Process constraint information-based operation process consistency detection method and device, electronic equipment and storage medium

Abstract

The application provides an operation process consistency detection method, device and system based on process constraint information, electronic equipment and storage medium. The method comprises the following steps: obtaining standardized operation process data, performing semantic analysis on the standardized operation process data to obtain process semantic analysis information; constructing process constraint information comprising operation nodes and node constraint relationships based on the process semantic analysis information; obtaining to-be-detected operation process data, performing semantic analysis on the to-be-detected operation process data to generate process feature representation; performing constraint compliance analysis on the to-be-detected operation process based on the association relationship between the process constraint information and the process feature representation to obtain an operation process consistency detection result. The application converts process execution logic into structured constraint information and performs consistency analysis based on node constraint relationships, thereby improving detection accuracy in a complex operation process scenario.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and data processing technology, and in particular to a method, apparatus, electronic device and computer-readable storage medium for detecting the consistency of operation processes based on process constraint information. Background Technology

[0002] With the development of information technology, data centers, industrial operation and maintenance systems, and other critical infrastructure systems typically need to perform operations such as equipment maintenance, fault handling, and emergency response according to preset operating procedures. These operating procedures are usually described in document form, including the operation objects, operation behaviors, execution conditions, and the relationships between different operation steps. To ensure that the operation process conforms to preset specifications, consistency checks are required on the operating procedure content to determine whether the actual process meets the corresponding requirements.

[0003] In existing technologies, consistency checks on operational processes typically employ manual review, rule-based template review, and generative model-based auxiliary analysis. Manual review relies on domain experts to examine process documents based on their experience; rule-based template review uses preset keywords, formatting rules, or fixed conditions to match and analyze process content; and generative model-based analysis utilizes the model's semantic understanding capabilities to analyze the process text and generate corresponding review results.

[0004] However, the above methods have at least the following problems: First, operational processes are usually described using natural language, containing a large number of implicit temporal relationships, dependencies, and execution conditions. Existing text-matching-based rule-based review methods struggle to accurately represent and analyze these logical relationships. Second, generative models primarily rely on their own knowledge for content generation, lacking structured expressions and verification mechanisms for domain-specific process constraints, which can easily lead to inconsistencies between analysis results and actual process specifications. Therefore, existing technologies lack a method that can convert operational processes into a structured representation containing logical constraints and perform consistency checks based on this structured representation. Summary of the Invention

[0005] This application aims to provide a method and apparatus for detecting the consistency of operational processes based on process constraint information, in order to solve the problem that in the prior art, operational processes are usually in the form of natural language, which makes it difficult for computers to directly understand the operational relationships and execution conditions in the process, resulting in the process consistency detection relying on human experience and making it difficult to achieve automated analysis.

[0006] To achieve the above objectives, this application provides a method for detecting operational process consistency based on process constraint information, comprising: Obtain standardized operation process data, and perform semantic parsing on the standardized operation process data to obtain process semantic parsing information. The process semantic parsing information includes operation object characteristics, operation behavior characteristics, execution condition characteristics, and process association characteristics. Based on the process semantic parsing information, process constraint information is constructed. The process constraint information includes operation nodes for representing process execution units, and node constraint relationships for representing the execution logic relationships between the operation nodes. Obtain the operation process data to be detected, perform semantic parsing on the operation process data to be detected, and generate a process feature representation, which is used to represent the process nodes and node relationships in the operation process to be detected. Based on the correlation between the process constraint information and the process feature representation, constraint compliance analysis is performed on the process node relationships in the operation process to be detected, so as to obtain the operation process consistency detection result.

[0007] Optionally, the process semantic parsing information includes entity information, behavior information, condition information, and association information; In the process of semantic parsing the standardized operation process data, the corresponding operation object entity is determined according to the process description content, the corresponding operation behavior is determined according to the action description content, and the execution conditions corresponding to the operation behavior and the association between different process elements are determined according to the context semantics.

[0008] Optionally, process constraint information is constructed based on the process semantic parsing information, including: An operation node is generated based on the characteristics of the operation object, the characteristics of the operation behavior, and the characteristics of the execution conditions. The operation node includes node identification information, operation object information, operation behavior information, execution condition information, and state change information. The state change information is used to indicate the process state change generated after the corresponding operation behavior is executed.

[0009] Optionally, the node constraint relationships include temporal constraints, dependency constraints, and conditional constraints; Specifically, timing constraints are generated based on the execution order relationship between multiple operation nodes, dependency constraints are generated based on the state transmission relationship, data association relationship or functional association relationship between multiple operation nodes, and condition constraints are generated based on the state requirements, environmental requirements or business constraints corresponding to the operation nodes.

[0010] Optionally, the method further includes: Candidate implicit constraint relationships are generated based on the relationships between operation objects, state changes, and execution conditions among multiple operation nodes. The validity of the candidate implicit constraint relationships is determined, and the candidate implicit constraint relationships that meet the preset judgment conditions are added to the process constraint information to update the node constraint relationships.

[0011] Optionally, semantic parsing is performed on the operation flow data to be detected to generate a flow feature representation, including: The data of the operation process to be detected is divided into process description units, and the operation object information, operation behavior information and execution condition information corresponding to each process description unit are extracted to generate the corresponding process node representation.

[0012] Optionally, generating the process feature representation includes: The operation object information, operation behavior information, and execution condition information in the process node representation are semantically normalized. The relationships between nodes are determined based on the normalized process node representation, so as to generate a process feature representation that includes process node information and node relationships.

[0013] Optionally, it also includes: Generate process evolution features based on the operational behavior change relationships and state change relationships in the operation process data to be detected; The process evolution features are associated with corresponding process nodes to form a process feature representation that includes information on dynamic changes in the process.

[0014] Optionally, based on the correlation between the process constraint information and the process feature representation, the following includes: Based on the semantic correspondence between the process node information in the process feature representation and the operation node in the process constraint information, a node association mapping relationship is established. Based on the node association mapping relationship, the node association relationship in the process feature representation is mapped to the corresponding node constraint relationship.

[0015] Optionally, constraint compliance analysis is performed on the process node relationships, including: Based on the node association mapping relationship, the execution order, node dependency relationship and execution conditions of the process nodes in the operation process to be detected are detected respectively; Based on the timing constraints, dependency constraints, and condition constraints, determine whether the operation process to be detected satisfies the process constraint information.

[0016] Optionally, obtaining the consistency detection result of the operation process includes: Identify the target process node in the operation process to be detected that does not satisfy the process constraint information; The process deviation type is determined based on the node constraint relationship corresponding to the target process node, and a consistency detection result including the target process node, deviation type, and corresponding violated constraint relationship is generated.

[0017] Optionally, it also includes: Generative semantic models are used to perform semantic parsing on the standardized operation process data or the operation process data to be detected in order to obtain process analysis results. Specifically, when performing semantic parsing using the generative semantic model, the process constraint information is input into the generative semantic model as constraint enhancement information.

[0018] Optionally, it also includes: The process analysis results output by the generative semantic model are converted into process analysis feature representations. Based on the operation nodes and node constraint relationships in the process constraint information, constraint verification is performed on the process analysis feature characterization to determine whether the process analysis results meet the process logic requirements.

[0019] Optionally, it also includes: Based on the constraint verification results, determine the non-compliant constraint items; Based on the non-compliance constraint, a result correction instruction is generated, and the process analysis result output by the generative semantic model is adjusted based on the result correction instruction to obtain the target analysis result that satisfies the process constraint information.

[0020] This application also provides an operation process consistency detection device based on process constraint information, including: The process data parsing module is used to acquire standardized operation process data and perform semantic parsing on the standardized operation process data to obtain process semantic parsing information. The process semantic parsing information includes operation object characteristics, operation behavior characteristics, execution condition characteristics, and process association characteristics. A process constraint construction module is used to construct process constraint information based on the process semantic parsing information. The process constraint information includes operation nodes for representing process execution units and node constraint relationships for representing the execution logic relationships between the operation nodes. The process feature representation module is used to acquire the operation process data to be detected, perform semantic parsing on the operation process data to be detected, and generate a process feature representation, which is used to represent the process nodes and node relationships in the operation process to be detected. The consistency detection module is used to perform constraint compliance analysis on the process node relationships in the operation process to be detected based on the correlation between the process constraint information and the process feature representation, so as to obtain the operation process consistency detection result.

[0021] This application also provides an electronic device, the electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the above-described operation flow consistency detection method based on process constraint information.

[0022] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described operation flow consistency detection method based on process constraint information.

[0023] Compared with the prior art, this application has at least the following beneficial effects: First, this application performs semantic parsing on standardized operating procedure data and converts operating behaviors, execution conditions, and relationships into process constraint information, enabling process knowledge that originally existed in the form of natural language to form a structured representation that can be processed by a computer, thus providing a data foundation for subsequent automated consistency detection.

[0024] Second, by constructing process feature representations and establishing a correlation mapping relationship between process feature representations and process constraint information, this application can analyze the operation process to be detected from multiple dimensions such as process elements, execution relationships and execution conditions. Compared with the traditional text matching-based auditing method, this improves the accuracy of complex process logic detection.

[0025] Third, this application utilizes process constraint information to perform consistency verification and correction on the output results of the generative semantic model, thereby controlling the semantic understanding ability of the generative model under domain process constraints, reducing the risk of the generated results deviating from the actual process specifications, and improving the reliability of automated review results. Attached Figure Description

[0026] Figure 1 A flowchart illustrating an operation process consistency detection method based on process constraint information provided in this application embodiment; Figure 2 A schematic diagram illustrating a process constraint information construction process provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the generation process of a feature representation of an operation flow to be detected, provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of a process for performing consistency checks based on process constraint information, as an embodiment of the present application. Figure 5 A schematic diagram of an operation process consistency detection device based on process constraint information provided in this application embodiment; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0027] The technical solution of this application will be described in detail below with reference to the accompanying drawings. It should be noted that the embodiments described in this application are only for illustrating the technical solution of this application and are not intended to limit the scope of protection of this application. Other embodiments obtained by those skilled in the art based on the technical content disclosed in this application without creative effort are all within the scope of protection of this application.

[0028] This application provides a method for detecting operational process consistency based on process constraint information, used to analyze the consistency between a standard operational process and an operational process to be tested. The method acquires standardized operational process data and performs semantic parsing on the data to obtain semantic parsing information representing process execution elements and process relationships. Based on this semantic parsing information, process constraint information is constructed, including operational nodes representing process execution units and node constraint relationships representing the execution logic relationships between these nodes. Further, semantic parsing is performed on the operational process data to be tested to generate process feature representations representing the structure of the process to be tested. Constraint compliance analysis is then performed based on the relationship between the process constraint information and the process feature representations to obtain the operational process consistency detection result. This approach transforms traditional process comparison methods based on text content or step lists into a structured analysis method based on process execution logic, improving consistency detection capabilities in complex operational process scenarios.

[0029] In practical applications, operational procedures are typically used to describe business operations with fixed execution specifications, such as data center operation and maintenance procedures, industrial equipment control procedures, energy system scheduling procedures, and emergency response procedures. These operational procedures are usually developed by domain personnel based on equipment operation requirements, safety management standards, and business execution rules, and are stored in the form of operating procedures, maintenance documents, execution plans, or historical operation records. Since the actual execution of a process involves not only multiple operational steps but also the execution order, status transmission, and conditional constraints between these steps, process consistency requires more than just determining whether two processes contain the same operational content; it also requires determining whether the operational nodes in the two processes are organized according to the same execution logic. For example, in a data center power supply switching process, although backup power supply status confirmation, power path switching, and load status checking are different operational steps, power path switching must be based on the completion of backup power supply status confirmation, and load status checking depends on the operating status after power path switching. Therefore, there are interrelated execution constraints between these operational steps.

[0030] To address the difficulty in identifying logical relationships within processes during existing consistency analysis, this application first performs semantic parsing on standardized operational process data. This transforms the process content, described in natural language, into semantic parsing information containing operational objects, operational behaviors, execution conditions, and process relationships. Furthermore, process constraint information is constructed based on this semantic parsing information. This process constraint information does not merely represent a set of process steps, but rather describes the execution logic within the standard process in the form of operational nodes and node constraint relationships. This enables the computer to determine whether the process under test meets standard requirements based on structured process relationships.

[0031] like Figure 1 As shown, this application provides a flowchart illustrating an operation process consistency detection method based on process constraint information. The method includes: acquiring standardized operation process data; performing semantic parsing on the standardized operation process data to obtain process semantic parsing information; constructing process constraint information based on the process semantic parsing information, the process constraint information including operation nodes and node constraint relationships; acquiring operation process data to be detected; performing semantic parsing on the operation process data to generate process feature representations; and performing constraint compliance analysis on the process node relationships in the operation process to be detected based on the correlation between the process constraint information and the process feature representations to obtain operation process consistency detection results.

[0032] Preferably, after obtaining standardized operation process data, preprocessing operations can be performed based on the data organization format, text structure information, and process description content of the standardized operation process data to determine multiple process description units. Each process description unit represents a basic execution process in the standardized operation process and can correspond to an independent operation step, a status confirmation process, or a process stage with a clear execution goal. For complex process descriptions that simultaneously contain multiple continuous operation behaviors, semantic decomposition can be performed based on changes in the operation object, changes in operation behavior, and status changes, so that different operation behaviors correspond to independent process description units, thereby avoiding the merging of multiple execution actions and the resulting loss of subsequent process logic analysis information. For example, the process description of "disconnecting the main power supply circuit and switching to the backup power supply mode after confirming the normal operation status of the backup power supply equipment" can be decomposed into multiple process description units such as "confirming the operating status of the backup power supply equipment," "disconnecting the main power supply circuit," and "switching to the backup power supply mode."

[0033] Preferably, after obtaining the process description unit, semantic parsing processing can be performed on the process description unit to obtain process semantic parsing information. The process semantic parsing information includes operation object features, operation behavior features, execution condition features, and process association features. The operation object features represent the target entity affected by the process operation; the operation behavior features represent the specific actions performed on the target entity; the execution condition features represent the state requirements that need to be met when performing the corresponding operation behavior; and the process association features represent the logical relationships between different process description units. By extracting the above process elements, the original process text can be converted into structured semantic information that can be used for subsequent process constraint construction.

[0034] Preferably, when extracting the features of the operation object, the corresponding operation object can be determined based on the entity description information in the process description unit, and a unified object representation can be generated by combining the object category, functional attributes, and contextual semantic information to reduce object recognition errors caused by name differences between different process documents. For example, in a data center operation and maintenance scenario, different textual expressions such as "backup power supply equipment," "backup power supply module," and "backup power supply unit" may correspond to the same functional entity. By combining equipment attributes and process context relationships to determine the correspondence between objects, the operation objects in process documents from different sources can be placed in a unified semantic space.

[0035] Preferably, when extracting operational behavior features, the corresponding operational behavior can be determined based on the action description information in the process description unit, and the actual technical meaning of the operational behavior can be determined by combining the operation object information and the process context, so as to avoid misjudgment of the operation meaning caused by relying solely on keyword matching. For example, different descriptions such as "switch to standby mode", "enter standby power supply state", and "execute standby power supply activation operation" all indicate a change in power supply operation state in the power supply switching process. By performing behavior normalization processing on different expression forms, a unified operational behavior representation can be generated.

[0036] Preferably, when extracting execution condition features, the execution conditions corresponding to the operation behavior can be determined based on the state description information, environmental constraint information, and business rule information in the process description unit. This ensures that the process semantic parsing information not only includes the execution action but also represents the state basis upon which the execution action depends. For example, when the process description is "Execute power switching operation after confirming that the backup power supply equipment voltage is stable," "backup power supply equipment voltage is stable" can be identified as the execution condition corresponding to the power switching operation, and this execution condition can be associated with the corresponding operation behavior, providing a data foundation for subsequently constructing dependency constraints and condition constraints between nodes.

[0037] Preferably, such as Figure 2 As shown, this application provides a schematic diagram of the process for constructing process constraint information. After obtaining process semantic parsing information, process constraint information can be constructed based on the process semantic parsing information. The process constraint information is used to represent the execution logic in a standardized operation process in the form of operation nodes and node constraint relationships. Operation nodes represent the basic execution units in the process execution, and node constraint relationships represent the execution logic relationships between different operation nodes. Compared to simply using a list of process steps to represent a standard process, this application introduces node constraint relationships, enabling the process constraint information to further describe the execution order, state associations, and conditional constraints between operation nodes, thereby forming a more complete representation of the process logic.

[0038] Preferably, when generating operation nodes, a corresponding process execution unit can be determined based on the characteristics of the operation object, the characteristics of the operation behavior, and the characteristics of the execution conditions, and an operation node corresponding to the process execution unit can be established. The operation node includes node identification information, operation object information, operation behavior information, execution condition information, and state change information. The node identification information uniquely identifies the corresponding operation node; the operation object information represents the target entity of the current operation; the operation behavior information represents the action performed on the target entity; the execution condition information represents the constraints that must be met before the operation behavior is executed; and the state change information represents the changes in the process state before and after the corresponding operation behavior is executed. By simultaneously storing the above information in the operation node, the operation node can describe not only the execution action itself but also the basis and result of the action execution, providing a complete data foundation for subsequently generating node constraint relationships.

[0039] For example, in a data center power supply switching process, standby power supply status detection nodes, power supply path switching nodes, and load operation check nodes can be generated based on the process semantic parsing results. The standby power supply status detection node operates on standby power supply equipment, its operation is status detection, the execution condition is meeting the detection requirements, and the status change information is generating a standby power supply status confirmation result. The power supply path switching node operates on the power supply path, its operation is power supply mode switching, the execution condition is that the standby power supply status meets the requirements, and the status change information is that the power supply status has changed from primary power supply mode to standby power supply mode. The load operation check node operates on the server load, its operation is operation status detection, the execution condition is that the power supply mode switching is complete, and the status change information is generating a load recovery confirmation result.

[0040] Preferably, after generating operation nodes, node constraint relationships can be further constructed based on the execution logic relationships between multiple operation nodes to form complete process constraint information. The node constraint relationships describe the association between different operation nodes, including the execution order relationship, state transmission relationship, and execution condition relationship. By constructing node constraint relationships, the process constraint information can not only record which operation nodes are included in the process, but also represent how each operation node coordinates its execution, thereby providing a constraint basis for subsequent consistency checks based on process logic.

[0041] Preferably, timing constraints can be generated based on the execution sequence relationship between multiple operation nodes. These timing constraints represent the sequential relationship between different operation nodes during process execution, limiting whether a certain operation node needs to be executed before or after another operation node. For example, in a data center power supply switching process, the backup power supply status detection node needs to be executed before the power supply path switching node, and the power supply path switching node needs to be executed before the load operation check node. Therefore, corresponding timing constraints can be generated based on the above execution sequence relationship. By introducing timing constraints, issues with step sequence adjustments in the process under test can be identified, avoiding misjudgments of process consistency simply because the process contains identical operation steps.

[0042] Preferably, dependency constraints can be generated based on the state transitivity, data association, or functional association between multiple operation nodes. These dependency constraints indicate whether the execution of a certain operation node depends on the state or processing results generated by other operation nodes. When the state change information generated after the execution of one operation node serves as the basis for the execution of another operation node, a corresponding dependency relationship can be established between the two operation nodes. For example, after the backup power supply status detection node executes, it generates the state change information "backup power supply status normal." This state change information can serve as a necessary condition for the execution of the power supply path switching node; therefore, a dependency constraint can be established between the backup power supply status detection node and the power supply path switching node. By introducing dependency constraints, problems such as missing necessary preliminary operations or violations of state transitivity in the process can be identified.

[0043] Preferably, condition constraints can be generated based on the execution condition information corresponding to the operation node. These condition constraints represent the state requirements, environmental requirements, or business restrictions that the operation node needs to meet during execution, enabling the process constraint information to describe the necessary conditions for the operation to be performed. For example, a device switching operation may require the device to be in normal operating condition, the communication link to be available, or the target resources to meet preset requirements; corresponding condition constraints can then be generated based on the aforementioned execution condition information. When the operation process to be detected contains a corresponding operation node but lacks the necessary execution conditions, logical defects in the process can be identified through condition constraints.

[0044] Preferably, by simultaneously constructing timing constraints, dependency constraints, and conditional constraints, the execution logic in the standard operating procedure can be described from multiple dimensions. Timing constraints describe the temporal execution relationship between operating nodes, dependency constraints describe the state or data association relationship between operating nodes, and conditional constraints describe the restrictions that operating nodes must meet to execute. Compared to representing the operating procedure solely through a list of process nodes, the process constraint information constructed in this application can more completely express the dynamic logical relationship during process execution, enabling subsequent consistency detection processes to analyze the process execution rules.

[0045] Preferably, in some embodiments, the node constraint relationships may further include implicit constraint relationships derived from process semantic information. In actual process documentation, some execution logic is not directly described in explicit statements, but is implicit in the object associations, state changes, and execution condition associations between multiple operation nodes. Therefore, candidate implicit constraint relationships can be generated based on the generated operation node information and the association characteristics between nodes, and the validity of the candidate implicit constraint relationships can be judged to improve the process constraint information.

[0046] Preferably, when generating candidate implicit constraint relationships, potential logical connections can be determined based on the relationships between operation objects, state changes, and execution conditions among multiple operation nodes. For example, when there is a correspondence between the state change information generated by one operation node and the execution condition information of another operation node, it can be determined that there may be a dependency relationship between the two operation nodes, and a corresponding candidate implicit constraint relationship can be generated. As another example, when multiple operation nodes perform continuous state adjustment operations on the same device object, it can be determined that there may be temporal relationships between the multiple operation nodes based on the operation object and the state change process.

[0047] Preferably, after obtaining the candidate implicit constraint relationships, the validity of the candidate implicit constraint relationships can be judged according to preset judgment conditions. Specifically, it can be determined whether the candidate implicit constraint relationship meets the inclusion conditions based on information such as the consistency of the operation objects corresponding to the candidate implicit constraint relationship, the continuity of state changes, the degree of matching of execution conditions, and the historical process execution status. When the candidate implicit constraint relationship meets the preset judgment conditions, it can be added to the process constraint information to expand the node constraint relationships.

[0048] For example, in data center equipment maintenance procedures, standard procedure text might only describe "execute equipment switching operation after completing equipment status confirmation," without explicitly stating that equipment status confirmation is a necessary prerequisite for equipment switching operation. By analyzing the status change information generated by the equipment status confirmation node and the execution condition information corresponding to the equipment switching node, a state dependency relationship can be identified between the two, and corresponding dependency constraints can be generated. By supplementing the above implicit constraint relationships, the procedure constraint information can be made closer to the actual execution logic.

[0049] Preferably, through the above-described process constraint information construction process, the explicit process descriptions and implicit execution logic in standardized operation process documents can be uniformly converted into a structured representation. Specifically, operation nodes carry the semantic information corresponding to the process execution units, and node constraint relationships carry the logical relationships between process execution units. This allows the generated process constraint information to serve as a standard constraint object in the subsequent consistency analysis of the operation process to be tested, improving the accuracy of abnormal process identification in complex process scenarios.

[0050] Preferably, after constructing the process constraint information, the operation process data to be detected can be further obtained, and semantic parsing can be performed on the operation process data to generate process feature representations. The operation process data to be detected represents the operation process information generated during actual execution, and can originate from real-time generated operation plans, historical execution records, manually input maintenance processes, or process instance data in the business system. By performing structured parsing on the operation process data to be detected, actual operation processes from different sources and with different expressions can be converted into a process representation form that matches the process constraint information, thereby providing a unified data foundation for subsequent consistency detection.

[0051] Preferably, when performing semantic parsing on the operation flow data to be tested, the process can first be divided into flow description units based on the flow description content in the data. Then, for each flow description unit, the corresponding operation object information, operation behavior information, and execution condition information are extracted to generate a corresponding flow node representation. This flow node representation describes the actual execution unit in the operation flow to be tested and corresponds to the operation nodes in the aforementioned flow constraint information. By performing structured parsing on the operation flow to be tested at the same granularity as the standard flow, node matching errors caused by different flow expression methods can be reduced.

[0052] For example, in a data center power supply switching scenario, the operation process to be detected might be described as "after confirming the backup power supply status, perform a power supply switch and check the server's operating status." Through process parsing, the backup power supply status confirmation node, the power supply switching node, and the server operation check node can be obtained, and the corresponding operation object, operation behavior, and execution condition information for each node can be extracted. Although the above process node representation originates from the actual process to be executed, it uses a data structure corresponding to the standard process constraint information, thus enabling it to participate in subsequent constraint compliance analysis.

[0053] Preferably, when generating the process feature representation, the operation object information, operation behavior information, and execution condition information in the process node representation can be semantically normalized, and a process feature representation containing node information and node relationships can be generated based on the association relationships between different process nodes. The process feature representation is used to represent the process structure information in the operation process to be detected, so that the operation process to be detected can be represented in a way that corresponds to the process constraint information.

[0054] The operation object information in the process feature representation is used to represent the target entity affected by the operation process to be detected; the operation behavior information is used to represent the actual action performed on the target entity; the execution condition information is used to represent the state or environmental requirements on which the corresponding action depends; and the node association relationship is used to represent the execution order, state association, and condition association relationship between different process nodes. By unifying the representation of the above information, errors in process judgment due to differences in description can be avoided when comparing only the process text content.

[0055] Preferably, during the generation of process feature representation, process evolution features can also be generated based on the operational behavior change relationships and state change relationships in the operation process to be detected, and the process evolution features are associated with corresponding process nodes to form a process feature representation containing dynamic change information. The process evolution features are used to describe the evolution process of the process state as the operational behavior changes during the execution of the operation process, so that the process feature representation can not only represent static process nodes, but also the state transition relationships during the process execution.

[0056] For example, in the power supply switching process, before the backup power supply status detection node is executed, the system is in the main power supply state; after the backup power supply status is confirmed, the system obtains the backup power supply availability state; after the power supply path switching operation is executed, the system enters the backup power supply state. The above different stages correspond to different process states. By extracting the state change process, a process evolution feature representing "main power supply state → backup power supply preparation state → backup power supply state" can be formed.

[0057] Preferably, by introducing process evolution features, the process feature representation can not only describe which operation steps the operation process to be tested includes, but also describe the state change relationships generated during the execution of each operation step. Compared with the method of analyzing the process based solely on the list of operation steps, the process feature representation generated in this application can more completely reflect the execution logic of the operation process to be tested, enabling the subsequent consistency detection process to identify problems where process steps exist but the state transition process does not meet the requirements.

[0058] Preferably, after generating the process feature representation, the association between the process feature representation and the process constraint information can be further established to determine the correspondence between the process nodes in the operation process to be detected and the operation nodes in the standard process. Specifically, semantic matching and relationship mapping can be performed on the process nodes to be detected and the operation nodes in the process constraint information based on the operation object characteristics, operation behavior characteristics, execution condition characteristics, and node association relationships, so that the operation process to be detected can be mapped to the constraint space corresponding to the standard process.

[0059] Preferably, when performing process node association mapping, semantic differences arising from different process expression methods can be matched. For example, an operation node in a standard process may be described as "execute backup power supply operation", while the corresponding description in the process to be tested may be "switch to backup operation mode". Although the text content is different, by analyzing the operation object, operation behavior and state change information, it can be determined that they represent the same process execution unit, and a corresponding node mapping relationship can be established.

[0060] Preferably, by generating the above-mentioned process feature representation, the actual operation process to be detected can be converted into a structured representation containing operation nodes, node relationships, and process evolution features, enabling the operation process to be detected to be compared with process constraint information in a unified semantic space. Compared with directly comparing the process text content, this application can perform analysis based on process structure and execution logic, improving the adaptability to complex operation processes, processes with different expression forms, and dynamically changing processes.

[0061] Preferably, after obtaining the process feature representation, process node mapping can be performed based on the association between the process feature representation and process constraint information to determine the correspondence between process nodes in the operation process to be detected and operation nodes in the standard operation process. Specifically, semantic matching can be performed on process nodes in the process feature representation and operation nodes in the process constraint information based on operation object information, operation behavior information, execution condition information, and node association relationships, and a node association mapping relationship can be established. By establishing a node association mapping relationship, the operation process to be detected can be transformed into the standard process logic space defined by the process constraint information, enabling the subsequent detection process to analyze the execution relationship between nodes, rather than directly comparing the surface differences between different process texts.

[0062] Preferably, when establishing node association mapping relationships, semantic matching processing can be performed to address the expression differences between the operation process to be detected and the standardized operation process. When there are textual differences between the operation description in the process to be detected and the operation description in the standard process, but the corresponding operation objects, operation behaviors, and state change relationships are consistent, it can be determined that they belong to the same process execution unit, and a corresponding node mapping relationship can be established. For example, the operation node description in the standard process is "execute backup power supply activation operation," while the corresponding description in the process to be detected is "switch to backup operation mode." Although they use different textual expressions, both indicate that the power supply mode has changed from the main power supply state to the backup power supply state. Therefore, a corresponding relationship can be established based on the operation object and state change information.

[0063] Preferably, after completing the process node association mapping, the node association relationships in the process feature representation can be mapped to the node constraint relationships in the process constraint information based on the node association mapping relationship, and then constraint compliance analysis can be performed on the process node relationships in the operation process to be tested. The constraint compliance analysis is used to determine whether the execution logic in the operation process to be tested meets the timing constraints, dependency constraints, and condition constraints required by the standard process, thereby determining whether there are process logic deviations in the operation process to be tested.

[0064] Preferably, during timing constraint detection, the execution order of nodes in the process feature representation can be compared with the timing constraint relationship in the process constraint information to determine whether the arrangement order of process nodes in the process to be tested conforms to the standard process requirements. When a change in the order of key operation nodes in the process to be tested is detected, an execution order anomaly can be determined based on the corresponding timing constraints. For example, the standard power supply switching process requires the backup power supply status detection to be performed first, followed by the power supply path switching operation. However, if the process to be tested directly performs the power supply path switching operation, then the timing constraints can be used to determine that the process to be tested has a non-compliant process order.

[0065] Preferably, when performing dependency constraint detection, the necessary preconditions for each operation node in the process to be tested can be determined based on the state change information corresponding to the process nodes and the data association between nodes. When the state condition required by a certain operation node is not generated by the preceding node, or the state result generated by the preceding node is not transmitted to the corresponding operation node, it can be determined that there is a missing dependency relationship in the current process based on the dependency constraints. For example, if the power supply path switching node depends on the "backup power supply status normal" result generated by the backup power supply status detection node, and if the process to be tested does not include the backup power supply status detection process, it can be determined based on the dependency constraints that the power supply switching operation lacks necessary preconditions.

[0066] Preferably, during the execution constraint detection, the execution condition information in the process feature representation can be matched with the condition constraints in the process constraint information to determine whether the operation node in the process to be detected meets the execution restriction requirements. The execution restriction requirements may include device status requirements, environmental status requirements, permission requirements, or business rule requirements. For example, a standard process requires that the device be in a stable operating state before a switching operation can be performed, but the process to be detected does not include a device status confirmation process; therefore, the operation node can be determined to not meet the execution requirements based on the condition constraints.

[0067] Preferably, by simultaneously performing timing constraint detection, dependency constraint detection, and condition constraint detection, the logical consistency between the operation process under test and the standard operation process can be analyzed from multiple dimensions. Timing constraints are used to determine whether the execution order of the process is correct, dependency constraints are used to determine whether the state transmission relationship of the process is complete, and condition constraints are used to determine whether the execution basis of the operation behavior meets the requirements. Compared to methods that only determine the existence of process steps, this application can identify various types of process anomalies, such as missing process steps, incorrect execution order, insufficient prerequisite dependencies, and unmet execution conditions.

[0068] Preferably, after completing the constraint compliance analysis, an operational process consistency detection result can be generated based on the analysis results. When it is determined that the operational process to be tested satisfies all node constraint relationships in the process constraint information, a detection result indicating process consistency can be generated; when a process node in the operational process to be tested is detected that does not satisfy the node constraint relationship, the target process node and the corresponding deviation type can be further determined, and a consistency detection result containing the target process node, deviation type, and corresponding violated constraint relationship can be generated.

[0069] Preferably, when generating deviation location information, the specific cause of the deviation can be determined based on the abnormal process node and its corresponding constraint relationship. For example, when it is detected that a power supply path switching node does not meet its dependency constraint, it can be determined that the node lacks the preceding status information provided by the backup power supply status confirmation node; when it is detected that a power supply path switching node does not meet its condition constraint, it can be determined that the node lacks the equipment status confirmation condition. By outputting the specific violated constraint relationship, the detection results can not only indicate whether the process is abnormal, but also locate the specific logical cause of the abnormality.

[0070] Through the aforementioned consistency detection process, this application can perform logical analysis of the operation process to be detected based on process constraint information, enabling computers to identify process anomalies that are difficult to detect using traditional text matching methods. For example, two processes may contain the same operation steps, but one process may have adjusted the order of key operations or lack a necessary status confirmation process. Traditional step matching methods may consider the two processes to be consistent, while this application can identify the above logical differences based on node constraint relationships, thereby improving the accuracy of consistency detection in complex process scenarios.

[0071] Preferably, in some embodiments, this application can also combine generative semantic models to perform semantic understanding and process analysis, thereby further improving the automated analysis capabilities in complex process data scenarios. The generative semantic model can be understood as a semantic processing model implemented based on computer programs, capable of performing semantic understanding, information association, logical reasoning, and result generation based on the input data content. Compared with traditional process parsing methods based on fixed rule matching or keyword retrieval, generative semantic models can understand the semantic relationships between different expression forms based on contextual information, thereby extracting information such as operation objects, operation behaviors, execution conditions, and process relationships from unstructured process data.

[0072] In practical applications, operational process data typically exists in the form of natural language text, operation records, maintenance logs, or business documents, containing a large number of domain terms, implicit logical relationships, and dependencies. For example, the same operational behavior may be expressed differently in different process data, and the same device status may be described differently depending on the application scenario. If traditional keyword matching-based methods are used for process parsing, it is difficult to identify the semantic equivalence between different expressions and to discover the execution logic implicit in the text. Therefore, this application utilizes a generative semantic model to perform semantic understanding of process data to obtain process analysis results that are closer to the actual business meaning.

[0073] Preferably, in this application, when the generative semantic model performs a process analysis task, it can first acquire the process data to be analyzed and perform a computer-processable data transformation on the process data. Specifically, the process text containing natural language descriptions can be converted into corresponding semantic representation data, enabling the computer to perform subsequent feature association calculations based on the semantic representation data. The semantic representation data can include entity information, action information, state information, and relationship information from the process description, used to represent the operation objects, execution behaviors, and logical associations involved in the process execution.

[0074] Preferably, after obtaining the semantic representation data, the generative semantic model can analyze the relationships between different semantic units based on the semantic association calculation process within the model to determine the possible corresponding process execution units in the process description. For example, for a process description containing "execute power switching operation after confirming that the backup power supply equipment is in normal condition", the generative semantic model can identify the equipment status confirmation behavior, power switching behavior, and the relationship between them based on the semantic association, and generate the corresponding process analysis results.

[0075] Preferably, in the process of generating process analysis results, the generative semantic model can predict candidate operation nodes, candidate node relationships, and candidate execution conditions related to the current process description based on the input content. Here, candidate operation nodes represent the process execution units determined by the model based on semantic understanding results; candidate node relationships represent the association methods between process nodes inferred by the model; and candidate execution conditions represent the operation behavior execution requirements identified or inferred by the model. Through the above processing, the generative semantic model can convert the original process text into process analysis results containing structured semantic information.

[0076] However, because generative semantic models typically generate outputs based on semantic relationships and probabilistic prediction mechanisms within existing data, their generation process is not entirely equivalent to information retrieval based on deterministic rules. In professional process analysis scenarios, when the input process data is incomplete, lacks domain rules, or has insufficient contextual information, generative semantic models may generate content inconsistent with the actual process based on existing general knowledge or statistical relationships, leading to uncertainty in the process analysis results. This uncertainty generated by generative semantic models can manifest in various forms in process consistency detection scenarios.

[0077] The first type is the illusion of process nodes, where the generative semantic model generates operation nodes that do not exist in the actual standard process. For example, in an equipment maintenance process, the standard process only specifies three steps: equipment status detection, parameter confirmation, and maintenance execution. However, the generative semantic model may add a step called equipment calibration, which is not actually specified, based on relevant domain knowledge. Since this step conforms to general domain knowledge but does not belong to the current standard process, it will cause the process analysis results to deviate from the actual business requirements. The second type is the illusion of process relationships, where the generative semantic model correctly identifies some process nodes but incorrectly establishes the relationships between nodes. For example, the standard process requires that equipment status detection be completed before the equipment switching operation can be performed. However, the generative semantic model may, due to incorrect semantic association inference, assume that the equipment switching operation can be performed independently, or incorrectly determine that there is a dependency relationship between two operation nodes that are not directly related. The above errors will cause subsequent process consistency checks to fail to accurately reflect the actual execution logic. The third type is the illusion of process conditions, where the generative semantic model misunderstands the process execution conditions. For example, a device switching operation actually requires multiple conditions to be met, such as the device operating normally, the communication link being available, and authorization verification being passed. However, the generative semantic model may omit some of these necessary conditions or generate non-existent constraints based on other similar processes. Since execution conditions directly affect whether the process can be executed safely, this type of error may lead to significant deviations in the process analysis results. The fourth type is the process evolution illusion, where the generative semantic model incorrectly infers state change relationships when analyzing the process execution. For example, in a continuous operation process, the state change generated by a certain operation node should serve as the basis for the execution of the next operation node. However, the generative semantic model may incorrectly determine the direction of the state change or skip intermediate state transitions, resulting in a mismatch between the generated process evolution relationship and the actual execution process.

[0078] Preferably, to address the potential erroneous generation results of the generative semantic model in professional process analysis scenarios, this application does not directly use the output results of the generative semantic model as the final basis for process analysis. Instead, it introduces the aforementioned constructed process constraint information to constrain the analysis process and output results of the generative semantic model. The process constraint information includes operation nodes and node constraint relationships. Operation nodes represent process execution units allowed in a standard process, and node constraint relationships represent the temporal relationships, dependencies, and conditional relationships between different process execution units.

[0079] Preferably, before the generative semantic model performs process analysis, the process constraint information can be converted into constraint feature representations that the model can process, and input into the generative semantic model along with the semantic representation data corresponding to the process data to be analyzed. Specifically, node constraint features can be generated based on the operation nodes in the process constraint information, relation constraint features can be generated based on the node constraint relationships, and the node constraint features and relation constraint features can be fused with the process semantic features, so that the generative semantic model can simultaneously refer to the process text information and standard process logic information when performing semantic reasoning.

[0080] Preferably, by introducing process constraint information during the model input stage, the scope of the generative semantic model's understanding of the process content can be limited. For example, when the generative semantic model identifies a candidate operation node based on the process text, it can combine the process constraint information to determine whether the candidate operation node belongs to a valid operation node in the standard process; when the generative semantic model infers a relationship between two process nodes, it can combine the node constraint relationship to determine whether the relationship satisfies the execution rules in the standard process. Through these methods, the possibility of the model generating incorrect process content solely based on linguistic associations can be reduced.

[0081] Preferably, during the semantic reasoning process of the generative semantic model, the process constraint information can not only serve as auxiliary information in the model input stage, but also participate in the model's judgment process on candidate process content. Specifically, when the generative semantic model generates candidate operation nodes, candidate node relationships, and candidate execution conditions based on the process semantic features, it can perform association matching and constraint compliance judgment on different candidate results based on the constraint feature representation corresponding to the process constraint information, so as to determine whether each candidate result meets the logical requirements corresponding to the standard process.

[0082] Preferably, during the candidate operation node generation process, the candidate operation nodes identified by the generative semantic model can be matched and analyzed based on the set of operation nodes in the process constraint information. When a candidate operation node can establish a correspondence with an existing operation node in the process constraint information, it can be determined that the candidate operation node belongs to a valid execution unit defined in the standard process; when a candidate operation node cannot establish a correspondence with any standard operation node, it can be further determined by combining the contextual semantic information whether it belongs to an equivalent node due to expression differences, or an abnormal node inferred by the generative semantic model based on general knowledge.

[0083] Preferably, when determining whether a candidate operation node is a valid execution unit, consistency of the operation object, consistency of the operation behavior, and consistency of the state change can be comprehensively considered. Specifically, consistency of the operation object is used to determine whether the target entity targeted by the candidate operation node matches the corresponding entity in the standard process; consistency of the operation behavior is used to determine whether the execution action described by the candidate operation node corresponds to the operation behavior in the standard process; and consistency of the state change is used to determine whether the state change generated after the execution of the candidate operation node conforms to the expected result in the standard process. This multi-dimensional matching method can avoid the misjudgment of normal process nodes as abnormal content due to differences in process text expression.

[0084] Preferably, during the candidate node relationship generation process, the node association relationships inferred by the generative semantic model can be restricted based on the node constraint relationships in the process constraint information. These node association relationships can include sequential execution relationships, state transit relationships, and conditional dependencies between nodes. When the generative semantic model generates an association relationship between two process nodes, it can determine whether this association relationship exists in the node constraint relationships corresponding to the standard process, or whether it can be derived from existing constraint relationships. If the corresponding constraint requirements cannot be met, it can be determined that the node association relationship is abnormal.

[0085] Preferably, for example, during the device switching process analysis, the generative semantic model identifies the "device status confirmation node" and the "device switching node" based on the text description, and further infers an execution dependency between them. If the process constraint information records that there is indeed a dependency relationship of "execute device switching after status confirmation" between the two, then this relationship can be retained; if the process constraint information indicates that there is no direct relationship between the two, or that the relationship needs to be established through other intermediate operation nodes, then the generated result can be marked as a relationship to be verified. In this way, the generative semantic model can avoid incorrectly constructing process execution logic based on semantic similarity.

[0086] Preferably, during the generation of candidate execution conditions, the model generation results can be verified based on the condition constraints in the process constraint information. Specifically, it can be determined whether the execution conditions identified by the generative semantic model include the necessary conditions required by the standard process, and whether there are any unnecessary conditions incorrectly supplemented by the model. When the model omits key execution conditions specified in the standard process, it can be determined that there is a condition missing problem; when the model generates restrictive conditions that do not exist in the actual process, it can be determined that there is a condition redundancy problem.

[0087] Preferably, by incorporating process constraint information into the generative semantic model reasoning process, a process semantic generation method constrained by business logic can be formed. Compared with the traditional generative semantic model that generates results solely based on input text and internal model knowledge, this application can introduce standard process constraints during the model generation stage, limiting the model generation space to actual process rules and reducing the probability of generating erroneous process content from the source.

[0088] Preferably, after completing the generative semantic model reasoning, the model output can be further processed by a structured transformation to generate process analysis feature representations corresponding to the process constraint information. Specifically, information on the operation object, operation behavior, execution conditions, and node associations can be extracted from the model output and converted into process analysis feature representations according to a preset data structure. By converting the model output in natural language form into a structured process representation, subsequent verification processes can be executed based on explicit process elements and logical relationships, without relying on human understanding of the model output.

[0089] Preferably, when validating the output results based on process analysis feature representation, the validity of process nodes in the model output can be checked first. Specifically, the process nodes in the process analysis feature representation can be matched with the operation nodes in the process constraint information to determine whether the process nodes in the model output fall within the scope of the standard process definition. When an abnormal process node that cannot be matched is detected in the model output, it can be determined that the generative semantic model may have generated process node illusion, and the corresponding abnormal node information can be recorded.

[0090] Preferably, when performing node relationship verification, the node association relationships in the process analysis feature representation can be detected based on the node constraint relationships in the process constraint information. Specifically, it can be detected whether the process nodes meet the requirements of temporal constraints, dependency constraints, and conditional constraints respectively. When it is detected that the order of process nodes in the model output is inconsistent with the order of the standard process, it can be determined that there is an abnormal temporal relationship; when it is detected that a process node lacks the state information provided by the necessary predecessor node, it can be determined that there is an abnormal dependency relationship; when it is detected that an operation node lacks the necessary constraints during execution, it can be determined that there is an abnormal conditional constraint.

[0091] Preferably, in some embodiments, the output result verification process can also be combined with process evolution characteristics for analysis. Specifically, the state change process recorded in the process analysis feature representation can be matched with the standard state evolution process corresponding to the process constraint information to determine whether the generative semantic model correctly understands the state transition relationship during process execution. For example, in a power supply switching process, the standard process requires the system state to change according to the process of "main power supply state - backup power supply preparation state - backup power supply state". If the model output result directly converts the main power supply state to the backup power supply state, omitting intermediate state changes, it can be determined that the output result has a state evolution understanding deviation based on the process evolution characteristics.

[0092] Preferably, through the above-described output verification process, the uncertainties that may arise from the generative semantic model can be converted into detectable non-compliance items. Compared to directly judging whether the model's output text is reasonable, this application establishes a structured detection basis through process constraint information, enabling the computer to automatically identify abnormal process nodes, erroneous process relationships, and non-compliant execution conditions in the model output, thereby improving the accuracy of generative semantic model output detection.

[0093] Preferably, after obtaining the non-compliant constraints, a result correction instruction can be generated based on the non-compliant constraints to adjust the output of the generative semantic model. The result correction instruction can determine the corresponding correction method based on the anomaly type, enabling the model to adjust for specific errors without needing to re-analyze all process data.

[0094] Preferably, when a process node corresponding to the non-compliant constraint becomes phantom, a result correction instruction for deleting, replacing, or remapping the abnormal process node can be generated. For example, when a generative semantic model generates a device calibration node that does not exist in the standard process based on industry-standard knowledge, it can determine that the node does not exist in the standard process based on the process constraint information and generate a correction instruction to delete the node.

[0095] Preferably, when the process relationship illusion corresponding to the non-compliance constraint is detected, a result correction instruction for adjusting the node association relationship can be generated. For example, when the generative semantic model incorrectly determines that there is a direct dependency relationship between two process nodes, but the process constraint information indicates that there are necessary intermediate nodes between them, a correction instruction for supplementing intermediate nodes or adjusting the execution order can be generated based on the corresponding node constraint relationship.

[0096] Preferably, when the process conditions corresponding to the non-compliance constraints are misleading, a result correction instruction can be generated to improve the execution condition information. For example, when the generative semantic model omits the status confirmation condition during the device switching process, the corresponding execution conditions can be supplemented according to the condition constraints, so that the final generated process analysis result meets the standard process requirements.

[0097] Preferably, after generating a result correction instruction based on non-compliant constraints, the result correction instruction can be fed back to the generative semantic model, allowing the generative semantic model to readjust the process analysis results according to the result correction instruction. Specifically, the generative semantic model can modify the abnormal content in the original output result based on the anomaly type, target process node, and corresponding constraint relationship contained in the correction instruction, and regenerate the target process analysis result that meets the process constraint requirements. Through the above feedback correction process, the output result of the generative semantic model can gradually approach the standard process logic requirements.

[0098] Preferably, in some embodiments, after the generative semantic model completes the output adjustment according to the result correction instruction, the adjusted process analysis result can be converted into a process analysis feature representation again, and a secondary verification can be performed based on the process constraint information. By repeatedly executing the process of "model generation - constraint detection - result correction - re-verification", the uncertainty in a single generation process can be further reduced, and the accuracy of the final process analysis result can be improved.

[0099] Preferably, during the secondary verification process, non-compliant constraints identified in the initial detection can be used as key detection targets to perform targeted verification on the corrected process analysis results. For example, if the initial detection finds that the generative semantic model is missing a certain preceding operation node, the focus after correction can be on whether the preceding operation node has been correctly added and whether the added node establishes a correct dependency relationship with subsequent operation nodes; if the initial detection finds that the process execution order is incorrect, the focus can be on whether the adjusted node order meets the corresponding timing constraints.

[0100] Preferably, through the aforementioned closed-loop processing method, this application can avoid the problem of relying solely on a single model output to determine the conclusion of process analysis. In the application of traditional generative semantic models, the model typically generates analysis results directly based on the input content, making it difficult for users or subsequent systems to determine whether the results conform to the rules of the professional domain. This application, by introducing process constraint information, makes the model output results verifiable and can automatically correct them based on constraint violations, thereby improving the stability of generative semantic models when applied to professional process analysis tasks.

[0101] Preferably, in some embodiments, the process constraint information can also be used to construct context-enhancing information for the generative semantic model. Specifically, based on the business scenario corresponding to the current process data to be analyzed, relevant operation nodes and node constraint relationships can be selected from pre-built process constraint information, and the relevant information and the process data to be analyzed can be input into the generative semantic model together. By selecting corresponding constraint content for the current process scenario, the influence of irrelevant process rules on the model reasoning process can be reduced, and the model analysis efficiency can be improved.

[0102] Preferably, in some embodiments, process constraint information can also participate in the generative semantic model processing through retrieval enhancement. Specifically, based on keywords, operation objects, and business types in the process data to be analyzed, standard process constraint content related to the current process can be retrieved from the process constraint information database, and the retrieved constraint content can be used as reference information in the model generation process. Through this method, the generative semantic model can dynamically acquire corresponding standard rules when facing different business processes, without having to input all process rules into the model simultaneously.

[0103] Preferably, in some embodiments, the generative semantic model can be domain-adapted according to different application scenarios to improve the model's understanding of specific process domains. The domain adaptation process can include domain terminology expansion, process description sample supplementation, prompt information optimization, or constraint information injection. Specifically, domain terminology expansion improves the model's ability to recognize professional object names and business expressions; process description sample supplementation enhances the model's understanding of specific process expressions; prompt information optimization guides the model to perform reasoning according to preset process analysis objectives; and constraint information injection integrates standard process logic into the model's analysis process.

[0104] Preferably, it should be noted that the generative semantic model in this application is not limited to a specific implementation form. It can be implemented based on different types of semantic generation models, such as pre-trained language models, domain-enhanced language models, or other models capable of performing semantic understanding and content generation processing. Regardless of the model structure used, as long as semantic analysis can be performed based on the input process data, and the process analysis results can be generated, verified, and corrected in combination with process constraint information, it is considered an implementation method of the technical solution of this application.

[0105] Preferably, through the above processing method, this application solves the problem of uncontrollable results when generative semantic models are applied to professional process analysis scenarios. Specifically, although traditional generative semantic models have strong natural language understanding capabilities, their generation process is based on probabilistic reasoning mechanisms, which may produce content inconsistent with actual business rules. This application constructs process constraint information and incorporates this information into the model input, model reasoning, and output verification processes, ensuring that the results generated by the generative semantic model are always constrained by standard process logic.

[0106] Preferably, compared with directly using generative semantic models to analyze process text, this application has at least the following technical effects: On the one hand, by introducing process constraint information during model reasoning, the problem of the model generating non-existent process content based on general knowledge can be reduced, and the accuracy of process node identification can be improved; on the other hand, by performing output verification based on node constraint relationships, erroneous process relationships and erroneous execution conditions in the model-generated results can be identified, and the accuracy of process logic analysis can be improved; furthermore, by generating correction instructions based on the violation of constraint terms, an automatic adjustment mechanism for model error output can be formed, improving the usability of generative semantic models in professional fields.

[0107] Preferably, in data center operation and maintenance scenarios, for example, when a generative semantic model analyzes a power switching process, the model may generate operations such as "automatic load migration" and "device re-initialization" that are not included in the current maintenance procedures, based on existing domain knowledge. By verifying the operation nodes in the process constraint information, it can be determined that these operation nodes do not exist in the standard power switching process and are treated as abnormal generated content. Simultaneously, when the model incorrectly determines that the "load status check node" can be executed before the "power path switching node," this logical error can be identified based on node timing constraints, and corrective instructions to adjust the node execution order can be generated.

[0108] Preferably, in the above application process, process constraint information actually serves as a bridge connecting business rules and the generative semantic model. On the one hand, process constraint information transforms standard process rules formulated by domain personnel into structured constraint relationships that computers can process, enabling business rules to participate in the model calculation process; on the other hand, the generative semantic model utilizes its own semantic understanding capabilities to process complex natural language process descriptions and limits the range of output results through process constraint information, thus forming a complementary relationship between the two. Therefore, this application can improve the ability of the generative semantic model to conform to actual business logic while retaining its flexible understanding capabilities.

[0109] Preferably, in summary, when combining generative semantic models with process analysis, this application does not simply replace traditional process analysis methods with generative semantic models. Instead, it addresses the uncertainty of results that exists in the application of generative semantic models in professional fields by introducing process constraint information to establish constraint enhancement, result detection, and feedback correction mechanisms. This enables the generative semantic model to complete process understanding and analysis tasks within the scope of standard process logic. Through this approach, the probability of the generative semantic model generating erroneous process information can be reduced, improving the accuracy and reliability of intelligent analysis results in complex operational process scenarios.

[0110] It should be noted that the generative semantic model in this application can be understood as an intelligent semantic processing model trained on a large-scale corpus. It can perform semantic understanding, information extraction, and content generation based on the input text information. Compared with traditional process parsing methods based on preset rules or keyword matching, the generative semantic model can understand complex natural language descriptions based on contextual information and extract latent semantic relationships from unstructured process data. Therefore, it is suitable for operational process analysis scenarios containing a large number of technical terms, complex execution conditions, and multi-layered logical relationships.

[0111] Preferably, when performing process analysis tasks, the generative semantic model can generate corresponding process analysis results based on the input process description. For example, for a description of an equipment maintenance process, the generative semantic model can identify the operation objects, operation behaviors, execution conditions, and operation purposes involved, and generate corresponding process node information or process relationship information. Compared with traditional process parsing methods, the generative semantic model can handle process descriptions with the same technical meaning under different expression forms, improving the ability to understand process semantics.

[0112] Preferably, however, since generative semantic models typically generate output based on statistical relationships in existing training data, their output does not always align with actual business rules. In professional process analysis scenarios, when input content lacks description, has insufficient domain knowledge, or incomplete contextual information, the generative semantic model may infer operational content that does not exist in the current process based on existing general knowledge, or incorrectly establish relationships between process nodes. For example, the generative semantic model may supplement operational steps not specified in the actual process, incorrectly determine the execution conditions of a certain operational node, or omit key constraints affecting the security of process execution. These issues can lead to inconsistencies between the output of the generative semantic model and the actual process logic.

[0113] Preferably, to address the potential uncertainties in the output results of the aforementioned generative semantic model, this application does not directly use the results generated by the generative semantic model as the basis for process analysis. Instead, it utilizes process constraint information to constrain the analysis process and output results of the generative semantic model. Specifically, the process constraint information is used to limit the valid operation nodes in the standard process and the execution logic relationships between nodes, enabling the process analysis results generated by the generative semantic model to be verified based on pre-determined process rules, thereby reducing the problem of erroneous process information generated by the model's own reasoning process.

[0114] Preferably, in this application, the generative semantic model can realize semantic understanding and content generation processing of process data based on a computer-executable program. When performing process analysis tasks, the generative semantic model can first obtain the process data to be processed and the process constraint information corresponding to the process data to be processed, and convert the process data to be processed into semantic representation data that can be recognized by a computer. Specifically, the input process text, operation records, or other forms of process description data can be processed by word segmentation, semantic encoding, and feature mapping to convert the original process description into a semantic feature representation containing word associations, entity information, and behavioral information.

[0115] Preferably, after obtaining the semantic feature representation, the generative semantic model can perform feature association calculations on the semantic feature representation based on a pre-trained parameter set to determine the association relationships between different semantic units. The semantic units can correspond to information such as operation objects, operation behaviors, execution conditions, and process relationships in the process description. By calculating the degree of association between multiple semantic units, the generative semantic model can determine the operation nodes that may be included in the current process description and the association relationships between these nodes.

[0116] Preferably, during the semantic reasoning process of the generative semantic model, the structured representation corresponding to the process constraint information can be converted into a constraint feature representation that the model can recognize, and the constraint feature representation is fused with the semantic feature representation corresponding to the process data. Specifically, node constraint features can be generated based on the operation node information in the process constraint information, and relation constraint features can be generated based on the node constraint relationships. The node constraint features and relation constraint features are then correlated with the process semantic features for calculation, so that the generative semantic model can simultaneously refer to the execution logic in the standard process when predicting process nodes and generating process relationships.

[0117] Preferably, after completing the constraint feature fusion, the generative semantic model can perform probabilistic inference calculations based on the fused semantic features to determine the generation probabilities corresponding to candidate process nodes, candidate operation relationships, and candidate execution conditions, and select the analysis results that meet the process constraint requirements according to preset screening conditions. The probabilistic inference process is not only based on the semantic relationships in the input process text but is also influenced by the constraint features corresponding to the process constraint information, thus reducing the possibility of the model generating content inconsistent with the standard process logic.

[0118] Preferably, when generating process analysis results, the generative semantic model can generate structured process analysis results based on the determined process nodes, node relationships, and execution condition information, and further convert the process analysis results into process analysis feature representations for consistency verification with process constraint information. When the process nodes, node relationships, or execution conditions generated by the generative semantic model do not meet the requirements of the process constraint information, it can be determined that the generated results are abnormal, and correction information can be generated based on the corresponding abnormality.

[0119] It should be noted that the generative semantic model used in this application can be implemented based on existing generative artificial intelligence technologies, and its specific model structure, training method, and parameter optimization process do not constitute a limitation of this application. In different embodiments, the generative semantic model can be implemented using different types of pre-trained language models, domain-enhanced language models, or other models with semantic understanding and content generation capabilities, and this application does not make specific limitations in this regard. Existing generative semantic models typically use large-scale training data for pre-training to enable the model to learn the relationships between different semantic units and generate corresponding semantic processing results based on input data. The training process can be implemented using existing model training methods, such as adjusting model parameters based on training samples, updating model parameters according to preset optimization goals, or performing model adaptation processing in conjunction with domain data. Since the above model training process belongs to existing technologies in the field of generative artificial intelligence, this application does not limit the specific training data construction method, model parameter setting method, training iteration process, and loss function type of the generative semantic model.

[0120] Preferably, in some embodiments, the generative semantic model can adopt different training strategies or model optimization methods according to different application requirements. For example, a basic semantic model can be obtained based on general domain training data, or it can be further combined with data from the target application domain to perform domain adaptation processing. Furthermore, the model's adaptability to specific tasks can be improved through prompt information optimization, knowledge enhancement, or other methods. However, regardless of the model training or optimization method used, as long as the generative semantic model can receive process data and perform semantic understanding, information association, and result generation processing, it can be applied to the technical solution of this application.

[0121] It is important to note that the technical solution of this application does not aim to improve the basic structure or training process of generative semantic models. Rather, it addresses the problems of uncertain output results and insufficient business logic constraints that exist when generative semantic models are applied to professional process analysis scenarios by introducing process constraint information to constrain the model's processing and output results. Specifically, this application constructs process constraint information containing operation nodes and node constraint relationships, enabling generative semantic models to perform semantic understanding and result generation within the scope of standard process logic. Furthermore, it reduces the impact of erroneously generated content on the process analysis results through constraint verification and result correction mechanisms.

[0122] Therefore, this application focuses on the application control mechanism of generative semantic models in specific business scenarios, rather than the training mechanism of the generative semantic models themselves. By converting domain process rules into computer-processable process constraint information and incorporating this constraint information into model input, semantic reasoning, and result verification processes, this application can improve the reliability and accuracy of generative semantic models in complex operational process analysis scenarios.

[0123] To further illustrate the specific application process of the technical solution in this application, the following example uses a data center power supply switching process consistency detection scenario. During data center operation, to ensure the continuous and stable operation of server clusters, storage devices, and network equipment, a primary power supply system and a backup power supply system are typically configured, and corresponding power supply switching procedures are established. When the primary power supply system malfunctions or requires planned maintenance, the data center power supply status needs to be switched to the backup power supply status according to the preset operating procedure. Since the power supply switching process involves multiple device status changes and multiple operating steps, if steps are omitted, the execution order is incorrect, or the execution conditions are insufficient during the process, it may lead to power supply switching failure, or even affect the continuous operation of data center services. Therefore, it is necessary to detect the consistency between the actual execution process and the standard power supply switching process.

[0124] Preferably, in this application scenario, the standard power switching operation procedure formulated by the data center operations and maintenance department can be obtained first as standardized operation process data. The standard power switching operation procedure may include multiple process steps such as backup power supply equipment status confirmation, power switching condition confirmation, primary power supply path disconnection, backup power supply path activation, and server operation status check. By performing semantic parsing on the above standard operation procedure, the operation object, operation behavior, and execution conditions corresponding to each process step can be determined, and corresponding process semantic parsing information can be generated.

[0125] Preferably, when parsing the standard power supply switching operation procedure, the descriptions such as "confirming the normal operating status of the backup power supply equipment," "confirming that the switching conditions are met," "disconnecting the main power supply path," "activating the backup power supply path," and "checking the server's operating status" can be divided into independent process description units, and corresponding operation nodes can be generated based on the semantic information in each process description unit. Specifically, the operation object corresponding to the backup power supply status confirmation node is the backup power supply equipment, the operation behavior is status detection, the execution condition is meeting the equipment detection requirements, and the status change information is generating a backup power supply status confirmation result; the operation object corresponding to the switching condition confirmation node is the switching environment, the operation behavior is condition confirmation, the execution condition is meeting the preset switching rules, and the status change information is generating a switching permission status; the operation object corresponding to the power supply path switching node is the power supply path, the operation behavior is power supply mode switching, and the execution condition is that the backup power supply status meets the requirements.

[0126] Preferably, after generating the aforementioned operation nodes, node constraint relationships can be further generated based on the execution logic relationships between the operation nodes. For example, timing constraints can be generated based on the execution order relationship of "backup power supply status confirmation node → switching condition confirmation node → power supply path switching node"; dependency constraints can be generated based on the state transitivity relationship of "the status confirmation result generated by the backup power supply status confirmation node serves as the basis for the execution of the power supply path switching node"; and condition constraints can be generated based on the constraint requirement that "the power supply path switching node needs to meet the requirements of normal backup power supply status and satisfied switching conditions". Through the above processing, flow constraint information containing operation nodes and node constraint relationships can be formed to represent the complete execution logic in the standard power supply switching process.

[0127] Preferably, during the actual execution of maintenance tasks, the operation process data to be tested submitted by the maintenance personnel can be obtained. For example, a temporary maintenance plan may record: "After confirming that the backup power supply equipment is in normal condition, perform the primary / backup power supply switch and check the server's operating status." After parsing the operation process to be tested, process feature representations corresponding to the backup power supply status confirmation node, power supply path switching node, and server operating status check node can be generated.

[0128] Preferably, after generating the process feature representation corresponding to the operation process to be tested, the process feature representation can be associated and mapped with the process constraint information corresponding to the standard power supply switching process to determine the correspondence between the nodes of the process to be tested and the nodes of the standard process operation. After node mapping, it can be found that although the process to be tested includes a backup power supply status confirmation node and a power supply path switching node, it lacks the switching condition confirmation node in the standard process. Furthermore, according to the node constraint relationship in the process constraint information, it can be determined that the switching condition confirmation node is a prerequisite node of the power supply path switching node, therefore, the current process to be tested has a missing process node problem.

[0129] Preferably, during constraint compliance analysis, the process to be tested can be further tested based on timing constraints, dependency constraints, and conditional constraints. For example, based on timing constraints, it can be determined whether the backup power supply status confirmation node in the process to be tested is located before the power supply path switching node; based on dependency constraints, it can be determined whether the power supply path switching node has obtained the necessary prerequisite status information; based on conditional constraints, it can be determined whether the backup power supply status and switching conditions meet the execution requirements. Through multi-dimensional constraint joint detection, process logic anomalies that cannot be detected by step comparison alone can be identified.

[0130] Preferably, when a missing switching condition confirmation node is detected in the process to be tested, a consistency detection result can be generated based on the corresponding node constraint relationship. The consistency detection result may include the target process node, the deviation type, and the violated constraint relationship. For example, a detection result such as "The power supply path switching node has a dependency constraint not satisfied, and the corresponding prerequisite constraint for the switching condition confirmation node is missing" can be generated. By outputting the specific violated constraint relationship, maintenance personnel can directly locate the cause of the process anomaly and adjust the operation plan based on the detection result.

[0131] Preferably, when further combining generative semantic model (GSM) with process analysis, the text of the operation process to be detected, process constraint information, and detection target can be input into the GSM to assist in extracting process nodes and node relationships. Subsequently, the output of the GSM can be converted into process analysis feature representations, and verification can be performed based on the process constraint information. When it is detected that the GSM has missed a switching condition confirmation node or incorrectly judged that the power supply switching process meets the requirements, a result correction instruction can be generated based on the non-compliance constraint, so that the GSM can supplement the missing process nodes or adjust the process logic relationship.

[0132] The consistency detection process of the data center power switching procedure described above verifies the application of the technical solution in actual operation and maintenance scenarios. This solution does not simply compare the similarity of process texts; instead, it constructs process constraint information containing operation nodes and node constraint relationships, converting the execution logic in the standard process into a structured representation, and then performs consistency analysis on the process to be tested based on this structured representation. Therefore, even if the process to be tested uses different textual expressions or has implicit logical gaps, this application can still identify corresponding anomalies based on the process execution constraint relationships, improving the consistency detection capability in complex operation process scenarios.

[0133] like Figure 5 As shown, this application also provides an operation process consistency detection device based on process constraint information, comprising: The process data parsing module is used to acquire standardized operation process data and perform semantic parsing on the standardized operation process data to obtain process semantic parsing information. The process semantic parsing information includes operation object characteristics, operation behavior characteristics, execution condition characteristics, and process association characteristics. A process constraint construction module is used to construct process constraint information based on the process semantic parsing information. The process constraint information includes operation nodes for representing process execution units and node constraint relationships for representing the execution logic relationships between the operation nodes. The process feature representation module is used to acquire the operation process data to be detected, perform semantic parsing on the operation process data to be detected, and generate a process feature representation, which is used to represent the process nodes and node relationships in the operation process to be detected. The consistency detection module is used to perform constraint compliance analysis on the process node relationships in the operation process to be detected based on the correlation between the process constraint information and the process feature representation, so as to obtain the operation process consistency detection result.

[0134] like Figure 6 As shown, an electronic device according to an embodiment of this application includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the operation flow consistency detection method based on process constraint information as described above in this application.

[0135] like Figure 7 As shown, this is a computer-readable storage medium according to an embodiment of the present application. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the operation flow consistency detection method based on process constraint information as described above in this application.

[0136] The above Figures 5-7 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A method for detecting operational process consistency based on process constraint information, characterized in that, include: Obtain standardized operation process data, and perform semantic parsing on the standardized operation process data to obtain process semantic parsing information. The process semantic parsing information includes operation object characteristics, operation behavior characteristics, execution condition characteristics, and process association characteristics. Based on the process semantic parsing information, process constraint information is constructed. The process constraint information includes operation nodes for representing process execution units, and node constraint relationships for representing the execution logic relationships between the operation nodes. Obtain the operation process data to be detected, perform semantic parsing on the operation process data to be detected, and generate a process feature representation, which is used to represent the process nodes and node relationships in the operation process to be detected. Based on the correlation between the process constraint information and the process feature representation, constraint compliance analysis is performed on the process node relationships in the operation process to be detected, so as to obtain the operation process consistency detection result.

2. The method for detecting operational process consistency based on process constraint information according to claim 1, characterized in that, The semantic parsing information of the process includes entity information, behavior information, condition information, and relationship information; In the process of semantic parsing the standardized operation process data, the corresponding operation object entity is determined according to the process description content, the corresponding operation behavior is determined according to the action description content, and the execution conditions corresponding to the operation behavior and the association between different process elements are determined according to the context semantics.

3. The method for detecting operational process consistency based on process constraint information according to claim 1, characterized in that, Based on the process semantic parsing information, process constraint information is constructed, including: An operation node is generated based on the characteristics of the operation object, the characteristics of the operation behavior, and the characteristics of the execution conditions. The operation node includes node identification information, operation object information, operation behavior information, execution condition information, and state change information. The state change information is used to indicate the process state change generated after the corresponding operation behavior is executed.

4. The method for detecting operational process consistency based on process constraint information according to claim 3, characterized in that, The node constraint relationships include temporal constraints, dependency constraints, and conditional constraints; Specifically, timing constraints are generated based on the execution order relationship between multiple operation nodes, dependency constraints are generated based on the state transmission relationship, data association relationship or functional association relationship between multiple operation nodes, and condition constraints are generated based on the state requirements, environmental requirements or business constraints corresponding to the operation nodes.

5. The method for detecting operational process consistency based on process constraint information according to claim 4, characterized in that, Also includes: Candidate implicit constraint relationships are generated based on the relationships between operation objects, state changes, and execution conditions among multiple operation nodes. The validity of the candidate implicit constraint relationships is determined, and the candidate implicit constraint relationships that meet the preset judgment conditions are added to the process constraint information to update the node constraint relationships.

6. The method for detecting operational process consistency based on process constraint information according to claim 1, characterized in that, Semantic parsing is performed on the data of the operation process to be detected to generate process feature representations, including: The data of the operation process to be detected is divided into process description units, and the operation object information, operation behavior information and execution condition information corresponding to each process description unit are extracted to generate the corresponding process node representation.

7. The method for detecting operational process consistency based on process constraint information according to claim 6, characterized in that, Generating the process feature representation includes: The operation object information, operation behavior information, and execution condition information in the process node representation are semantically normalized. The relationships between nodes are determined based on the normalized process node representation, so as to generate a process feature representation that includes process node information and node relationships.

8. An operation process consistency detection device based on process constraint information, characterized in that, include: The process data parsing module is used to acquire standardized operation process data and perform semantic parsing on the standardized operation process data to obtain process semantic parsing information. The process semantic parsing information includes operation object characteristics, operation behavior characteristics, execution condition characteristics, and process association characteristics. A process constraint construction module is used to construct process constraint information based on the process semantic parsing information. The process constraint information includes operation nodes for representing process execution units and node constraint relationships for representing the execution logic relationships between the operation nodes. The process feature representation module is used to acquire the operation process data to be detected, perform semantic parsing on the operation process data to be detected, and generate a process feature representation, which is used to represent the process nodes and node relationships in the operation process to be detected. The consistency detection module is used to perform constraint compliance analysis on the process node relationships in the operation process to be detected based on the correlation between the process constraint information and the process feature representation, so as to obtain the operation process consistency detection result.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the operation process consistency detection method based on process constraint information as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the operation flow consistency detection method based on process constraint information as described in any one of claims 1 to 7.

Images