[0051] S1014. Input the attention weighted features into the fourth convolutional layer, perform adaptive convolution, and obtain an adaptive feature representation.

[0052] Exemplarily, the convolution kernel size and step size are dynamically calculated according to the size of the attention-weighted feature. The calculated convolution kernel size and step size are used to perform a convolution operation on the attention-weighted feature. An activation function is applied to the convolution result to obtain an adaptive feature representation.

[0053] S1015. Input the adaptive feature representation into the pooling layer to obtain a reduced-dimensional feature vector.

[0054] Exemplarily, the pooling window size and step size are calculated according to the size of the adaptive feature representation and the target output size. The adaptive feature representation is pooled using the calculated pooling window size and step size to obtain a feature vector after dimensionality reduction.

[0055] S1016. Construct a multi-task learning model based on the dimension-reduced feature vector to obtain a multi-task feature representation.

[0056] Exemplarily, the reduced-dimensional feature vector is input to multiple parallel task-specific layers. The task-related loss function is calculated for the output of each task-specific layer. The loss functions of multiple tasks are weighted summed to obtain a multi-task loss. The model parameters are optimized based on the multi-task loss to obtain a multi-task feature representation.

[0057] S1017. Initialize meta parameters according to the multi-task feature representation to obtain an initial meta model, perform gradient update on the initial meta model to obtain a task-adaptive meta model, use the task-adaptive meta model to predict new samples, and obtain an error detection factory model.

[0058] Exemplarily, the multi-task feature representation is used as the input of the meta-learning model. The conditional probability distribution of the meta-learning model is calculated. The log-likelihood function is maximized to optimize the meta-parameters. The optimized meta-parameters are combined with the multi-task learning model to obtain the error detection factory model.

[0059] In some embodiments, the first time series feature, the second time series feature, and the third time series feature are fused to obtain a fused time series feature vector, including:

[0060] The first time series feature, the second time series feature, and the third time series feature are self-attention calculated respectively to obtain three attention weighted features. Specifically, the query matrix Q, the key matrix K, and the value matrix V are constructed according to the first time series feature, the second time series feature, and the third time series feature. Calculate the query matrix Q, the key matrix K, and the value matrix V, where Q=XW q , K = XW k , V = XW v , X is the input feature, W q , W k , W v is a learnable parameter matrix, and calculates the attention weight matrix A=softmax( ), where d is the weighted sum Z = AV calculated for the feature dimension, and Z is multi-headedly concatenated to obtain the attention weighted feature.

[0061] A multilayer perceptron is constructed based on the three attention-weighted features, and the output of the multilayer perceptron is transformed nonlinearly to obtain a nonlinear feature representation. Specifically, the three attention-weighted features are concatenated into the input matrix X, and two hidden layers h1=ReLU(XW1+b1) and h2=ReLU(h1W2+b2) are constructed. The output layer y=h2W3+b3 is batch normalized to obtain a nonlinear feature representation.

[0062] Dynamic convolution kernels are applied to nonlinear feature representations to perform adaptive convolution operations and obtain adaptive feature maps. Specifically, the convolution kernel parameters θ=f(x) are dynamically generated according to the input features, where f is a parameter generation network. Dynamic convolution kernels of three sizes, 3×3, 5×5, and 7×7, are constructed. Convolution operations y=x*θ are performed on the input features respectively, and the convolution results of different scales are weighted and fused to obtain adaptive feature maps.

[0063] A graph neural network is constructed based on the adaptive feature graph, and the node representation of the graph neural network is aggregated to obtain graph-level features. Specifically, the feature graph is converted into graph structure data, each pixel point is used as a node, and the edge connection relationship E(i, j) = sim(h i ,h j ), where sim is the similarity function, and the graph attention layer is used to update the node features h'i=σ(Σj∈N i α ij W j ), and obtain graph-level feature representation through global average pooling.

[0064] A residual connection network is constructed based on the graph-level features, and the intermediate layer features of the residual connection network are transferred across layers to obtain multi-scale fusion features. Specifically, a network containing three residual blocks is constructed, each residual block contains two layers of convolution, and a jump connection y=F(x)+x is added to each residual block. Features at different layers are dimensionally aligned through 1×1 convolution, and features of multiple layers are weighted and combined to obtain multi-scale fusion features.

[0065] Adaptive pooling is performed on the multi-scale fusion features, and weighted average is performed according to the feature importance to obtain the fused time series feature vector. Specifically, the feature importance score s=σ(Wx+b) is calculated, and the feature map is globally averaged pooled in the spatial dimension. The features of different channels are weighted according to the importance score, and the weighted features are concatenated to obtain the final fused time series feature vector.

[0066] Through the above method, local details and global dependencies in time series data can be captured, feature changes at multiple scales can be adapted, and the expressive power of feature representation can be improved.

[0067] In some embodiments, the error data stream is classified and rated to obtain an error classification rating result, including: performing multimodal feature extraction on the error data stream to obtain an error feature vector; constructing a heterogeneous graph neural network based on the error feature vector, performing message passing and aggregation on the nodes, and obtaining a graph embedding representation; applying an adaptive attention mechanism to the graph embedding representation to obtain context-aware features; constructing a hierarchical Bayesian network based on the context-aware features, inferring the conditional probability distribution, and obtaining an error category probability distribution; applying fuzzy set theory to the error category probability distribution, performing membership calculation, and obtaining an error severity score; and performing multi-level classification of errors based on the error severity score and a preset rating threshold to obtain an error classification rating result.

[0068] For example, multimodal feature extraction can capture multi-dimensional information of error data. The construction of heterogeneous graph neural networks and the message passing mechanism help explore the potential correlation between errors. The application of adaptive attention mechanism further enhances the perception of contextual information. The introduction of hierarchical Bayesian networks enables the system to perform error reasoning based on probabilistic models, improving the interpretability of classification. The application of fuzzy set theory provides a flexible mathematical tool for quantitative assessment of error severity. Multi-level classification based on preset thresholds ensures the accuracy and operability of rating results. The combination of this series of steps not only improves the accuracy of error classification, but also enhances the system's adaptability to complex error scenarios, laying a solid foundation for the subsequent formulation of error recovery strategies.

[0069] In some embodiments, a dynamic error recovery strategy is decomposed into multiple atomic operations, and the atomic operations are mapped to target recovery operations according to a target operating system, including: performing semantic analysis on the dynamic error recovery strategy to obtain a semantic tree structure, decomposing the dynamic error recovery strategy according to the semantic tree structure to obtain multiple atomic operation sequences; determining an atomic operation dependency graph according to the multiple atomic operation sequences, and parallelizing the multiple atomic operation sequences according to the atomic operation dependency graph to obtain a parallelized atomic operation set; performing platform adaptability analysis on each atomic operation in the parallelized atomic operation set according to a preset cross-platform operation mapping library to obtain a platform adaptability score; screening the parallelized atomic operation set according to the platform adaptability score to obtain an atomic operation subset compatible with the target platform; performing similarity matching on the atomic operation subset compatible with the target platform and the system call library of the target operating system to determine a candidate system call set; performing call chain optimization according to a preset call chain optimization algorithm and the candidate system call set to generate a target recovery operation code; performing static code analysis on the target recovery operation code to obtain a code dependency graph; modularly encapsulating the target recovery operation code according to the code dependency graph to obtain a target recovery operation.

[0070] In some embodiments, call chain optimization is performed according to a preset call chain optimization algorithm and a candidate system call set to generate a target recovery operation code, including: performing time and space complexity analysis on the candidate system call set to obtain a call complexity matrix; performing multi-objective optimization on the candidate system call set according to the call complexity matrix to obtain a Pareto optimal solution set; applying a genetic algorithm to the Pareto optimal solution set to obtain a global optimal call chain; constructing an abstract syntax tree according to the global optimal call chain, and performing code generation conversion on the abstract syntax tree to obtain a target recovery operation code.

[0071] Exemplarily, the first step is to perform time and space complexity analysis on the candidate system call set to obtain a call complexity matrix. This step provides quantitative basic data for the subsequent optimization process, so that different system calls can be objectively compared and evaluated.

[0072] Step 2: Use the complexity matrix obtained in the first step to perform multi-objective optimization and obtain the Pareto optimal solution set. This step seeks a balance between time and space complexity, provides a series of possible optimal solutions, and provides multiple options for subsequent further optimization.

[0073] Step 3: Use genetic algorithm to process the Pareto optimal solution set and obtain the global optimal call chain. This step further screens and optimizes the Pareto optimal solution set based on the second step, and selects a global optimal call chain from multiple possible optimal solutions, providing clear guidance for the final code generation.

[0074] Step 4: Build an abstract syntax tree based on the global optimal call chain and perform code generation conversion to obtain the target recovery operation code. This step converts the optimization results of the previous steps into actual executable code, completing the conversion from theoretical optimization to practical application.

[0075] In some embodiments, a target operating system is coordinated to execute a target recovery operation through a distributed transaction manager, including: decomposing the target recovery operation into tasks to obtain multiple subtasks; constructing a directed acyclic graph based on the multiple subtasks, and topologically sorting the directed acyclic graph to obtain a task execution sequence; applying a distributed consistency protocol to the task execution sequence to obtain a global consistency state; and scheduling resources for the target operating system based on the global consistency state to obtain a final execution result.

[0076] This method improves the manageability of complex recovery operations through task decomposition, ensures the correct execution order of subtasks by using directed acyclic graphs and topological sorting, ensures data consistency across system operations by applying distributed consistency protocols, optimizes system resource utilization and improves the execution efficiency of recovery operations by scheduling resources based on the global consistency state. Overall, it realizes the intelligence and automation of the error handling process of cross-platform RPA robots and enhances the fault tolerance and reliability of the system.

[0077] See also Figure 2 , Figure 2 The embodiment of the present application also provides a schematic block diagram of an intelligent error handling system for a cross-platform RPA robot, wherein the intelligent error handling system 200 for a cross-platform RPA robot is used to execute the aforementioned intelligent error handling method for a cross-platform RPA robot. The intelligent error handling system 200 for a cross-platform RPA robot can be configured in a server.

[0078] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0079] like Figure 2 As shown, the intelligent error handling system 200 of the cross-platform RPA robot includes: an error capture module 201, a classification and rating module 202, an error recovery module 203, an operation mapping module 204 and an execution coordination module 205.

[0080] The error capture module 201 is used to create error detection factories for the operation data of various operating systems respectively, and capture error data streams through the error detection factories;

[0081] The classification and rating module 202 is used to classify and rate the error data stream to obtain an error classification and rating result;

[0082] The error recovery module 203 is used to jointly optimize the error classification and rating results according to the preset rule engine and the preset reinforcement learning model to obtain a dynamic error recovery strategy;

[0083] An operation mapping module 204, used to decompose the dynamic error recovery strategy into multiple atomic operations, and map the atomic operations into target recovery operations according to the target operating system;

[0084] The execution coordination module 205 is used to coordinate the target operating system to execute the target recovery operation through the distributed transaction manager.

[0085] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement an intelligent error handling method for a cross-platform RPA robot such as any one of the embodiments of the present application when executing the computer program.

[0086] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements an intelligent error handling method for a cross-platform RPA robot such as any one of the embodiments of the present application.

[0087] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An intelligent error handling method for a cross-platform RPA robot, characterized in that: The method comprises: Creating error detection factories for operation data of various operating systems respectively, and capturing error data streams through the error detection factories; Classify and rate the erroneous data stream to obtain an error classification and rating result; According to a preset rule engine and a preset reinforcement learning model, the error classification and rating results are jointly optimized to obtain a dynamic error recovery strategy; The dynamic error recovery strategy is semantically analyzed to obtain a semantic tree structure, and the dynamic error recovery strategy is decomposed according to the semantic tree structure to obtain a plurality of atomic operation sequences; an atomic operation dependency graph is determined according to the plurality of atomic operation sequences, and the plurality of atomic operation sequences are parallelized according to the atomic operation dependency graph to obtain a parallelized atomic operation set; platform adaptability analysis is performed on each atomic operation in the parallelized atomic operation set according to a preset cross-platform operation mapping library to obtain a platform adaptability score; the parallelized atomic operation set is screened according to the platform adaptability score to obtain a target platform compatible atomic operation subset; and the target platform compatible atomic operation subset is compared with the target platform compatible atomic operation subset. The system call library of the target operating system is matched for similarity to determine a candidate system call set; the time and space complexity analysis of the candidate system call set is performed to obtain a call complexity matrix; the multi-objective optimization of the candidate system call set is performed according to the call complexity matrix to obtain a Pareto optimal solution set; the genetic algorithm is applied to the Pareto optimal solution set to obtain a global optimal call chain; an abstract syntax tree is constructed according to the global optimal call chain, and the abstract syntax tree is converted to code generation to obtain a target recovery operation code; static code analysis is performed on the target recovery operation code to obtain a code dependency graph; modular packaging of the target recovery operation code is performed according to the code dependency graph to obtain a target recovery operation; The target recovery operation is task-decomposed to obtain a plurality of subtasks; a directed acyclic graph is constructed based on the plurality of subtasks, and the directed acyclic graph is topologically sorted to obtain a task execution sequence; a distributed consistency protocol is applied to the task execution sequence to obtain a global consistency state; resources of the target operating system are scheduled based on the global consistency state to obtain a final execution result.

2. The cross-platform RPA robot intelligent error handling method according to claim 1, characterized in that: The error detection factories are created for the operation data of various operating systems respectively, including: Inputting the operating data into a first convolution sublayer, a second convolution sublayer, and a third convolution sublayer respectively to obtain a first time series feature, a second time series feature, and a third time series feature; Performing feature fusion on the first time series feature, the second time series feature and the third time series feature to obtain a fused time series feature vector; Performing a linear transformation on the fused time series feature vector to obtain a query matrix, a key matrix and a value matrix, calculating attention weights according to the query matrix, the key matrix and the value matrix to obtain weighted features, and performing multi-head concatenation and linear transformation on the weighted features to obtain attention weighted features; Inputting the attention weighted features into the fourth convolutional layer, performing adaptive convolution, and obtaining an adaptive feature representation; Inputting the adaptive feature representation into a pooling layer to obtain a reduced-dimensional feature vector; Constructing a multi-task learning model based on the dimension-reduced feature vector to obtain a multi-task feature representation; Initialize meta-parameters according to the multi-task feature representation to obtain an initial meta-model, perform gradient update on the initial meta-model to obtain a task-adaptive meta-model, use the task-adaptive meta-model to predict new samples, and obtain an error detection factory model.

3. The intelligent error handling method for a cross-platform RPA robot according to claim 2, characterized in that: The step of fusing the first time series feature, the second time series feature, and the third time series feature to obtain a fused time series feature vector includes: Performing self-attention calculation on the first time series feature, the second time series feature, and the third time series feature respectively to obtain three attention weighted features; Constructing a multilayer perceptron according to the three attention weighted features, and performing a nonlinear transformation on the output of the multilayer perceptron to obtain a nonlinear feature representation; Applying a dynamic convolution kernel to the nonlinear feature representation, performing an adaptive convolution operation, and obtaining an adaptive feature map; Constructing a graph neural network according to the adaptive feature graph, aggregating node representations of the graph neural network to obtain graph-level features; Constructing a residual connection network according to the graph-level features, performing cross-layer feature transfer on the intermediate layer features of the residual connection network, and obtaining multi-scale fusion features; The multi-scale fusion features are weighted averaged to obtain a fused time series feature vector.

4. The cross-platform RPA robot intelligent error handling method according to claim 1, characterized in that: Classifying and rating the erroneous data stream to obtain an error classification and rating result includes: Performing multimodal feature extraction on the error data stream to obtain an error feature vector; Constructing a heterogeneous graph neural network according to the error feature vector, performing message passing and aggregation on nodes, and obtaining a graph embedding representation; Applying an adaptive attention mechanism to the graph embedding representation to obtain context-aware features; Constructing a hierarchical Bayesian network based on the context-aware features, reasoning about the conditional probability distribution, and obtaining an error category probability distribution; Applying fuzzy set theory to the probability distribution of the error categories, calculating the membership degree, and obtaining the error severity score; According to the error severity score and a preset rating threshold, the errors are classified into multiple levels to obtain an error classification rating result.

5. An intelligent error handling system for a cross-platform RPA robot, characterized in that: The cross-platform RPA robot intelligent error handling system is used to execute the cross-platform RPA robot intelligent error handling method according to any one of claims 1 to 4, and the cross-platform RPA robot intelligent error handling system includes: An error capture module, used to create error detection factories for the operation data of various operating systems respectively, and to capture error data streams through the error detection factories; A classification and rating module, used for classifying and rating the erroneous data stream to obtain an error classification and rating result; An error recovery module, used to jointly optimize the error classification and rating results according to a preset rule engine and a preset reinforcement learning model to obtain a dynamic error recovery strategy; An operation mapping module is used to perform semantic analysis on the dynamic error recovery strategy to obtain a semantic tree structure, decompose the dynamic error recovery strategy according to the semantic tree structure to obtain multiple atomic operation sequences; determine an atomic operation dependency graph according to the multiple atomic operation sequences, parallelize the multiple atomic operation sequences according to the atomic operation dependency graph to obtain a parallelized atomic operation set; perform platform adaptability analysis on each atomic operation in the parallelized atomic operation set according to a preset cross-platform operation mapping library to obtain a platform adaptability score; screen the parallelized atomic operation set according to the platform adaptability score to obtain a target platform compatible atomic operation subset; and perform platform adaptability analysis on the atomic operations compatible with the target platform. The subset is matched with the system call library of the target operating system for similarity to determine a candidate system call set; the candidate system call set is analyzed for time and space complexity to obtain a call complexity matrix; the candidate system call set is optimized for multiple objectives according to the call complexity matrix to obtain a Pareto optimal solution set; a genetic algorithm is applied to the Pareto optimal solution set to obtain a global optimal call chain; an abstract syntax tree is constructed according to the global optimal call chain, and the abstract syntax tree is converted to code generation to obtain a target recovery operation code; a static code analysis is performed on the target recovery operation code to obtain a code dependency graph; the target recovery operation code is modularly packaged according to the code dependency graph to obtain a target recovery operation; An execution coordination module is used to decompose the target recovery operation into multiple subtasks; construct a directed acyclic graph based on the multiple subtasks, topologically sort the directed acyclic graph, and obtain a task execution sequence; apply a distributed consistency protocol to the task execution sequence to obtain a global consistency state; and schedule resources for the target operating system based on the global consistency state to obtain a final execution result.

Images