Business process learning and automatic execution method and system of adaptive simulation engine

Abstract

This invention relates to the field of business process automation technology, providing a method and system for business process learning and automated execution using an adaptive simulation engine. The method acquires operation behavior sequences, interface state changes, and rule document data; performs stage division and establishes mapping relationships to generate a structured process representation; determines the target execution branch based on matching current state parameters with triggering conditions; achieves cross-system interaction and data transfer; and optimizes execution strategies by extracting performance indicators and anomaly features from execution records. This invention enables intelligent learning and automatic optimization of business processes, improving system operating efficiency and reducing cross-system operational complexity.

Description

Technical Field

[0001] This invention relates to the field of business process automation technology, and in particular to a method and system for business process learning and automated execution using an adaptive simulation engine. Background Technology

[0002] As enterprises deepen their digital transformation, business process automation has become a key means to improve operational efficiency. Business processes typically involve collaboration between multiple information systems. Manually operating these cross-system processes is not only time-consuming and labor-intensive but also prone to errors. Traditional business process automation technologies are mainly based on predefined rules and fixed process paths, using process engines or Robotic Process Automation (RPA) tools to automate specific tasks.

[0003] Existing technologies in the field of business process automation mainly include process mining technology, RPA technology, and rule-based workflow engines. Process mining technology identifies the execution path of business processes by analyzing system logs; RPA technology achieves automation at the interface level by simulating human operations; and rule-based workflow engines guide the automatic execution of business processes through a set of predefined rules.

[0004] Existing technologies for business process learning and automated execution suffer from the following shortcomings and deficiencies: First, they struggle to adapt to dynamic changes in business processes. Traditional automation methods rely on fixed process definitions; when business rules change or system interfaces are updated, manual reconfiguration of the process model is required, lacking adaptability and the ability to dynamically adjust execution strategies based on actual performance. Second, cross-system process automation is difficult. Existing technologies primarily target process automation within a single system, struggling to effectively handle data transfer and context switching between multiple systems, resulting in low levels of automation across business processes and inefficient inter-system collaboration. Third, existing technologies lack intelligent optimization mechanisms for process execution. Traditional automation technologies cannot automatically extract performance metrics and anomalies from historical execution data, making it difficult to self-optimize operation sequences and parameters, leading to low automation efficiency and poor stability. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for business process learning and automated execution of an adaptive simulation engine, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of the present invention provides a method for business process learning and automated execution of an adaptive simulation engine, comprising:

[0007] Acquire operational behavior sequence data, interface state change data, and rule document data during the execution of the business process;

[0008] The operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships is determined, and a structured process representation is generated.

[0009] Collect current process status parameters, match the current process status parameters with the trigger conditions in the structured process representation, and determine the target execution branch based on the matching result;

[0010] Based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology to generate cross-system execution records.

[0011] Based on the cross-system execution records, the execution performance indicators and abnormal features of each operation node are extracted. Based on the execution performance indicators and abnormal features, an operation sequence adjustment strategy and an operation parameter optimization strategy are generated. The strategy is used to update the structured process representation.

[0012] The operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data, and the triggering conditions in the process semantic relationships is determined, generating a structured process representation, including:

[0013] Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data, and align the interface element location features with the state transition sequence in the interface state change data in time sequence to determine the interface state differences before and after each operation node is triggered;

[0014] Semantic parsing is performed on the rule document data to identify the logical relationship between the described conditional expressions and the execution actions, and key entities and their attribute constraints are extracted as semantic representations of the triggering conditions;

[0015] By calculating the semantic similarity between the operation type features of the operation node and the semantic representation of the triggering condition, a first mapping association between the operation node and the triggering condition is established.

[0016] By analyzing the degree of matching between the interface state differences and the entity attribute constraints in the triggering conditions, a second mapping association between interface state change data and triggering conditions is established;

[0017] The first mapping association and the second mapping association are merged to form a ternary mapping relationship between operation nodes, interface state change data and trigger conditions, and a structured process representation containing node identifiers, state transition rules and trigger condition constraints is generated based on the ternary mapping relationship.

[0018] Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data, and align the interface element location features with the state transition sequence in the interface state change data in time sequence to determine the interface state differences before and after each operation node is triggered, including:

[0019] The operation behavior sequence data is parsed to extract the recorded interface element identification information and interaction action type information. The interface element identification information is converted into interface element positioning features containing position coordinates and hierarchical relationships, and the interaction action type information is classified into operation type features.

[0020] The interface state change data is obtained by retrieving a sequence of interface snapshots recorded in chronological order. By comparing the structural and content changes between adjacent interface snapshots, a state transition sequence describing the interface evolution process is generated.

[0021] Based on the execution timestamps of each operation node in the operation behavior sequence data, the state transition node corresponding to the time of each operation node is located in the state transition sequence, and a temporal correspondence between the operation node and the state transition node is established;

[0022] For each operation node, extract the preceding and following interface snapshots of the state transition node corresponding to its time sequence. Based on the difference between the interface element positioning features in the preceding and following interface snapshots, determine the interface state difference before and after the operation node is triggered.

[0023] Collect current process status parameters, match the current process status parameters with the trigger conditions in the structured process representation, and determine the target execution branch based on the matching result, including:

[0024] By monitoring the completion status of the currently executing operation node, the element attribute values ​​of the current interface, and the field content of the current business data, the current process status parameters that characterize the current process status are collected.

[0025] Extract subsequent candidate branches associated with the current operation node and their corresponding triggering conditions from the structured process representation, wherein each triggering condition consists of a condition expression and a constraint range;

[0026] The interface element attribute values ​​in the process state parameters are compared with the judgment criteria of the condition expression in the trigger conditions to calculate the degree of conformity between the trigger conditions of each candidate branch and the current state.

[0027] Candidate branches that meet the trigger condition constraints are selected based on the degree of compliance. When there are multiple candidate branches that meet the conditions, a unique target execution branch is determined based on the branch priority recorded in the structured process representation.

[0028] Based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology, and cross-system execution records are generated, including:

[0029] Based on the sequence of operation nodes corresponding to the target execution branch in the structured process representation, the system identifier and interface element positioning features associated with each operation node are extracted, and the system switching node is identified by analyzing the changes in the system identifiers of adjacent operation nodes;

[0030] For each system switching node, the interface state change data corresponding to that node is obtained from the mapping relationship in the structured process representation, and the interface element features that identify the system interaction entry point are extracted as the access entry point of the target system;

[0031] During system switching, the interface element corresponding to the access point is located by simulating user interaction behavior, and an activation operation is performed to complete the switch from the source system to the target system;

[0032] After the system switch is completed, the business data fields that need to be transferred from the source system to the target system are determined based on the data dependency relationships recorded in the structured process representation.

[0033] The business data fields are filled into the corresponding interface elements of the target system by simulating data entry operations, and the cross-system operation trajectory of each system switch is recorded as the cross-system execution record.

[0034] Based on the cross-system execution records, the execution performance indicators and anomaly characteristics of each operation node are extracted. Based on the execution performance indicators and anomaly characteristics, operation sequence adjustment strategies and operation parameter optimization strategies are generated, including:

[0035] The switching time and operation execution status of each operation node are extracted from the cross-system execution records. The execution time of each operation node is obtained by calculating the time interval between adjacent switching times. The retry frequency is obtained by counting the number of times marked as failure or retry in the operation execution status. The execution time and retry frequency are used as execution performance indicators.

[0036] Identify operation nodes marked as abnormal in the cross-system execution records, extract the system identifier, interface element features, and business data fields corresponding to the operation node, and combine them to form abnormal features describing the scenario in which the abnormality occurred;

[0037] Based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm to generate an operation order adjustment strategy and an operation parameter optimization strategy. The operation order adjustment strategy is used to reduce the total execution time of the process, and the operation parameter optimization strategy is used to reduce the frequency of operation retries.

[0038] Based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm, generating an operation order adjustment strategy and an operation parameter optimization strategy, including:

[0039] The reinforcement learning state representation is determined based on the execution performance metrics and the anomaly features;

[0040] Based on the reinforcement learning state representation, adjusting the execution position of the operation node in the operation node sequence is taken as the first action type, and modifying the interface element positioning method and operation execution interval of the operation node is taken as the second action type, thus establishing an action space;

[0041] Execute the first action type in the action space, swap the order of operation nodes with data dependencies in the operation node sequence, calculate the total execution time of the process after the swap, and use the reduction in execution time as the first reward feedback;

[0042] Update the reinforcement learning strategy parameters based on the first reward feedback, select the node order adjustment scheme that minimizes the total execution time of the process, and generate the operation order adjustment strategy;

[0043] Execute the second action type in the action space, adjust the recognition rules of the corresponding interface elements and the waiting time for the operation trigger for the operation node corresponding to the abnormal feature, calculate the adjusted operation retry frequency, and use the reduction in retry frequency as the second reward feedback;

[0044] The reinforcement learning strategy parameters are updated based on the second reward feedback. The parameter configuration scheme that minimizes the frequency of operation retries is selected, and an operation parameter optimization strategy is generated.

[0045] A second aspect of the present invention provides a business process learning and automated execution system for an adaptive simulation engine, comprising:

[0046] The first unit is used to acquire operation behavior sequence data, interface state change data, and rule document data during the execution of the business process;

[0047] The second unit is used to divide the operation behavior sequence data into stages, identify operation nodes in each process stage, extract process semantic relationships from the rule document data, determine the mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships, and generate a structured process representation;

[0048] The third unit is used to collect the current process status parameters, match the current process status parameters with the triggering conditions in the structured process representation, and determine the target execution branch based on the matching results;

[0049] The fourth unit is used to determine the interaction entry points of each system in the cross-system process based on the structured process representation and the target execution branch, and to perform system switching and data transfer through simulation operation technology to generate cross-system execution records;

[0050] The fifth unit is used to extract the execution performance indicators and abnormal features of each operation node based on the cross-system execution records, and to generate operation sequence adjustment strategies and operation parameter optimization strategies based on the execution performance indicators and abnormal features. The strategies are used to update the structured process representation.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] processor;

[0053] Memory used to store processor-executable instructions;

[0054] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0056] This invention achieves a structured representation of the process by intelligently dividing the sequence data of operational behaviors into stages and identifying operational nodes, and combining the semantic relationships in the rule document, thus solving the problem of insufficient understanding of business processes in traditional process automation.

[0057] This invention can perform real-time matching between the current process status parameters and the triggering conditions in the structured process representation, intelligently determine the target execution branch, make the process execution dynamically adaptable, and effectively cope with process change scenarios.

[0058] This invention solves the technical problem that traditional automation tools struggle to connect systems in cross-system business scenarios by identifying and determining the interaction entry points of each system in a cross-system process and using simulation operation technology to achieve automatic switching and data transfer between systems.

[0059] This invention extracts the execution performance indicators and anomaly characteristics of operation nodes based on cross-system execution records, automatically generates operation sequence adjustment strategies and parameter optimization strategies, realizes continuous updating and optimization of process representation, and improves the stability and efficiency of the system.

[0060] This invention eliminates the need for extensive manual writing of automation scripts. By automatically learning and adaptively executing business processes, it significantly reduces the implementation cost and maintenance difficulty of business process automation. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating the business process learning and automated execution method of the adaptive simulation engine according to an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of the process for determining the target execution branch in an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0065] Figure 1 This is a flowchart illustrating the business process learning and automated execution method of the adaptive simulation engine according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0066] Acquire operational behavior sequence data, interface state change data, and rule document data during the execution of the business process;

[0067] The operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships is determined, and a structured process representation is generated.

[0068] Collect current process status parameters, match the current process status parameters with the trigger conditions in the structured process representation, and determine the target execution branch based on the matching result;

[0069] Based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology to generate cross-system execution records.

[0070] Based on the cross-system execution records, the execution performance indicators and abnormal features of each operation node are extracted. Based on the execution performance indicators and abnormal features, an operation sequence adjustment strategy and an operation parameter optimization strategy are generated. The strategy is used to update the structured process representation.

[0071] In one optional implementation, the operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data, and the triggering conditions in the process semantic relationships is determined, generating a structured process representation, including:

[0072] Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data, and align the interface element location features with the state transition sequence in the interface state change data in time sequence to determine the interface state differences before and after each operation node is triggered;

[0073] Semantic parsing is performed on the rule document data to identify the logical relationship between the described conditional expressions and the execution actions, and key entities and their attribute constraints are extracted as semantic representations of the triggering conditions;

[0074] By calculating the semantic similarity between the operation type features of the operation node and the semantic representation of the triggering condition, a first mapping association between the operation node and the triggering condition is established.

[0075] By analyzing the degree of matching between the interface state differences and the entity attribute constraints in the triggering conditions, a second mapping association between interface state change data and triggering conditions is established;

[0076] The first mapping association and the second mapping association are merged to form a ternary mapping relationship between operation nodes, interface state change data and trigger conditions, and a structured process representation containing node identifiers, state transition rules and trigger condition constraints is generated based on the ternary mapping relationship.

[0077] This invention divides the operation behavior sequence data into stages, identifies operation nodes in the process stages, extracts process semantic relationships from rule documents, and establishes a mapping between operation nodes, interface state changes, and triggering conditions, thereby generating a structured process representation.

[0078] First, the operation sequence data is divided into stages. Operation sequence data typically includes a series of actions performed by the user in the software system, such as clicking a button, entering text, and selecting options. Stage division is based on clustering of operation actions according to time intervals and semantic relationships. Specifically, the time interval between adjacent operations is calculated; when the time interval exceeds a preset threshold (e.g., 30 seconds), it indicates the start of a different stage. Simultaneously, the semantic relationships between operations are analyzed, such as whether the operation objects belong to the same functional module, thus dividing the operation sequence into multiple stages. For example, in an electronic form processing workflow, this includes stages such as form filling, information confirmation, and submission for approval.

[0079] Next, we identify the operation nodes in each process stage. Operation nodes refer to the specific actions performed by the user on the interface, such as clicking the submit button or entering text in a text box. When identifying operation nodes, we extract the type of each operation (click, input, drag, etc.), the target element (button, text box, drop-down menu, etc.), and the operation parameters (input text content, selected option value, etc.).

[0080] Then, semantic relationships of the process are extracted from the rule document data. Rule documents typically describe business rules and process logic in natural language, such as "the submit button will only be activated after the user fills in the required fields" or "advanced approval is only required when the amount is greater than 1000". Through natural language processing technology, conditional statements (such as "if", "when", "only when", etc.) and result statements (such as "then", "only then", "required", etc.) in the document are identified, and the logical relationships between the triggering conditions and execution actions described therein are extracted.

[0081] Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data. Interface element location features include the element's XPath path, CSS selector, element ID, or other unique identifier. Operation type features include operation types such as click, input, and drag, as well as related parameters. For example, for the operation "Enter admin in the username text box", the interface element location feature is " / / input[@id='username']", the operation type is "input", and the parameter is "admin".

[0082] The system aligns the location features of interface elements with the state transition sequence in the interface state change data to determine the differences in interface state before and after each operation node is triggered. The interface state change data records the changes in interface element attributes (such as visibility, enabled status, content, etc.) over time. The system aligns the operation behavior with the interface state change using timestamps to calculate the differences in the interface before and after each operation. For example, after clicking the "Submit" button, the form state changes from "Editable" to "Read-only," and a "Submission Successful" message appears.

[0083] Semantic parsing is performed on the rule document data to identify the logical relationship between the described conditional expressions and the execution actions. Natural language processing techniques, such as dependency parsing and semantic role labeling, are used to extract the conditional and result parts from the rule description. The conditional part typically contains key entities (such as forms or amounts) and their attribute constraints (such as required fields being filled or amounts being greater than 1000), while the result part describes the system behavior or state changes after the conditions are met.

[0084] By calculating the semantic similarity between the operation type features of the operation node and the semantic representation of the triggering condition, a first mapping relationship between the operation node and the triggering condition is established. The semantic similarity calculation can be based on word vector models, such as Word2Vec or BERT. For example, "click the submit button" in the rule document and "click the button with the id 'submit'" in the operation sequence have high semantic similarity and a mapping relationship can be established.

[0085] By analyzing the degree of matching between interface state differences and entity attribute constraints in trigger conditions, a second mapping relationship is established between interface state change data and trigger conditions. For example, the condition "all required fields are filled" in the rule "when all required fields are filled, the submit button becomes available" can be mapped to the interface state change "the submit button changes from disabled to enabled".

[0086] The first and second mapping associations are merged to form a ternary mapping relationship between operation nodes, interface state change data, and triggering conditions. During the fusion process, if the operation node and the triggering condition have high semantic similarity, and the interface state change caused by the operation is consistent with the result described in the triggering condition, then an association is established between the three. For example, the operation of "clicking the submit button" matches the triggering condition of the rule "the system enters the review state after clicking submit," and at the same time, the interface state changes from "editing" to "under review," thus forming a mapping relationship among the three.

[0087] Based on a ternary mapping relationship, a structured process representation is generated, including node identifiers, state transition rules, and trigger condition constraints. This structured representation can be in JSON or XML format. Each node contains a unique identifier, an operation type description, preconditions, post-state changes, and associated business rules. This structured representation can be used for process automation, process validation, or generating process documentation. For example, the structured representation of a form submission node includes a node ID, submission operation, precondition "all required fields are filled," state change "form status changes from edit to submitted," and business rule "approval is required after submission."

[0088] The above methods can transform unstructured sequences of operational behaviors and rule documents into structured process representations, facilitating process analysis, optimization, and automated execution, thereby improving the efficiency and accuracy of business process management.

[0089] In one optional implementation, the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data are extracted, and the interface element location features are time-aligned with the state transition sequence in the interface state change data to determine the interface state differences before and after each operation node is triggered, including:

[0090] The operation behavior sequence data is parsed to extract the recorded interface element identification information and interaction action type information. The interface element identification information is converted into interface element positioning features containing position coordinates and hierarchical relationships, and the interaction action type information is classified into operation type features.

[0091] The interface state change data is obtained by retrieving a sequence of interface snapshots recorded in chronological order. By comparing the structural and content changes between adjacent interface snapshots, a state transition sequence describing the interface evolution process is generated.

[0092] Based on the execution timestamps of each operation node in the operation behavior sequence data, the state transition node corresponding to the time of each operation node is located in the state transition sequence, and a temporal correspondence between the operation node and the state transition node is established;

[0093] For each operation node, extract the preceding and following interface snapshots of the state transition node corresponding to its time sequence. Based on the difference between the interface element positioning features in the preceding and following interface snapshots, determine the interface state difference before and after the operation node is triggered.

[0094] First, the operation behavior sequence data is parsed to extract interface element identification information and interaction action type information. This data is typically stored in structured data formats, such as JSON or XML, containing detailed information for each user operation. During parsing, for each operation node, interface element identification information such as elementId, elementPath, and elementType, as well as interaction action type information such as click, scroll, and input, are extracted.

[0095] The extracted UI element identifier information is converted into UI element positioning features that include position coordinates and hierarchical relationships. Specifically, the position of the corresponding UI element in the view tree is queried based on elementId or elementPath, and its absolute coordinates (x, y), relative coordinates (relX, relY), width and height (width, height), as well as its depth and parent-child relationship (parentId, childrenIds) in the view hierarchy are obtained. At the same time, attributes such as the element's visibility and clickable area are recorded to form complete UI element positioning features.

[0096] Interaction type information is categorized into operation type features, mainly including: click (tap, click, doubleClick), swipe (swipe, scroll, drag), input (input, textChange), long press (longPress), and gesture (gesture, pinch). For each operation type, its unique parameters are further extracted. For example, input operations need to record the input text content, while swipe operations need to record the coordinates of the starting and ending points, speed, and direction.

[0097] Next, the sequence of interface snapshots recorded chronologically from the interface state change data is obtained. Interface snapshots typically contain the interface structure tree (DOM tree or view tree), element attribute information, style information, etc. Adjacent interface snapshots are compared and analyzed to identify structural and content changes. Structural changes refer to the addition, deletion, or hierarchical changes of interface elements, which are identified by comparing the differences in the view tree structure between two snapshots. Content changes refer to changes in the text, image, style, and other attributes of elements, which are identified by comparing the differences in the attribute values ​​of the elements.

[0098] By comparing the differences between adjacent snapshots, a state transition sequence describing the evolution of the interface is generated. Each state transition node includes: the interface state ID before the transition, the interface state ID after the transition, the timestamp of the transition, the set of changed elements involved in the transition (changedElements), the change type (changeType), and the change content description (changeDescription). Change types include: element addition (ElementAdded), element deletion (ElementRemoved), element attribute change (AttributeChanged), element content change (ContentChanged), element position change (PositionChanged), etc.

[0099] Based on the execution timestamps of each operation node in the operation behavior sequence data, the state transition node corresponding to the time of each operation node is located in the state transition sequence. Specifically, for the timestamp T_O of operation node O, the state transition node S with the closest timestamp and greater than T_O is searched in the state transition sequence. If the time difference (T_S - T_O) is less than a preset threshold (e.g., 300 milliseconds), it is considered that the state transition node S was triggered by operation node O, and a correspondence between O and S is established; if the time difference is too large, there is an intermediate state that has not been recorded correctly, and exception handling is required.

[0100] For each operation node with an established time-series correspondence, extract the preceding and following interface snapshots of its corresponding state transition node. Based on the previously extracted interface element location features, locate the elements involved in the operation in the preceding and following interface snapshots, and analyze the changes of these elements in the preceding and following snapshots.

[0101] Specifically, a click operation can cause different types of state changes, such as a change in the state of the target element (e.g., button highlighting), triggering the loading of a new interface (e.g., page navigation), or displaying a new element (e.g., a pop-up dialog box). By comparing the attribute differences of the target element and its related elements in the snapshots before and after, these state changes are identified and categorized as: element state change (ElementStateChange), page navigation (PageNavigation), content update (ContentUpdate), and layout adjustment (LayoutAdjustment).

[0102] In practical applications, such as mobile application automated testing scenarios, this method can be used to analyze the differences in the interface state before and after the test tool executes the click of the submit button. By extracting the element positioning features (such as the button's ID and coordinates) from the operation node and combining them with time-aligned state transition data, it can be determined that this click operation caused a series of interface changes, such as form submission, page redirection, and loading of new content. This information is of great value for understanding the actual effect of the operation, verifying the correctness of the function, and diagnosing the reasons for operation failure.

[0103] The differences in the interface state before and after the operation node is triggered are output in structured data format, including: operation ID, operation type, operation element characteristics, preceding state ID, succeeding state ID, set of state change types, set of changed elements, and their specific change descriptions. This data can be used for subsequent applications such as operation intent analysis, behavior pattern mining, and automated script generation.

[0104] The above methods can accurately extract the correlation between user actions and interface state changes, providing a data foundation for understanding user interaction intentions and application response mechanisms, and supporting more efficient human-computer interaction analysis and automated testing.

[0105] In one optional implementation, current process state parameters are collected, and the current process state parameters are matched with triggering conditions in the structured process representation. Based on the matching result, the target execution branch is determined, including:

[0106] By monitoring the completion status of the currently executing operation node, the element attribute values ​​of the current interface, and the field content of the current business data, the current process status parameters that characterize the current process status are collected.

[0107] Extract subsequent candidate branches associated with the current operation node and their corresponding triggering conditions from the structured process representation, wherein each triggering condition consists of a condition expression and a constraint range;

[0108] The interface element attribute values ​​in the process state parameters are compared with the judgment criteria of the condition expression in the trigger conditions to calculate the degree of conformity between the trigger conditions of each candidate branch and the current state.

[0109] Candidate branches that meet the trigger condition constraints are selected based on the degree of compliance. When there are multiple candidate branches that meet the conditions, a unique target execution branch is determined based on the branch priority recorded in the structured process representation.

[0110] Figure 2 This is a schematic diagram illustrating the process of determining the target execution branch according to an embodiment of the present invention. Figure 2As shown, firstly, multi-dimensional monitoring is used to collect current process status parameters. Specifically, listeners are set up to continuously monitor the status of currently executing operation nodes, including node completion flags, execution progress, and execution results. For example, for form submission operations, monitoring includes whether the form was successfully submitted, whether the form data was received by the system, and the response code returned by the server. Simultaneously, element recognition technology is used to capture the element attribute values ​​of the current interface in real time, such as whether buttons are clickable, whether input boxes are empty, the currently selected value in a dropdown list, and the content of pop-up prompts. Furthermore, data interfaces are used to access the field content of current business data, such as order status, user balance, and inventory quantity. These data collectively constitute a parameter set representing the current process status.

[0111] Next, the subsequent candidate branches and their triggering conditions associated with the current operation node are extracted from the structured process representation. The structured process representation is typically stored in the form of a directed graph or decision tree, where each node represents an operation step, and the connections between nodes represent the execution path. Starting from the current operation node, all directly connected downstream nodes are identified as candidate branches. For each candidate branch, its corresponding triggering condition is extracted. Each triggering condition consists of two parts: a condition expression and a constraint range. The condition expression defines the judgment logic, such as order amount > 100, sufficient inventory status, or the page containing the text "payment successful." The constraint range defines the range that the result of the condition expression must satisfy, such as a Boolean condition must be true, a numerical condition must be within a specific range, or a string condition must match a specific pattern.

[0112] Subsequently, the process status parameters and trigger conditions are matched and compared. For each candidate branch's trigger condition, the current process status parameters are substituted into the condition expression for calculation. For example, if the trigger condition is "(Inventory Quantity 100) AND (User Level = 'VIP')", the results of the sub-expressions are calculated separately and then logically combined. During the matching process, a fuzzy matching algorithm is used to calculate the degree of conformity between each candidate branch's trigger condition and the current state. For exact matching conditions (such as equality comparison), the conformity is 100% or 0%; for fuzzy conditions (such as text similarity), the similarity between the actual value and the expected value is calculated as the conformity.

[0113] Based on the matching results above, candidate branches that meet the trigger condition constraints are selected. First, candidate branches with a compliance rate below the lower limit of the trigger condition constraint range are filtered out. If the compliance rate reaches or exceeds the upper limit of the constraint range, the candidate branch is considered to fully meet the trigger condition. When multiple candidate branches meet the conditions, the final target execution branch is determined based on the predefined branch priorities in the structured process representation. Branch priorities are set based on business importance, execution frequency, or expected results. For example, exception handling branches typically have higher priorities, while regular business processing branches are ranked according to the importance of the business scenario.

[0114] In practical applications, let's take the order processing flow of an e-commerce platform as an example. When the "payment confirmation" node is reached, parameters such as payment result, account balance, and inventory status are monitored. Assume the structured process representation defines three candidate branches: payment success branch (condition: the page contains the text "payment successful"), insufficient balance branch (condition: account balance < order amount), and insufficient inventory branch (condition: product inventory < order quantity). When the user completes the payment operation, the page displays a "payment successful" message, and it is confirmed that the account balance and inventory are sufficient. At this point, the trigger condition for the "payment success branch" is met 100%, and it is designated as the target execution branch, continuing with subsequent order confirmation and logistics arrangement operations.

[0115] Through the above, precise matching of current process status parameters with triggering conditions in structured process representation is achieved, ensuring that the process executes according to the expected path in complex business scenarios, thereby improving the accuracy and robustness of process automation.

[0116] In one optional implementation, based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology to generate cross-system execution records, including:

[0117] Based on the sequence of operation nodes corresponding to the target execution branch in the structured process representation, the system identifier and interface element positioning features associated with each operation node are extracted, and the system switching node is identified by analyzing the changes in the system identifiers of adjacent operation nodes;

[0118] For each system switching node, the interface state change data corresponding to that node is obtained from the mapping relationship in the structured process representation, and the interface element features that identify the system interaction entry point are extracted as the access entry point of the target system;

[0119] During system switching, the interface element corresponding to the access point is located by simulating user interaction behavior, and an activation operation is performed to complete the switch from the source system to the target system;

[0120] After the system switch is completed, the business data fields that need to be transferred from the source system to the target system are determined based on the data dependency relationships recorded in the structured process representation.

[0121] The business data fields are filled into the corresponding interface elements of the target system by simulating data entry operations, and the cross-system operation trajectory of each system switch is recorded as the cross-system execution record.

[0122] In the process of automating cross-system processes, the technical implementation of determining the interaction entry points of each system and realizing system switching and data transfer based on structured process representation and target execution branches is as follows.

[0123] First, based on the sequence of operation nodes corresponding to the target execution branch in the structured process representation, the system identifiers and interface element location features associated with each operation node are extracted. Specifically, the sequence of operation nodes corresponding to the target execution branch is queried from the data structure of the structured process representation. This sequence is stored in the form of an ordered list, and each operation node contains attributes such as node identifier, associated system identifier, operation type, and interface element location features of the operation object. The interface element location features can be stored in the form of XPath expressions, CSS selectors, or image feature descriptions for subsequent precise location of interface elements.

[0124] System switching nodes are identified by analyzing changes in the system identifiers of adjacent operation nodes. Specifically, the sequence of operation nodes is traversed, and the system identifiers of adjacent nodes are compared. When a change in the system identifier is detected, the current node is marked as a system switching node. For example, if the system identifier of the nth node in the operation sequence is SystemA, and the system identifier of the (n+1)th node is SystemB, then the nth node is marked as a system switching node switching from SystemA to SystemB.

[0125] For each identified system switching node, the interface state change data corresponding to that node is obtained from the mapping relationship in the structured process representation. The structured process representation contains the mapping relationship between operation nodes and interface state changes, and stores the interface state differences before and after each operation. The interface state change data corresponding to the system switching node is retrieved from the mapping relationship. This data includes the interface snapshot before the operation, the interface snapshot after the operation, and records of interface element changes.

[0126] This process extracts interface element features that identify system interaction entry points from interface state change data. Specifically, it analyzes key elements in the interface state change data, such as interactive elements like menu items, buttons, and links, and combines these with predefined system interaction entry feature templates (such as specific icons, text labels, or location features) to identify interface elements that can trigger system switching. The positioning features of these elements (such as the element's XPath, coordinate position, and appearance features) are then extracted as entry information for accessing the target system.

[0127] When switching systems, the interface elements corresponding to the access points are located by simulating user interaction behavior. Based on the interface element location features extracted in the previous steps, matching elements are searched in the source system interface. The location process can employ various techniques: DOM-based element search (suitable for web applications), image recognition-based element matching (suitable for desktop applications), or coordinate positioning (suitable for fixed-layout interfaces). When element location is uncertain, a fuzzy matching algorithm can be used, combined with auxiliary information such as the element's text content and relative positional relationships, to improve location accuracy.

[0128] After locating the target element, an activation operation is performed to complete the switch from the source system to the target system. The activation operation typically simulates a mouse click event, but depending on the characteristics of different systems, it may also need to simulate complex interactive behaviors such as keyboard shortcuts and drag-and-drop. After performing the activation operation, the interface changes are monitored to confirm that the target system has been successfully loaded. If the system switch fails, a retry or alternative solution is executed according to the preset exception handling strategy.

[0129] After the system switchover is complete, based on the data dependencies recorded in the structured process representation, the business data fields that need to be transferred from the source system to the target system are determined. These data dependencies are stored in the structured process representation as a directed graph, where nodes represent data fields and edges represent data flow directions. By querying this graph structure, the list of data fields flowing from the source system to the target system and their mapping relationships can be obtained.

[0130] For the business data fields that need to be transmitted, the corresponding data values ​​are extracted from the source system. Data extraction can be achieved through methods such as obtaining text from interface elements, calling data interfaces, or directly querying the database. The extracted data undergoes necessary format conversion and validation to ensure it conforms to the data specifications of the target system.

[0131] Business data fields are populated into the corresponding interface elements of the target system by simulating data entry operations. Appropriate data entry methods are used based on the type of the target system's interface elements (such as input boxes, drop-down menus, date pickers, etc.). For example, for text input boxes, keyboard input events are simulated; for drop-down menus, the sequence of clicks to expand and select options is simulated. During the data entry process, adjustments are made in real time based on the response status of the interface elements to ensure correct data entry.

[0132] Throughout the process, the cross-system operation trajectory for each system switch is recorded as a cross-system execution log. The execution log includes the system switch timestamp, source system identifier, target system identifier, interaction entry point information, transmitted data fields and their values, execution status, and other information. These records are stored in structured log format for subsequent execution auditing, problem investigation, and process optimization.

[0133] In practical applications, taking the enterprise financial reimbursement process as an example, when a user needs to switch from the OA system to the financial system, the above technical solution can automatically identify the "financial system entry" button in the OA system, simulate the click operation to complete the system switch, and automatically fill in the reimbursement form number, amount, department and other information in the OA system into the corresponding form in the financial system, which greatly improves the execution efficiency and accuracy of cross-system business processes.

[0134] In one optional implementation, the execution performance indicators and anomaly characteristics of each operation node are extracted based on the cross-system execution records, and an operation sequence adjustment strategy and operation parameter optimization strategy are generated based on the execution performance indicators and anomaly characteristics, including:

[0135] The switching time and operation execution status of each operation node are extracted from the cross-system execution records. The execution time of each operation node is obtained by calculating the time interval between adjacent switching times. The retry frequency is obtained by counting the number of times marked as failure or retry in the operation execution status. The execution time and retry frequency are used as execution performance indicators.

[0136] Identify operation nodes marked as abnormal in the cross-system execution records, extract the system identifier, interface element features, and business data fields corresponding to the operation node, and combine them to form anomaly features describing the scenario in which the anomaly occurred;

[0137] Based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm to generate an operation order adjustment strategy and an operation parameter optimization strategy. The operation order adjustment strategy is used to reduce the total execution time of the process, and the operation parameter optimization strategy is used to reduce the frequency of operation retries.

[0138] First, the switching time and execution status of each operation node are extracted from the cross-system execution records. In practice, cross-system execution records are usually stored in log files, containing information such as the timestamp of the operation, operation type, target system identifier, interface element identifier, business data, and execution status. By parsing these log files, the switching time of each operation node, i.e., the start and end times of the operation, and the execution status of the operation (such as success, failure, retry, etc.) can be obtained.

[0139] Next, the execution time of each operation node is obtained by calculating the time interval between adjacent switching moments. Specifically, for operation node A, its execution time is equal to the difference between its end time and start time. For example, if the start time of operation node A is 2023-05-10 14:30:25.512 and the end time is 2023-05-10 14:30:28.765, then the execution time of this node is 3.253 seconds. This calculation method can be applied to all operation nodes to obtain a complete execution time dataset.

[0140] Meanwhile, by counting the number of times the operation execution status is marked as failed or retried, the retry frequency of each operation node can be obtained. During cross-system execution, due to factors such as network latency and slow system response, some operations require multiple attempts to complete successfully. By analyzing the status markers in the execution record, the retry frequency of each operation node can be calculated. For example, if a form submission operation succeeds after 3 retries in one process execution, then the retry frequency of that operation node is 3.

[0141] The execution time and retry frequency obtained above are used as performance indicators of the operation node. These indicators directly reflect the execution efficiency and stability of the operation node. The shorter the execution time, the faster the operation is executed; the lower the retry frequency, the more stable the operation is executed.

[0142] Next, identify operation nodes marked as abnormal in the cross-system execution records. Abnormal states include various situations such as execution failure, timeout, and data inconsistency. When an abnormal node is found, extract the system identifier (such as ERP system, CRM system, etc.), interface element features (such as button ID, form name, etc.), and business data fields (such as order number, customer name, etc.) corresponding to the operation node, and combine this information to form an abnormal feature describing the scenario in which the abnormality occurred.

[0143] For example, for an operation node that creates customer information in a CRM system, if an anomaly occurs, the extracted anomaly characteristics include: the system identifier is CRM-V3.5, the interface element characteristic is customer_creation_form, and the business data fields are customer_name=XYZ and customer_type=enterprise. These anomaly characteristics describe in detail the specific environment and conditions under which the anomaly occurred, which is helpful for subsequent analysis and optimization.

[0144] Based on the extracted execution performance metrics and anomaly features, a reinforcement learning algorithm is used to optimize the execution order and operation parameters of the operation node sequence. The reinforcement learning model regards the operation node sequence as the state space, the adjustment of the operation order and the change of parameters as the action space, and the reduction of the total execution time and the reduction of the retry frequency as reward signals.

[0145] In reinforcement learning, a state-action value function Q(s,a) is first established, representing the expected cumulative reward for taking action a in state s. The Q function is continuously updated using algorithms such as Q-learning or Deep Q-Network to find the optimal policy. Specifically, the performance metrics and anomaly features of the operation nodes can be used as state feature inputs. After processing by the neural network, the Q-value of each action is output, and the action with the highest Q-value is selected for execution.

[0146] Through the reinforcement learning process described above, operation sequence adjustment strategies and operation parameter optimization strategies are generated. The operation sequence adjustment strategy mainly targets execution time metrics. By rearranging the execution order of operation nodes, waiting time and system switching overhead are reduced, thereby lowering the total process execution time. For example, if it is found that multiple consecutive operations in system A are interspersed with operations in system B, resulting in frequent system switching, the operation sequence can be adjusted to concentrate the execution of operations in the same system as much as possible, reducing the number of system switching.

[0147] Operation parameter optimization strategies primarily target retry frequency metrics. By adjusting operation parameter values, waiting times, and retry strategies, the failure rate can be reduced, thereby decreasing the number of retries. For example, if a form submission operation frequently retryes due to data validation failures, the format or value range of input parameters can be optimized to improve the first-time success rate. If a slow system response leads to timeout retries, the waiting time parameter can be appropriately increased to avoid unnecessary retries.

[0148] The above methods use a data-driven approach to automatically analyze the execution performance and anomalies of cross-system operations and generate optimization strategies, effectively improving the execution efficiency and stability of cross-system operation processes and reducing manual intervention and maintenance costs.

[0149] In one optional implementation, based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm, generating an operation order adjustment strategy and an operation parameter optimization strategy, including:

[0150] The reinforcement learning state representation is determined based on the execution performance metrics and the anomaly features;

[0151] Based on the reinforcement learning state representation, adjusting the execution position of the operation node in the operation node sequence is taken as the first action type, and modifying the interface element positioning method and operation execution interval of the operation node is taken as the second action type, thus establishing an action space;

[0152] Execute the first action type in the action space, swap the order of operation nodes with data dependencies in the operation node sequence, calculate the total execution time of the process after the swap, and use the reduction in execution time as the first reward feedback;

[0153] Update the reinforcement learning strategy parameters based on the first reward feedback, select the node order adjustment scheme that minimizes the total execution time of the process, and generate the operation order adjustment strategy;

[0154] Execute the second action type in the action space, adjust the recognition rules of the corresponding interface elements and the waiting time for the operation trigger for the operation node corresponding to the abnormal feature, calculate the adjusted operation retry frequency, and use the reduction in retry frequency as the second reward feedback;

[0155] The reinforcement learning strategy parameters are updated based on the second reward feedback. The parameter configuration scheme that minimizes the frequency of operation retries is selected, and an operation parameter optimization strategy is generated.

[0156] First, the state representation for reinforcement learning needs to be determined based on performance metrics and anomaly characteristics. This state representation includes: the execution time distribution of the current operation sequence, data dependencies between nodes, UI element loading status, execution anomaly frequency statistics, and system resource usage. This information is encoded in vector form to form the state space S. For example, for a sequence containing n operation nodes, an (n×m)-dimensional state matrix can be constructed, where each row represents an operation node, and each column represents different state characteristics of that node, such as execution time, anomaly frequency, and dependencies with other nodes.

[0157] Next, based on the reinforcement learning state representation, an action space is established, containing two types of actions: the first type is adjusting the execution position of the operation node in the operation node sequence; the second type is modifying the UI element positioning method and operation execution interval of the operation node. For the first type of action, it can be defined as a node position swapping operation, that is, selecting two nodes from the sequence and swapping their positions; for the second type of action, it can be defined as adjusting the UI element selector of a specific node (such as switching from XPath to CSS selector) or adjusting the waiting time before the operation is triggered (such as changing from fixed waiting to dynamic waiting).

[0158] Subsequently, the first action type is executed in the action space. First, the data dependencies in the sequence of operation nodes are analyzed to construct a dependency graph. Based on the dependency graph, node pairs with parallel execution potential are selected for sequential swapping experiments. For example, if nodes A and C both depend on the output of node B, but there is no dependency between A and C, their execution order can be swapped. After each swap, the total execution time is recorded and calculated, and the reduction in execution time is used as the first reward. For example, if a swap reduces the total execution time from 120 seconds to 100 seconds, the reward value is 20.

[0159] Based on the initial reward feedback, reinforcement learning algorithms such as policy gradient or Q-learning are used to update the policy parameters. The policy function π(a|s) represents the probability of choosing action a in state s. The policy parameter θ is optimized using gradient ascent to maximize the expected cumulative reward. After multiple rounds of iterative learning, the node order adjustment scheme that minimizes the total execution time of the process is selected, generating the operation order adjustment policy. This policy can be represented as the optimal arrangement of the node sequence.

[0160] After determining the operation sequence adjustment strategy, the second action type is executed in the action space. For operation nodes corresponding to abnormal characteristics, the recognition rules of the corresponding interface elements and the waiting time for operation triggering are adjusted. For example, for operation nodes that frequently retries due to inaccurate positioning of interface elements, the following adjustments can be tried: replacing the absolute path selector with a relative selector; adding multiple positioning conditions; and adjusting the waiting strategy from a fixed duration to waiting for specific conditions to be met. After each adjustment, the operation retry frequency is calculated, and the reduction in retry frequency is used as the second reward feedback. For example, if a parameter adjustment reduces the number of retries from 15 to 3, the reward value is 12.

[0161] Based on the feedback from the second reward, the policy parameters are updated using the same reinforcement learning algorithm. After multiple rounds of learning iterations, the parameter configuration scheme that minimizes the frequency of operation retries is selected to generate an operation parameter optimization strategy. This strategy includes the optimal UI element identification rules and the best waiting time configuration for each abnormal node.

[0162] Taking the automated form filling process as an example, the original operation sequence includes five nodes: opening the page, clicking the login button, entering the username, entering the password, and clicking submit. By applying this method, it was found that there is no strong dependency between entering the username and entering the password, and they can be executed in parallel. Therefore, the adjusted sequence was generated: opening the page, clicking the login button, [entering username, entering password], and clicking submit. The operations within the brackets [] can be executed in parallel, optimizing the process execution time from the original 8 seconds to 5 seconds. Simultaneously, for the login button node, which often fails to recognize elements due to page loading delays, the original strategy of waiting a fixed 1 second was optimized to wait for the button element to become clickable, and a multi-selector (ID + text content) was used for positioning, reducing the number of retries for this node from an average of 4 to 0.

[0163] This reinforcement learning-based optimization method not only improves the efficiency of operational processes but also enhances their stability, enabling automated processes to better adapt to different operating environments and state changes.

[0164] The adaptive simulation engine business process learning and automated execution system of this invention includes:

[0165] The first unit is used to acquire operation behavior sequence data, interface state change data, and rule document data during the execution of the business process;

[0166] The second unit is used to divide the operation behavior sequence data into stages, identify operation nodes in each process stage, extract process semantic relationships from the rule document data, determine the mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships, and generate a structured process representation;

[0167] The third unit is used to collect the current process status parameters, match the current process status parameters with the triggering conditions in the structured process representation, and determine the target execution branch based on the matching results;

[0168] The fourth unit is used to determine the interaction entry points of each system in the cross-system process based on the structured process representation and the target execution branch, and to perform system switching and data transfer through simulation operation technology to generate cross-system execution records;

[0169] The fifth unit is used to extract the execution performance indicators and abnormal features of each operation node based on the cross-system execution records, and to generate operation sequence adjustment strategies and operation parameter optimization strategies based on the execution performance indicators and abnormal features. The strategies are used to update the structured process representation.

[0170] A third aspect of the present invention provides an electronic device, comprising:

[0171] processor;

[0172] Memory used to store processor-executable instructions;

[0173] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0174] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0175] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A business process learning and automated execution method for an adaptive simulation engine, characterized in that, include: Acquire operational behavior sequence data, interface state change data, and rule document data during the execution of the business process; The operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships is determined, and a structured process representation is generated. Collect current process status parameters, match the current process status parameters with the trigger conditions in the structured process representation, and determine the target execution branch based on the matching result; Based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology to generate cross-system execution records. Based on the cross-system execution records, the execution performance indicators and abnormal features of each operation node are extracted. Based on the execution performance indicators and abnormal features, an operation sequence adjustment strategy and an operation parameter optimization strategy are generated. The strategy is used to update the structured process representation. The process of extracting execution performance metrics and anomaly characteristics of each operation node based on the cross-system execution records, and generating operation sequence adjustment strategies and operation parameter optimization strategies based on the execution performance metrics and anomaly characteristics, includes: The switching time and operation execution status of each operation node are extracted from the cross-system execution records. The execution time of each operation node is obtained by calculating the time interval between adjacent switching times. The retry frequency is obtained by counting the number of times marked as failure or retry in the operation execution status. The execution time and retry frequency are used as execution performance indicators. Identify operation nodes marked as abnormal in the cross-system execution records, extract the system identifier, interface element features, and business data fields corresponding to the operation node, and combine them to form abnormal features describing the scenario in which the abnormality occurred; Based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm to generate an operation order adjustment strategy and an operation parameter optimization strategy. The operation order adjustment strategy is used to reduce the total execution time of the process, and the operation parameter optimization strategy is used to reduce the frequency of operation retries.

2. The method according to claim 1, characterized in that, The operation behavior sequence data is divided into stages, operation nodes in each process stage are identified, and process semantic relationships are extracted from the rule document data. The mapping relationship between the operation nodes, the interface state change data, and the triggering conditions in the process semantic relationships is determined, generating a structured process representation, including: Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data, and align the interface element location features with the state transition sequence in the interface state change data in time sequence to determine the interface state differences before and after each operation node is triggered; Semantic parsing is performed on the rule document data to identify the logical relationship between the described conditional expressions and the execution actions, and key entities and their attribute constraints are extracted as semantic representations of the triggering conditions; By calculating the semantic similarity between the operation type features of the operation node and the semantic representation of the triggering condition, a first mapping association between the operation node and the triggering condition is established. By analyzing the degree of matching between the interface state differences and the entity attribute constraints in the triggering conditions, a second mapping association between interface state change data and triggering conditions is established; The first mapping association and the second mapping association are merged to form a ternary mapping relationship between operation nodes, interface state change data and trigger conditions, and a structured process representation containing node identifiers, state transition rules and trigger condition constraints is generated based on the ternary mapping relationship.

3. The method according to claim 2, characterized in that, Extract the interface element location features and operation type features corresponding to each operation node in the operation behavior sequence data, and align the interface element location features with the state transition sequence in the interface state change data in time sequence to determine the interface state differences before and after each operation node is triggered, including: The operation behavior sequence data is parsed to extract the recorded interface element identification information and interaction action type information. The interface element identification information is converted into interface element positioning features containing position coordinates and hierarchical relationships, and the interaction action type information is classified into operation type features. The interface state change data is obtained by retrieving a sequence of interface snapshots recorded in chronological order. By comparing the structural and content changes between adjacent interface snapshots, a state transition sequence describing the interface evolution process is generated. Based on the execution timestamps of each operation node in the operation behavior sequence data, the state transition node corresponding to the time of each operation node is located in the state transition sequence, and a temporal correspondence between the operation node and the state transition node is established; For each operation node, extract the preceding and following interface snapshots of the state transition node corresponding to its time sequence. Based on the difference between the interface element positioning features in the preceding and following interface snapshots, determine the interface state difference before and after the operation node is triggered.

4. The method according to claim 1, characterized in that, Collect current process status parameters, match the current process status parameters with the trigger conditions in the structured process representation, and determine the target execution branch based on the matching result, including: By monitoring the completion status of the currently executing operation node, the element attribute values ​​of the current interface, and the field content of the current business data, the current process status parameters that characterize the current process status are collected. Extract subsequent candidate branches associated with the current operation node and their corresponding triggering conditions from the structured process representation, wherein each triggering condition consists of a condition expression and a constraint range; The interface element attribute values ​​in the process state parameters are compared with the judgment criteria of the condition expression in the trigger conditions to calculate the degree of conformity between the trigger conditions of each candidate branch and the current state. Candidate branches that meet the trigger condition constraints are selected based on the degree of compliance. When there are multiple candidate branches that meet the conditions, a unique target execution branch is determined based on the branch priority recorded in the structured process representation.

5. The method according to claim 1, characterized in that, Based on the structured process representation and the target execution branch, the interaction entry points of each system in the cross-system process are determined. System switching and data transfer are performed through simulation operation technology, and cross-system execution records are generated, including: Based on the sequence of operation nodes corresponding to the target execution branch in the structured process representation, the system identifier and interface element positioning features associated with each operation node are extracted, and the system switching node is identified by analyzing the changes in the system identifiers of adjacent operation nodes; For each system switching node, the interface state change data corresponding to that node is obtained from the mapping relationship in the structured process representation, and the interface element features that identify the system interaction entry point are extracted as the access entry point of the target system; During system switching, the interface element corresponding to the access point is located by simulating user interaction behavior, and an activation operation is performed to complete the switch from the source system to the target system; After the system switch is completed, the business data fields that need to be transferred from the source system to the target system are determined based on the data dependency relationships recorded in the structured process representation. The business data fields are filled into the corresponding interface elements of the target system by simulating data entry operations, and the cross-system operation trajectory of each system switch is recorded as the cross-system execution record.

6. The method according to claim 1, characterized in that, Based on the execution performance metrics and the anomaly characteristics, the execution order and operation parameters of the operation node sequence are optimized using a reinforcement learning algorithm, generating an operation order adjustment strategy and an operation parameter optimization strategy, including: The reinforcement learning state representation is determined based on the execution performance metrics and the anomaly features; Based on the reinforcement learning state representation, adjusting the execution position of the operation node in the operation node sequence is taken as the first action type, and modifying the interface element positioning method and operation execution interval of the operation node is taken as the second action type, thus establishing an action space; Execute the first action type in the action space, swap the order of operation nodes with data dependencies in the operation node sequence, calculate the total execution time of the process after the swap, and use the reduction in execution time as the first reward feedback; Update the reinforcement learning strategy parameters based on the first reward feedback, select the node order adjustment scheme that minimizes the total execution time of the process, and generate the operation order adjustment strategy; Execute the second action type in the action space, adjust the recognition rules of the corresponding interface elements and the waiting time for the operation trigger for the operation node corresponding to the abnormal feature, calculate the adjusted operation retry frequency, and use the reduction in retry frequency as the second reward feedback; The reinforcement learning strategy parameters are updated based on the second reward feedback. The parameter configuration scheme that minimizes the frequency of operation retries is selected, and an operation parameter optimization strategy is generated.

7. An adaptive simulation engine-based business process learning and automated execution system, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to acquire operation behavior sequence data, interface state change data, and rule document data during the execution of the business process; The second unit is used to divide the operation behavior sequence data into stages, identify operation nodes in each process stage, extract process semantic relationships from the rule document data, determine the mapping relationship between the operation nodes, the interface state change data and the triggering conditions in the process semantic relationships, and generate a structured process representation; The third unit is used to collect the current process status parameters, match the current process status parameters with the triggering conditions in the structured process representation, and determine the target execution branch based on the matching results; The fourth unit is used to determine the interaction entry points of each system in the cross-system process based on the structured process representation and the target execution branch, and to perform system switching and data transfer through simulation operation technology to generate cross-system execution records; The fifth unit is used to extract the execution performance indicators and abnormal features of each operation node based on the cross-system execution records, and to generate operation sequence adjustment strategies and operation parameter optimization strategies based on the execution performance indicators and abnormal features. The strategies are used to update the structured process representation.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images