A text-to-state machine method and system for programmable configuration assistance in thermal power units

By using a text-to-state-machine method for thermal power unit programmable configuration assistance, the source text is parsed and verified, implicit steps are automatically completed and measurement points/commands are bound, and a configuration assistance description file is generated. This solves the problems of long translation cycle and unreliability of thermal power unit operation procedures to control logic, and achieves the structural integrity and executability of the logic.

CN122840077BActive Publication Date: 2026-11-17ZHEJIANG BAIMA LAKE LABORATORY CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611339159.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-01
Publication Date
2026-11-17
Estimated Expiration
2046-09-01

AI Technical Summary

Technical Problem

In existing technologies, the translation cycle from operating procedures to control logic of thermal power units is long, relies on human experience, and the generated logic structure is unreliable. It also lacks connection with the actual measurement point/command system, resulting in the generated state machine being unexecutable or unverifiable.

Method used

A text-to-state-machine method for programmable configuration of thermal power units is adopted. By parsing the source text to generate candidate structured operation description units, verifying them according to a preset mode, mapping and generating a state machine, automatically completing the hidden steps, binding executable measurement points and commands, and generating configuration-aided description files and conversion reports.

Benefits of technology

It shortens the configuration development cycle, improves the structural integrity and executability of the generated logic, ensures that the product is bound to the real measurement point/command system, provides executability annotation and dual-source playback verification, and avoids generating unexecutable or unverifiable state machines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122840077B_ABST
    Figure CN122840077B_ABST
Patent Text Reader

Abstract

This invention discloses a text-to-state-machine method and system for programmable configuration assistance in thermal power units, comprising: parsing source text to generate candidate structured operation description units, and performing pattern verification according to the structured description mode; generating a state machine containing nodes and transition edges based on the verified structured operation description units and mapping according to the mapping rule base; identifying implicit steps not explicitly stated in the source text according to domain completion rules, and automatically completing the corresponding nodes; binding action items in the state machine to corresponding writable commands, and binding condition items to corresponding readable measurement points, the binding process including multi-candidate disambiguation; marking the executability of nodes according to the binding results, and downgrading nodes that trigger degradation conditions to retain manual step nodes that connect the preceding and following transitions; generating configuration assistance description files and conversion reports. This invention anchors the generation logic to the real measurement point / command system, makes executability verifiable, and improves the structural integrity, executability, and reviewability of configuration assistance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, and in particular to a text-to-state machine method and system for programmable configuration assistance of thermal power units. Background Technology

[0002] In DCS configuration software, the APS / sequential control logic of thermal power units is manually constructed by control engineers based on operating procedures, startup curves, interlocking protection setting sheets, etc., using function block diagrams and sequential control diagrams. The entire process from logic development to commissioning of an APS typically takes months, involving a large workload and relying heavily on the engineer's experience. Operating procedures are natural language text written for operators, while APS / sequential control configuration is machine-executed logic. The former contains numerous natural language descriptions such as "confirm a parameter is normal," "wait for equipment to stabilize," and "if abnormal, pause and manually confirm." The latter requires a machine-readable, verifiable, configurable, and traceable control logic structure, such as functional group structures, step nodes, state transitions, completion criteria, waiting / timeout conditions, abnormal jump conditions, measurement points, and command references. A significant gap exists between the two: the expression from human-generated procedural text to machine-processable programmable control logic. Currently, the translation from procedures to control logic relies heavily on manual labor, resulting in long cycles, high costs, experience dependence, and a lack of structured intermediate products. When generating control logic directly from a language model, the output format and content are not structurally constrained, making it impossible to guarantee the structural correctness of the generated logic at the semantic level of thermal power plant operation. Furthermore, operating procedure texts are written for operators and generally omit implicit steps that are self-evident to humans. However, when converting these texts into control logic, these implicit steps are omitted, resulting in incomplete logic that cannot function as an executable state machine. Moreover, the generated logic is disconnected from the actual unit's measurement / command system, and even if a state machine is generated, there is a lack of explicit distinction regarding the executableness of nodes.

[0003] The Chinese patent document "A Model-Based Code Generation Method and Related Apparatus" (publication number CN120560623A, publication date August 29, 2025) discloses a method for generating code based on a given requirements document. This method uses an AI model to analyze the business processes within the requirements document and represents them using a state machine. It achieves an interactive interface that represents the business processes based on states, and interactive actions based on events that trigger state transitions. Further analysis of the state machine representation using the AI ​​model automatically generates code, eliminating the need for manual coding and improving efficiency. However, this technology is only a general method for converting natural language descriptions into control code. In scenarios involving converting operating procedures for thermal power units into state machines, it still misses implicit steps omitted in the text, leading to unreliable and incomplete logic. It also fails to address the issue of associating with specific power plant measurement point libraries and control command libraries. Furthermore, the lack of explicit differentiation of node executability results in seemingly complete outputs that are actually unexecutable or unverifiable, making them unsuitable for engineering review and difficult for engineers to edit. Summary of the Invention

[0004] This invention aims to overcome the problems in existing technologies when converting text such as operating procedures and APS control steps of thermal power units into state machines, including long manual translation cycles, unreliable and incomplete logic structures directly generated from models, and a lack of distinction between generated logic and the actual measurement point / command system and executable functionality. It provides a text-to-state-machine method and system assisted by the programmable configuration of thermal power units.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A text-to-state machine method for programmable configuration-assisted operation of thermal power units includes: Parse the source text to generate candidate structured operation description units, and perform pattern verification according to the preset structured description pattern; The structured operation description unit is verified, and a state machine containing nodes and transition edges is generated according to the mapping rule library. Based on the domain completion rules, the implicit steps not explicitly stated in the source text are identified, and the corresponding nodes are automatically completed. Bind the action items in the state machine to the corresponding writable commands, and bind the condition items to the corresponding readable test points. The binding process includes multi-candidate disambiguation. Based on the binding results, the nodes are marked as executable. Nodes that trigger the degradation conditions are downgraded to manual step nodes that retain the connection between the previous and subsequent migrations. Generate configuration auxiliary description files and conversion reports.

[0006] This invention converts texts such as thermal power unit operating procedures, unit automatic start-stop (APS) programmable sequence steps, and operation tickets into state machines for auxiliary development of programmable control configuration. The generated state machines and configuration auxiliary description files are used by engineers for review, secondary editing, or conversion to target configuration tools to assist in the configuration development and modification of APS / sequential control logic. It can also be used for scenarios such as generating sequential control logic drafts, organizing deep peak-shaving operation sequences, and structuring operation tickets. Unlike existing general text-to-code generation technologies, this invention targets the complex step control of thermal power unit APS / sequential control—a multi-functional control system with layered breakpoints and condition adaptation. The target output is a state machine and configuration auxiliary description file bound to the actual unit's measurement point / command system and marked with executable characteristics, for engineers to review and edit, rather than directly issued code or instructions.

[0007] Preferably, a dual-source playback verification step is also included before generating the configuration auxiliary description file and conversion report: Acquire the event sequence of successfully completed historical operations, including digital event records and analog trend data; The event is associated with the state machine node based on the bound measurement point identifier; the switch quantity event drives the switch quantity node migration in time sequence; the analog quantity trend data is compared with the comparison threshold of the condition item within the corresponding event time window to determine the analog quantity condition; Check the preset validation items, output a list of inconsistent nodes and trace back to the source text entry.

[0008] Preferably, the generated candidate structured operation description unit is implemented by a rule template channel, a language model channel, or a combination of the two; For candidate structured operation description units that fail the verification, feedback information carrying the reason for the failure is generated and fed back to the generation stage for regeneration or rollback to the rule template channel; those that fail to be regenerated after a preset number of times are marked as manually edited items.

[0009] Preferably, the nodes of the state machine include operation nodes and confirmation nodes, as well as at least one of waiting nodes, branch nodes, parallel nodes, exception handling nodes, and manual step nodes; the migration edges between nodes represent at least one of execution order, condition satisfaction, timeout, parallel splitting and merging, and exception jump relationship.

[0010] Preferably, the domain completion rules include a mapping from action type to confirmation signal type, a default timeout duration configured according to device category, and a default jump strategy for abnormal handling nodes. The automatic completion of the corresponding nodes involves generating corresponding confirmation nodes or timeout handling nodes based on the identified implicit steps, and recording the applied completion rule identifier, action type, device category, and corresponding source text position for each completion node.

[0011] As a preferred method, for action items, a control command with a writable attribute is retrieved from the control command information database and bound; For the condition item, retrieve and bind measurement points with readable attributes from the measurement point information database; The multi-candidate disambiguation process involves matching and generating a set of candidates to be bound in the control command information database and the measurement point information database; after filtering based on preset hard filtering conditions, weighted scoring is performed according to preset scoring indicators and the scores are sorted in descending order; when the highest score is unique and greater than or equal to the preset binding threshold, and the difference between the highest score and the second highest score is greater than or equal to the score difference interval threshold, the binding is successful.

[0012] Preferably, the step of annotating the nodes with executability based on the binding result includes at least the following: Action nodes are marked as candidates for automatic execution when they are bound to writable commands and meet their security attribute requirements. When a node is confirmed to be bound to a readable measurement point, it is marked as an automatic verification candidate. When the binding is ambiguous and cannot be uniquely determined, retain the candidate binding relationship and mark it as pending engineer review.

[0013] Preferably, the degradation conditions include: The set of candidates to be bound after hard filtering is empty; or The highest score is lower than a preset binding threshold; or The difference between the highest and second-highest scores is less than the score difference interval threshold; or The read / write attributes of the successfully bound candidate do not match the action or condition to be bound; Mark the reason for the downgrade after downgrading.

[0014] As a preferred option, the process of mapping and generating a state machine also includes mapping the functional group hierarchy of the text into a hierarchical state machine, expanding the functional group nodes of the main state machine into sub-state machines, and sending the termination conditions of the sub-state machines back to the main state machine as subsequent transition conditions. Configure the manual step nodes generated by the downgrade as APS breakpoint type nodes. Suspend them after the previous node is completed, output the suspension status and the content to be manually executed to the breakpoint field of the configuration auxiliary description file, and continue the process through subsequent migration edges after manual confirmation.

[0015] As a preferred approach, for conditional items bound to readable measurement points, their comparison thresholds are parsed according to the following priority: explicit values ​​in the source text; set values ​​obtained by matching the set value sheet or interlocking protection set value library; historical operation data statistical baseline; if none of these can be determined, they are marked as manually edited items; the source level used for parsing is written into the conversion traceability record in the conversion report. For operation nodes whose action type is adjustment, their confirmation nodes are condition items and bound to readable measurement points. The comparison target value of the confirmation node is preferentially taken from the target parameter field value recorded in the structured operation description unit corresponding to the operation node. Only when the target parameter field is a non-numerical expression or is defaulted, the comparison threshold is parsed according to priority.

[0016] A text-to-state machine system with programmable configuration assistance for thermal power units includes: The parsing module parses the source text to generate candidate structured operation description units; The verification module performs pattern verification on candidate structured operation description units; The mapping completion module maps the validated structured operation description units to generate a state machine, completing the implicit content not explicitly stated in the source text; Bind the disambiguation module, bind writable commands to action items, bind readable test points to condition items, and perform multi-candidate disambiguation; The degradation annotation module annotates nodes for executability based on the binding results, and downgrades nodes that trigger degradation conditions to manual step nodes. The generation module generates configuration auxiliary description files and conversion reports.

[0017] Compared to existing technologies, this invention offers the following advantages: It transforms the manual translation of procedures into APS / sequential control logic into an automated auxiliary process that produces structured intermediate representations and executable, definable state machines. This improves the structural integrity and reproducibility of the generated logic, leaving verifiable intermediate products, shortening the configuration development cycle, and reducing reliance on personal experience. Through measurement point / command binding, disambiguation, and degradation mechanisms, the generated logic is anchored to the actual measurement point / command system of specific units. Steps that cannot be implemented are explicitly downgraded to manual steps rather than silently failing, ensuring the engineering usability of the product. Through executability annotation, it explicitly distinguishes between automatic execution candidates, automatic verification candidates, and manual actions. The system incorporates various states such as manual confirmation, pending measurement points, pending commands, and nodes that cannot be automatically executed, avoiding the generation of seemingly complete but actually unexecutable or unverifiable state machines. Implicit step completion compensates for omitted confirmation, waiting, and timeout handling in the procedures, ensuring the generated state machine is complete and executable. Dual-source playback verification of historical event sequences utilizes the actual operation of the unit to identify unreachable migrations or incorrect confirmation sequences in the generated logic, improving its reliability. Intelligent parsing results, such as language models, are restricted to the candidate stage and must undergo pattern verification before entering the mapping process, preventing the direct generation of uncontrollable control logic. The output is positioned as configuration assistance rather than direct execution, with clear safety boundaries and engineering feasibility. Attached Figure Description

[0018] Figure 1 This is a flowchart of a text-to-state machine method for programmable configuration assistance of thermal power units in this invention.

[0019] Figure 2 This is a flowchart of the pattern verification process for candidate structured operation description units in this invention.

[0020] Figure 3 This is a flowchart of the binding of measurement points and commands and the degradation of executability in this invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted that all data in the present invention were collected under the premise of legality and compliance.

[0022] like Figure 1 As shown, a text-to-state machine method for programmable configuration assistance in thermal power units includes: Parse the source text to generate candidate structured operation description units, and perform pattern verification according to the preset structured description pattern; The structured operation description unit is verified, and a state machine containing nodes and transition edges is generated according to the mapping rule library. Based on the domain completion rules, the implicit steps not explicitly stated in the source text are identified, and the corresponding nodes are automatically completed. Bind the action items in the state machine to the corresponding writable commands, and bind the condition items to the corresponding readable test points. The binding process includes multi-candidate disambiguation. Based on the binding results, the nodes are marked as executable. Nodes that trigger the degradation conditions are downgraded to nodes that retain the manual steps connecting the previous and subsequent migrations, and the reason for the degradation is marked. Generate configuration auxiliary description files and conversion reports containing conversion traceability records.

[0023] This invention converts texts such as thermal power unit operating procedures, unit automatic start-stop (APS) programmable sequence steps, and operation tickets into state machines for auxiliary development of programmable control configuration. The generated state machines and configuration auxiliary description files are used by engineers for review, secondary editing, or conversion to target configuration tools to assist in the configuration development and modification of APS / sequential control logic. It can also be used for scenarios such as generating sequential control logic drafts, organizing deep peak-shaving operation sequences, and structuring operation tickets. Unlike existing general text-to-code generation technologies, this invention targets the complex step control of thermal power unit APS / sequential control—a multi-functional control system with layered breakpoints and condition adaptation. The target product is a state machine and configuration auxiliary description file bound to the actual unit's measurement point / command system and marked with executable characteristics, for engineers to review and edit, rather than directly issued code or instructions. Before entering the execution phase, the state machine generated by this invention must undergo pre-execution verification, manual confirmation, and plant-level interlocking protection logic constraints, without changing or replacing the inherent interlocking protection logic of the plant-level control system.

[0024] In the final deliverable generation stage, in addition to generating configuration auxiliary description files and conversion reports, candidate execution plans, measurement point binding tables, command binding tables, lists requiring manual confirmation, lists of non-automatic execution, playback verification reports, and project review reports can also be generated. Configuration auxiliary description files are used for engineer review, secondary editing, or conversion to target configuration tools; candidate execution plans are only used as input for external verification systems or engineer review; conversion reports include the backtracking of each node of the state machine to the source text entry, structured operation description units, and the applied mapping rules or completion rules (i.e., conversion tracing records).

[0025] As a specific implementation, before parsing the source text, it is necessary to first obtain the source file and the necessary equipment information library, measurement point information library, control command information library, structured description pattern and mapping rule library required for state machine transformation.

[0026] Specifically, the source text includes the operating procedure text of the thermal power unit, the APS program control sequence description, or the operation ticket text, which is in natural language or semi-structured text.

[0027] The equipment information database includes equipment code, equipment name, system to which it belongs, equipment level, equipment type, associated measurement points, and allowed start / stop conditions.

[0028] The measuring points in the measuring point information database have readable attributes determined by their signal type, including measuring point number, description, signal type (AI / AO / DI / DO, etc.), dimensions, range, associated process subsystem, associated equipment, and readable attributes. The control commands in the control command information database have writable attributes determined by their signal type, including command code, command name, target object, command type, writable attributes, write-allowed conditions, security level, and associated equipment.

[0029] In this invention, the readable and writable attributes of measurement points or commands are determined based on their signal types. Specifically, the correspondence is as follows: input signals (such as DI (digital input) and AI (analog input) correspond to readable attributes, used to read equipment status or process measurement values ​​to determine condition items; output signals (such as DO (digital output) and AO (analog output) correspond to writable attributes, used to issue instructions or set values ​​to the equipment to execute action items. Correspondingly, status and measurement points (DI / AI) in the measurement point information database have readable attributes, while instruction and setting points (DO / AO) in the control command information database have writable attributes. During binding, condition items must be bound to measurement points with readable attributes, and action items must be bound to control commands with writable attributes. If an item can only be bound to an object with a mismatched attribute (e.g., an action item can only be bound to a read-only measurement point), it is determined to be a mismatched attribute and enters node degradation. The above correspondence is configurable and can be adjusted according to the actual signal system of the unit, but it must remain consistent within the same conversion task after configuration.

[0030] The structured description pattern specifies the field composition, field value range, referenced object type, functional group hierarchy, and step dependency validity conditions of the structured operation description unit.

[0031] The field composition includes at least one of the following: function group identifier, function subgroup identifier, device object, action type, execution prerequisite, target parameter, completion criterion, waiting condition, timeout condition, exception handling, manual confirmation mark, readable measurement point reference, writable command reference, and source text location; the referenced object type includes device, measurement point, control command, and parameter; the step dependency legality condition includes acyclic step dependency and the step or state corresponding to the execution prerequisite being reachable.

[0032] The mapping rule base contains mapping rules from structured operation description units to state machine nodes and domain completion rules (such as mapping from action type to acknowledgment signal type, default timeout configuration by device category, default jump strategy for exception handling, etc.).

[0033] As a specific example, such as Figure 2 As shown, the generation of candidate structured operation description units from parsed source text can be achieved by rule template channels, language model channels, or a combination of both. When performing pattern validation according to the preset structured description pattern, the structured operation description unit that passes the validation can enter the subsequent mapping generation and implicit step completion process; for the candidate structured operation description unit that fails the validation, feedback information carrying the reason for failure is generated and fed back to the generation stage for regeneration or rollback to the rule template channel; items that fail to be regenerated after a preset number of times are marked as manual editing items.

[0034] It should be noted that the process of generating a state machine from structured operation description units in this invention is deterministic. That is, under the same input text, the same structured description pattern, and the same mapping rule base, the mapping generation process produces a uniquely determined state machine output, without any random sampling. The language model channel is only used to generate candidate structured operation description units and must pass pattern verification before entering the mapping generation process; its output does not directly constitute any product. The pattern verification in this invention targets the semantics of thermal power plant operations, not the code syntax.

[0035] Specifically, the rule template channel performs structured extraction based on procedural sentence templates and domain terminology; the language model channel calls the language model to generate candidate structured operation descriptions, the engineering implementation of which includes: prompting the construction of fragments containing structured description patterns, domain terminology, and a small number of sample examples to constrain the output structure; imposing structured format constraints on the output, accepting only candidates that conform to the skeleton of the structured description pattern; the confidence level is determined based on the field-level consistency of multiple independent generation results of the same text, and when the confidence level is lower than a preset threshold, the corresponding text is processed by the rule template channel; the failure reason of pattern validation is used as an additional input for the next round of generation and participates in regeneration; maintaining the evaluation set of procedural entries and expected structured results, and performing regression evaluation after changes to the candidate generation channel or rule base, with relevant statistics (first-time interception rate, pass rate after feedback, etc.) included in the conversion report; the output of the language model channel is only a candidate and must pass pattern validation before entering the mapping generation.

[0036] The above-mentioned specific technical solutions were prototyped and verified. The language model independently generated 5 candidates for each procedure item. The differences in field-level descriptions among the candidates ranged from 21.6% to 30.8% (4 ways of writing object names, mixed use of OPEN / START action types, different wording of completion criteria, and different selection of optional fields). The threshold value and comparison direction remained consistent in multiple generation processes, indicating that the language model is stable for numerical facts but random for descriptions. After pattern verification and mapping, all candidates converged to a unique state machine output.

[0037] Pattern validation includes at least two of the following: field integrity validation, field value range validation, reference object consistency validation, abnormal branch integrity validation, and step dependency validity validation.

[0038] Feedback information generated by failed pattern validation includes the failure field identifier, failure reason category (field missing / value out of domain / referenced object not existing / absent abnormal branch / illegal dependency, etc.), and repair suggestions. This feedback information serves as additional input during regeneration. For example, if a procedure paragraph, "After starting the device, confirm normal operation; if abnormal, handle it," is generated as a candidate through the language model channel, pattern validation will find that it lacks completion criteria, abnormal triggering conditions, and abnormal jump targets, generating corresponding feedback. If the regeneration still does not meet the requirements of the structured description pattern, the process will revert to the rule template channel, breaking down the step into action nodes, manual confirmation nodes, and nodes for handling abnormalities to be added, with missing fields marked as items requiring manual confirmation. Structured operation description units that fail validation after a preset number of regenerations are marked as manually editable items and submitted to engineers for manual completion in the configuration auxiliary interface. The completed results are then re-entered for validation and mapping.

[0039] As a specific implementation, in the process of mapping generation and implicit step completion, a state machine containing nodes and transition edges is generated by a validated structured operation description unit based on a mapping rule base. The nodes of the state machine include operation nodes and confirmation nodes, as well as at least one of waiting nodes, branch nodes, parallel nodes, exception handling nodes, and manual step nodes. The transition edges between nodes represent at least one of execution order, condition fulfillment, timeout, parallel splitting and merging, and exception jump relationships. Based on domain completion rules, implicit steps not explicitly stated in the source text are identified, and the corresponding nodes are automatically completed.

[0040] The mapping rule base contains mapping rules from structured operation description units to state machine nodes and domain completion rules. The mapping rules are pre-configured by domain experts according to thermal power operation specifications and can be updated and modified in the future. For example, in the mapping of action types to node templates: the node template generated by the action type "start (pump / fan type)" is "operation node + operation feedback confirmation node + timeout handling node", the node template generated by the action type "open (electric door type)" is "operation node + open to position confirmation node + timeout handling node", the node template generated by the action type "adjust to target value" is "operation node + analog quantity reaches target confirmation node (AI comparison) + timeout handling node", and the node template generated by the action type "put into operation (automatic / interlocking)" is "operation node + put into operation status confirmation node".

[0041] Domain completion rules include the mapping from action type to confirmation signal type (start corresponds to running feedback DI, enable corresponds to on-position feedback DI, adjust to target value corresponds to the corresponding analog quantity reaching the target), the default timeout duration configured by device category, and the default jump strategy for abnormal handling nodes.

[0042] In the process of mapping structured operation description units to generate state machines, implicit steps for completing operation nodes are performed in two layers according to the domain completion rules in the mapping rule base: The first layer (implicit step identification) determines the confirmation steps, waiting steps, or timeout handling steps that are not explicitly stated in the source text but should exist according to the semantics of thermal power operation, based on the action type and equipment category of the operation node. The second layer (Completion and Source Tracing) generates corresponding confirmation nodes (confirming whether the action is in place), waiting conditions (waiting before reaching the goal), and timeout handling nodes (handling actions or confirmations that have timed out) for the identified implicit steps. It also records the application completion rule identifier, action type, device category, and corresponding source text location for each completion node.

[0043] Domain completion rules are pre-configured by domain experts based on thermal power plant operation specifications and stored in the mapping rule base, rather than being automatically generated by the system during the conversion process. These rules are configurable and can be updated in a versioned manner after review by control engineers, but remain fixed within the same conversion task. Completion based on these domain rules ensures that the generated state machine is semantically complete and executable in thermal power plant operation. This two-layer structure makes completion nodes traceable, enabling the automatic discovery and completion of implicit steps missing in the source text during the mapping from the specification text to the state machine, while also providing completion traceability.

[0044] Furthermore, the process of mapping to generate a state machine also includes mapping the functional group hierarchy of the text into a hierarchical state machine, expanding the functional group nodes of the main state machine into sub-state machines, and transmitting the termination conditions of the sub-state machines back to the main state machine as subsequent transition conditions.

[0045] As a specific example, such as Figure 3 As shown, a binding and disambiguation process is performed on the action items and condition items in the generated state machine. For action items, control commands with writable attributes are retrieved and bound from the control command information database; for condition items, measurement points with readable attributes are retrieved and bound from the measurement point information database. The writable / readable attributes are determined based on the signal type of the measurement point or command.

[0046] When multiple candidates exist, multi-candidate disambiguation is performed. Multi-candidate disambiguation involves matching and generating a set of candidates to be bound in the control command information database and the measurement point information database; filtering based on preset hard filtering conditions; weighting and sorting the candidates in descending order according to preset scoring indicators; and binding is successful when the highest score is unique and greater than or equal to the preset binding threshold, and the difference between the highest score and the second highest score is greater than or equal to the difference interval threshold.

[0047] For binding failures, the following criteria apply: if the candidate set to be bound is empty after hard filtering, it is judged as "no object to bind"; if the candidate set to be bound is not empty but the highest score is lower than the preset binding threshold, it is judged as "no valid binding"; if the difference between the highest score and the second highest score is less than the score difference interval threshold, it is judged as "multiple candidates undecided"; in addition, if the read / write attribute of a successfully bound candidate does not match the item to be bound (e.g., the action item can only match read-only measurement points, the condition item can only match write-only commands), it is judged as "attribute mismatch".

[0048] Based on the binding results, executability is marked. Nodes that fail to bind or whose measurement point / command attributes do not meet the requirements are automatically downgraded to manual step nodes, and the reason for downgrading is marked. Manual step nodes retain the migration relationship with the preceding and following nodes. This mechanism anchors the generated operation logic to the actual measurement point / command system of a specific unit and transforms the defects of incomplete or unimplementable generated logic into a controllable downgrading and explicit marking mechanism.

[0049] Furthermore, for the condition items bound to the measuring point, their comparison thresholds are parsed according to the following priority: explicit numerical values ​​in the procedure text, where the procedure directly provides a numerical value or range, the value is taken; matching of the setting sheet or interlocking protection setting library, searching the setting sheet / interlocking protection setting library by measuring point identifier or "equipment + parameter type" and taking the corresponding setting; historical operation data statistical baseline, taking the stable range of the measuring point in the corresponding process stage during historical successful operations as the threshold; if parsing fails, it is marked as a manually edited item.

[0050] The source level used for parsing is written into the conversion traceability record of the conversion report for engineers to review. For operation nodes with the action type of adjustment, the confirmation node is a condition item and bound to a readable measurement point. The comparison target value of the confirmation node is preferentially taken from the target parameter field value recorded in the structured operation description unit corresponding to the operation node (such as the water level setting value given by the procedure). Only when the target parameter field is a non-numerical expression or is defaulted will the value be parsed according to the aforementioned priority.

[0051] As a specific implementation, the multi-candidate disambiguation process when binding action items and condition items in the state machine is as follows.

[0052] Candidate generation. In the measurement point information database and control command information database, matching is performed using device codes or KKS-type coding rules, combined with matching of the normalized descriptive text of the operation object terms. The union of these matches is taken as the set of candidates to be bound. The above code matching can employ exact matching or fuzzy matching based on code similarity reaching a preset recall threshold. Descriptive text matching can use existing text similarity calculation methods such as edit distance, term overlap, or text vector similarity; candidates are included once the similarity reaches the preset recall threshold. This step aims for high recall and can directly use the aforementioned existing matching methods. Precise candidate determination and disambiguation are completed by subsequent hard filtering and soft scoring.

[0053] Hard filtering. The candidate set to be bound is filtered based on the process subsystem affiliation, signal type, read / write attributes, and dimensions, eliminating non-compliant candidates. Hard filtering is an item-by-item hard constraint; the candidate set to be bound must simultaneously satisfy all constraints to be retained. Violation of any constraint results in rejection. Specific rejection conditions are as follows: Candidates in the set whose process subsystem is inconsistent with the process subsystem of the item to be bound and do not belong to the preset allowed associated subsystems are rejected; candidate signal types do not match the requirements of the item to be bound (condition items require input signals AI / DI, action items require output signals AO / DO), are rejected; read / write attributes do not match the type of the item to be bound (condition items must be bound to readable attributes, action items must be bound to writable attributes), are rejected; for condition items involving numerical comparisons, candidate measurement points whose dimensions or ranges are inconsistent with the parameter type of the condition item (e.g., pressure conditions matching temperature or level measurement points), are rejected.

[0054] It should be noted that the above-mentioned elimination based on signal type and read / write attributes is aimed at candidates in the candidate set to be bound that are clearly marked and do not clearly match the item to be bound. For candidates in the candidate set to be bound that have missing, unmarked, or uncertain attribute fields, they are temporarily retained in the hard filtering stage and will be determined by the attribute back check stage after binding. If the back check confirms that its read / write attributes do not match the item to be bound, it is judged as "attribute mismatch" and enters node downgrading.

[0055] Soft scoring and ranking. The remaining candidates after hard filtering are weighted and ranked in descending order based on factors such as naming similarity, device hierarchy distance, upstream and downstream relationships, consistency of context-bound objects, and historical review records. Relevant data can come from the bound knowledge base or the engineer's review write-back process.

[0056] Judgment: If the candidate set to be bound after hard filtering is not empty, there is a unique highest score after soft scoring and its score is not lower than the binding threshold, and the difference between the highest score and the second highest score is not less than the score difference interval threshold, the binding is successful; if the difference between the highest score and the second highest score is less than the score difference interval threshold (there are similar competing candidates), it is judged as "multiple candidates undecided", that is, "binding is ambiguous", downgraded and the candidate list is retained, marked for engineer review; if the candidate set to be bound after hard filtering is empty, it is judged as "no binding object", downgraded to a manual step node, marked as a node to be supplemented with measurement points / commands. If the candidate set to be bound is not empty but the highest score is lower than the binding threshold, it is judged as "no valid binding"; in addition, for candidates that have been successfully bound according to the above judgment, attribute back-check must also be performed. If its read / write attribute determined by the signal type does not match the action item or condition item to be bound, it is judged as "attribute mismatch", and it is also downgraded to a manual step node and the reason for the downgrade is marked.

[0057] The scoring elements and weights are configurable. The recall threshold, binding threshold, and difference interval threshold are also pre-set configurable items and remain fixed within the same conversion task.

[0058] During technical verification, five hard disambiguation queries containing approximate interference candidates (each query had more than two approximate candidate measurement points in the database) were verified. All five queries were uniquely and correctly bound: for example, "circulating pump outlet main pipe pressure" correctly matched the circulating water system measurement point under the interference of approximate main pipe pressure measurement points in both open circulating water and closed cooling water systems, based on the process subsystem attribution scoring; "condensate pump operation" correctly matched the operating status DI point under the interference of approximate descriptions such as the sealing water solenoid valve. This indicates that the binding rate comes from the scoring judgment of the disambiguation mechanism rather than being a necessary result of the measurement point database configuration.

[0059] As a specific example, performing executability annotation on nodes based on the binding result includes at least the following: Action nodes are marked as candidates for automatic execution when they are bound to writable commands and meet their security attribute requirements. When an action node is not bound to a writable command, it is marked as a manual action or cannot be executed automatically. When a node is confirmed to be bound to a readable measurement point, it is marked as an automatic verification candidate. If a node is not bound, mark it as a point requiring manual confirmation or to be supplemented for measurement. When the binding is ambiguous and cannot be uniquely determined (multiple candidates pending), the candidate binding relationship is retained and marked as pending engineer review.

[0060] Degradation conditions include no bindable objects, no valid bindings, multiple undecided candidates, and attribute mismatches. Specifically, these include: an empty set of candidates to be bound after hard filtering; the highest score being lower than a preset binding threshold; the difference between the highest and second-highest scores being less than a score difference interval threshold; or a successfully bound candidate whose read / write attributes, determined by the signal type, do not match the action or condition to be bound. After degradation, the reason for degradation is marked. Manual step nodes are represented by specific node types in the state machine, retaining predecessor and successor migration edges. During the configuration assistance phase, engineers can replace manual step nodes with specific operations or confirm them as manually executed steps. After manual confirmation, the process continues via the successor migration edge.

[0061] The manual step nodes generated during degradation can be configured as APS breakpoint nodes (manual confirmation gates). Specifically, after the preceding automatic steps are completed, the state machine is suspended at this breakpoint node, and its suspended state and the content to be manually executed are output to the breakpoint field of the configuration auxiliary description file (to be represented as an APS breakpoint during configuration). After manual confirmation by engineers or operators, the process can continue through subsequent migrations. In the dual-source playback verification of historical event sequences, this breakpoint node participates in playback as a forced pause point: playback to the breakpoint node must match the corresponding manual confirmation event in the historical sequence to continue; otherwise, an inconsistent message "Manual confirmation missing at the breakpoint" is output. Therefore, APS breakpoint nodes are not just optional annotations, but a node type with definite behavior in both product output and playback verification.

[0062] As a specific implementation, a dual-source playback verification step is also included before generating the configuration auxiliary description file and conversion report: Acquire event sequences of historically successfully completed operations, including switch event records (SOE event records, including millisecond-level timestamps of device start-up, status status, and confirmation signals) and analog trend data (historical trend data, such as analog sampling sequences, including outlet pressure, deaerator water level, etc.). Events are associated with state machine nodes according to time sequence and based on the bound measurement point identifier; the migration of switch quantity nodes is driven by switch quantity events in time sequence; the analog quantity trend data is compared with the comparison threshold of the condition item within the corresponding event time window to determine the analog quantity condition; Check the preset validation items, output a list of inconsistent nodes and trace back to the source text entry.

[0063] Existing technologies often rely on engineers reviewing each line or on formal verification at the code level to determine the rationality of the generated control logic. They lack methods to verify the feasibility of the generated logic migration path and confirm the correctness of the sequence using the actual historical operation of the unit. To address this issue, this invention sets up a playback verification step, using bound measurement point identifiers as event-node association keys. It employs dual-source playback of switch event sequences and analog trend data, comparing analog conditions using thresholds within the event time window, considering sampling period alignment and dead zones, and tracing inconsistencies back to source text entries to form a corrective closed loop. The generated operational logic is anchored to the actual unit measurement point / command system (executability determination, disambiguation, and degradation labeling), implicit steps omitted in the regulations are supplemented according to thermal power industry rules, and the executability of the generated state machine is verified using event sequences and trend data from the unit's historical operation.

[0064] Specifically, the process of determining the analog quantity time window in this invention is as follows: if the analog quantity trend value does not reach the comparison threshold of the corresponding condition item within the time window (or only fluctuates within the dead zone), then the playback of the confirmation node is determined to be unsatisfactory, and the result is output to the list of inconsistent nodes (e.g., "the state machine requires condition X to be satisfied first at a certain confirmation node, but X appears after this step in the historical sequence") and backtracks to locate the corresponding source text entry, indicating that there may be an order problem in the generation logic at this point.

[0065] The preset verification items include: migration edge reachability, driven by historical events, whether the state machine can move from the initial node to the terminal node along the migration edge; confirmation condition satisfaction order, whether the confirmation signals of each confirmation node appear in the historical sequence in the order required by the state machine; timeout window in place, whether the waiting / timeout node is correctly triggered or in place within its time window; and abnormal jump matching, whether the abnormal jumps defined by the state machine match the triggering events in the historical sequence.

[0066] When a manual step node is configured as an APS breakpoint node, it participates in playback as a forced pause point. When playback reaches this node, it must match the corresponding manual confirmation event in the historical sequence before it can continue. Simulation event sequences or manually constructed test event sequences can be used as supplementary playback sources to cover paths not covered by historical data and output a prompt for supplementary verification.

[0067] As an optional embodiment, the generated configuration auxiliary description file can be converted into the import format of at least one target configuration tool according to preset conversion rules, so that engineers can import and edit it. The configuration auxiliary description file is not limited to the proprietary configuration format of a specific DCS manufacturer, and serves as an intermediate description file for conversion by the target configuration tool, review by engineers, or secondary editing.

[0068] In addition, the generated configuration auxiliary description file or candidate execution plan can be output to the operation sequence verification and admission system for structured verification, and items that fail the verification are marked as feedback to the conversion report of this invention.

[0069] Furthermore, a review result write-back step can be set up. Engineers review the fields, state machine nodes, test point / command binding relationships, degradation nodes, and replay verification issues for pattern validation. After versioning and recording, the results are written back to the template library (a step to assist in parsing and generating candidate structured operation description units), the pattern correction rule library (to assist in correcting structured description patterns), or the binding knowledge base. When the binding review results are written back to the binding knowledge base, they participate in the scoring as a "historical review record" element in the subsequent binding disambiguation soft scoring, allowing disambiguation accuracy to improve with accumulated reviews, with the two features mutually supporting each other.

[0070] Furthermore, when the procedure text corresponds to a low-load deep peak shaving operation, the operating condition category is identified during the parsing phase, and at least one of the following is added to the structured operation description unit: low-load operating condition marker, load change rate field, minimum equipment combination field, combustion stability confirmation field, environmental parameter confirmation field, and low-load safety confirmation node. During the mapping generation phase, the corresponding operating condition constraint rule set is loaded to participate in mapping generation and pattern verification. If a low-load safety confirmation node cannot be bound to a readable measurement point, it is marked as a measurement point to be supplemented or requiring manual confirmation.

[0071] A text-to-state machine system with programmable configuration assistance for thermal power units includes: The text acquisition module acquires the source text, as well as the device information database, measurement point information database, control command information database, structured description pattern, and mapping rule database. The parsing module parses the source text to generate candidate structured operation description units; The verification module performs pattern verification on candidate structured operation description units and generates feedback information or marks manually edited items for those that fail the verification. The mapping completion module generates a state machine by mapping the validated structured operation description units according to the mapping rule base, and identifies and completes the implicit content not explicitly expressed in the source text. Bind the disambiguation module, bind writable commands to action items, bind readable test points to condition items, and perform multi-candidate disambiguation; The degradation annotation module annotates nodes for executability based on the binding results, downgrades nodes that trigger degradation conditions to manual step nodes, marks the reasons for degradation, and retains their preceding and following migration edges. The playback verification module, based on the records of digital events and analog trend data of the historical successful operation process of thermal power units, performs dual-source playback verification of the state machine and outputs a list of inconsistent nodes. The generation module generates configuration auxiliary description files and conversion reports. The conversion report includes the backtracking of each node of the state machine to the source text entry, the structured operation description unit, and the applied mapping rules or domain completion rules. The review write-back module writes back the engineer's review results to the template library, pattern correction rule library, or binding knowledge base after versioning and recording.

[0072] The following example illustrates the specific process of this invention by converting the sequential control procedure of a first-stage auxiliary machine of a supercritical coal-fired power unit. This embodiment covers three mapping paths: switch quantity action, analog quantity adjustment, and parallel step.

[0073] Step 1: Data Acquisition. Acquire the relevant procedure text, equipment information database, measurement point information database, control command information database, structured description schema, and mapping rule database.

[0074] Step 2: Generation of candidate structured operation description units. For the procedure entry "Start condensate pump A, confirm that its outlet pressure is established, open the recirculation valve; simultaneously start condensate pump B; adjust the deaerator water level to the target value", candidate structured operation description units are generated through the rule template channel and / or language model channel.

[0075] Unit U1, Operation Object = Condensate Pump A, Action Type = Start-up, Abnormal Handling = (Not Available); Unit U2, Confirmation Condition = Condensate Pump A Outlet Pressure Established; Unit U3, Operation Object = Condensate Pump A Recirculation Valve, Action Type = Open, Execution Prerequisite = U2; Unit U4, Operation Object = Condensate Pump B, Action Type = Start-up, Parallel to Unit U1; Unit U5, Operation Object = Deaerator Water Level, Action Type = Adjust to Target Value, Target Parameter = Water Level Setpoint.

[0076] The structured operation description unit is constrained by the structured description pattern. The value range, necessity, and reference object type of each field are all specified by the structured description pattern. The test point / command reference field is filled in during the binding stage and is a placeholder for binding during the generation stage.

[0077] Step 3, Schema Validation and Feedback. Validate according to the structured description schema: Field completeness: If the exception handling field of U1 is missing, the verification prompt should be to fill in the default or mark it; Reference object consistency: U2 "outlet pressure" must be able to correspond to the pressure type measuring point in the measuring point library, and U5 "water level set value" must be able to correspond to the target parameter; Exception branch completeness: If a unit contains "If an exception occurs, then handle it" but lacks the exception trigger condition or jump target, then the exception branch is judged to be incomplete; Step dependency legality: If U3 references U2 in the premise, the dependency is acyclic and the premise is reachable, then the verification passes; U1 and U4 are marked as parallel and have no mutual dependency, then the verification passes.

[0078] If a unit is generated through the language model channel but fails to be validated (e.g., the action type value is not within the value range), the reason for the failure will be fed back, and the candidate generation stage will regenerate or roll back the rule template channel; if it still fails after repeating the preset number of times, it will be marked as a manually edited item, and the pattern validation in this step is a structural validation of the semantic level of thermal power operation.

[0079] Step 4: Mapping Generation and Implicit Step Completion. The validated structured operation description units are mapped into a state machine based on the mapping rule base. The implicit steps are then completed as follows: For U1 "Start Condensate Pump A", the confirmation node "Confirm Condensate Pump A Operation Feedback" is automatically generated by mapping "Action Type = Start → Confirmation Signal Type = Operation Feedback" and the timeout handling node is generated by mapping "Default Timeout for Pump Equipment = Tb seconds".

[0080] For U3 "Open recirculation door", an automatic "Confirm recirculation door is open in place" confirmation node and timeout handling are generated.

[0081] For U4 "Start condensate pump B", a parallel node and a merge node are generated: U1 and U4 are parallel branches, and after the operation feedback confirmation of both is satisfied, they continue through the merge edge.

[0082] For U5 "Deaerator water level adjusted to target value", an analog quantity comparison confirmation node is generated according to "Action type = Adjustment → Confirmation signal type = Corresponding analog quantity reaches target (AI comparison)". The comparison target value of this confirmation node is preferentially taken from the target parameter field of the structured operation description unit (here, the water level setpoint given by the procedure); only when the target parameter field is a non-numerical expression such as "normal" or "stable" or is the default, the comparison threshold is resolved by the threshold resolution priority chain.

[0083] Each completion node records the completion traceability, including the applied completion rule identifier, action type, device category, and corresponding source text location (forming part of the transformation traceability record).

[0084] Step 5: Bind measurement points to commands, disambiguate, mark executability and downgrade.

[0085] The U1 action "Start Condensate Pump A" is bound to a writable command (such as the start / stop command for condensate pump A) in the control command information library; its confirmation node "Operation Feedback" is bound to a readable status measurement point (the DI point of condensate pump A's operating status) in the measurement point information library; if both are successfully bound, the action node is marked "Automatic Execution Candidate" and the confirmation node is marked "Automatic Verification Candidate".

[0086] The U2 condition "Outlet pressure established" is bound to a readable pressure measurement point (condensate pump A outlet pressure AI point).

[0087] The U5 condition "Deaerator water level reaches target" is bound to a readable water level measuring point (deaerator water level AI point), and the target parameter is obtained by comparing the threshold.

[0088] Assuming that the U3 "recirculation gate" has no writable instructions in the current command library (e.g., if the gate is a manual gate), the operation node is automatically downgraded to a manual step node, marked with the downgrade reason "no bound writable command," while retaining its transition relationship with preceding and following nodes. The downgraded state machine segment is: Preceding confirmation node (condensate pump A outlet pressure established, marked as automatic verification candidate) → Manual action node (open recirculation gate, downgrade reason: no bound writable command) → Manual confirmation node (confirm recirculation gate is fully open) → Subsequent node (continue). This embodiment does not simply mark "unexecutable," but transforms unexecutable steps into reviewable and continueable manual nodes, while retaining the transition connection with preceding and following nodes. If a condition item has multiple candidate measurement points, it enters the multi-candidate disambiguation scoring stage.

[0089] Step 6: Product and Transformation Report Generation. Generate configuration auxiliary description files (including state machine nodes, migration edges, bound measurement points / commands, executability annotations, and degradation markers) and / or candidate execution plans, as well as measurement point binding tables, command binding tables, and lists awaiting manual confirmation; generate transformation reports, where each node contains transformation traceability records (source text entries, structured description units, and applied mapping rules / completion rules).

[0090] In the actual verification process, the prototype system generated outputs from 11 real procedures; the measurement point / command library was constructed based on the anonymized configuration of the real equipment in the procedures; the data is reproducible. A total of 152 candidate structured operation description units were generated. The initial verification interception rate during the pattern validation phase was 15.1% (intercepting 23 candidates with non-compliant structures, including 21 complex compound sentences and 10 missing fields, etc.). After feedback and repair, the pass rate increased to 86.2%, and manually edited items were marked at 13.8%. There were 315 source text nodes, including 121 explicit confirmation nodes (80 in-place confirmation, 36 other confirmations, and 5 operation feedback confirmations), while there were 0 timeout handling nodes. This method automatically completed 155 timeout handling nodes, accounting for 33.0% of the final state machine nodes (470), indicating that the procedures generally involve writing actions and confirmations. However, the system systematically omits timeout handling; there are 287 items to be bound (142 action items and 145 condition items), with a binding rate of 78.7%, an automatic decision rate of 78.7%, a multi-candidate unresolved rate of 9.1%, and a downgrade rate of 12.2% (downgraded items were all verified to be manual devices such as steam traps without remote operation conditions, and the downgrade judgment was correct). The proportion of nodes that can be automatically handled (automatic execution candidates + automatic verification candidates) is about 79%; all 11 procedures have been converted and connected end-to-end, and the prototype produces a draft of the state machine to be reviewed for a single procedure in seconds, while the existing manual process of sorting out the equivalent configuration logic of each procedure usually takes man-days, which is a significant improvement compared to the existing technology.

[0091] In contrast, the method described in the background directly generates the state machine from the same procedure text using a language model, without pattern verification, implicit step completion, measurement / command binding, or executability annotation—the latter directly generating control logic from the language model. This method results in zero timeout handling nodes, zero measurement / command bindings, and zero executability annotations. For branch / parallel structured structures, the "OR" branch is simply merged into the node name without expansion; and zero transformation traceability records. While the generated product structure is readable on the surface, it lacks timeout handling, measurement / command binding, executability differentiation, and transformation traceability, making it impossible to determine its executability on a specific unit. In contrast, the method of this invention anchors the generated logic to a real measurement / command system, makes executability verifiable, and allows for dual-source playback verification of historical events, significantly improving the structural integrity, executability, and reviewability of configuration assistance.

[0092] The above embodiments are further elaborations and descriptions of the present invention to facilitate understanding, and are not intended to limit the present invention in any way. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for converting text into a state machine with the aid of programmed configuration of a thermal power unit, characterized in that, include: Parse the source text to generate candidate structured operation description units, and perform pattern verification according to the preset structured description pattern; The structured operation description unit is verified, and a state machine containing nodes and transition edges is generated according to the mapping rule library. Based on the domain completion rules, the implicit steps not explicitly stated in the source text are identified, and the corresponding nodes are automatically completed. The action items in the state machine are retrieved from the control command information base and bound to control commands with writable attributes, and the condition items are retrieved from the measurement point information base and bound to measurement points with readable attributes. The binding process includes multi-candidate disambiguation. Multi-candidate disambiguation involves matching and generating a set of candidates to be bound from the control command information database and the measurement point information database. After filtering based on preset hard filtering conditions, the scores are weighted according to preset scoring indicators and sorted in descending order; the binding is successful when the highest score is unique and greater than or equal to the preset binding threshold, and the difference between the highest score and the second highest score is greater than or equal to the score difference interval threshold. Based on the binding results, the nodes are marked as executable. Nodes that trigger the degradation conditions are downgraded to manual step nodes that retain the connection between the previous and subsequent migrations. Generate configuration auxiliary description files and conversion reports.

2. The method of claim 1, wherein the method further comprises: It also includes a dual-source playback verification step before generating the configuration auxiliary description file and conversion report: Acquire the event sequence of successfully completed historical operations, including digital event records and analog trend data; Events are associated with state machine nodes based on the bound measurement point identifiers; switch quantity events drive the migration of switch quantity nodes in time sequence; The analog quantity condition is determined by comparing the analog quantity trend data with the comparison threshold of the condition item within the corresponding event time window. Check the preset validation items, output a list of inconsistent nodes and trace back to the source text entry.

3. The text-to-state machine method with the auxiliary sequence control of the thermal power generating unit according to claim 1 or 2, characterized in that, The generated candidate structured operation description unit is implemented by a rule template channel, a language model channel, or a combination of both. For candidate structured operation description units that fail the verification, feedback information carrying the reason for the failure is generated and fed back to the generation stage for regeneration or rolled back to the rule template channel. Items that fail to regenerate after a preset number of attempts are marked as manually edited.

4. The method of claim 1, wherein the method further comprises: The nodes of the state machine include operation nodes and confirmation nodes, as well as at least one of waiting nodes, branch nodes, parallel nodes, exception handling nodes, and manual step nodes; the migration edges between nodes represent at least one of the following relationships: execution order, condition satisfaction, timeout, parallel splitting and merging, and exception jump.

5. A text-to-state machine method for programmable configuration assistance of thermal power units according to claim 1 or 4, characterized in that, The domain completion rules include the mapping from action type to confirmation signal type, the default timeout duration configured according to device category, and the default jump strategy for abnormal handling nodes; The automatic completion of the corresponding nodes involves generating corresponding confirmation nodes or timeout handling nodes based on the identified implicit steps, and recording the applied completion rule identifier, action type, device category, and corresponding source text position for each completion node.

6. A text-to-state machine method for programmable configuration assistance of thermal power units according to claim 1, 2, or 4, characterized in that, The hard filtration conditions are: If the process subsystem to which a candidate belongs in the candidate set to be bound is inconsistent with the process subsystem to which the item to be bound belongs, and it does not belong to the preset allowed associated subsystem, it will be removed; Candidate signals whose types do not match the requirements of the item to be bound are rejected. Candidates whose read / write attributes do not match the type of the item to be bound are removed; For conditional items involving numerical comparisons, candidates whose dimensions or ranges are inconsistent with the parameter type of the conditional item should be eliminated. The elimination based on signal type and read / write attributes targets candidates in the candidate set to be bound that are clearly marked and do not clearly match the item to be bound. Candidates with missing, unmarked, or uncertain attribute fields are temporarily retained and their attributes are checked after successful binding.

7. The text-to-state machine method for programmable configuration assistance of thermal power units according to claim 6, characterized in that, The step of marking nodes as executable based on the binding result includes at least the following: Action nodes are marked as candidates for automatic execution when they are bound to writable commands and meet their security attribute requirements. When a node is confirmed to be bound to a readable measurement point, it is marked as an automatic verification candidate. When the binding is ambiguous and cannot be uniquely determined, retain the candidate binding relationship and mark it as pending engineer review.

8. A text-to-state machine method for programmable configuration assistance in thermal power units according to claim 6, characterized in that, The downgrade conditions include: The set of candidates to be bound after hard filtering is empty; or The highest score is lower than a preset binding threshold; or The difference between the highest and second-highest scores is less than the score difference interval threshold; or After a successful binding of a candidate, the read / write attribute determined by the signal type does not match the action or condition item to be bound after the attribute check is performed. Mark the reason for the downgrade after downgrading.

9. A text-to-state machine method for programmable configuration assistance of thermal power units according to claim 1 or 4, characterized in that, The process of mapping to generate a state machine also includes mapping the functional group hierarchy of the text to a hierarchical state machine, expanding the functional group nodes of the main state machine into sub-state machines, and sending the termination conditions of the sub-state machines back to the main state machine as subsequent transition conditions. Configure the manual step nodes generated by the downgrade as APS breakpoint type nodes. Suspend them after the previous node is completed, output the suspension status and the content to be manually executed to the breakpoint field of the configuration auxiliary description file, and continue the process through subsequent migration edges after manual confirmation.

10. A text-to-state machine method for programmable configuration assistance of thermal power units according to claim 6, characterized in that, For conditional items bound to readable measurement points, their comparison thresholds are parsed according to the following priority: explicit values ​​in the source text; set values ​​obtained by matching the set value sheet or interlocking protection set value library; historical operation data statistical baseline; if none of these can be determined, they are marked as manually edited items; the source level used for parsing is written into the conversion traceability record in the conversion report. For operation nodes whose action type is adjustment, their confirmation nodes are condition items and bound to readable measurement points. The comparison target value of the confirmation node is preferentially taken from the target parameter field value recorded in the structured operation description unit corresponding to the operation node. Only when the target parameter field is a non-numerical expression or is defaulted, the comparison threshold is parsed according to priority.

11. A text-to-state machine system assisted by programmable configuration of a thermal power unit, applicable to the text-to-state machine method as described in any one of claims 1-10, characterized in that, include: The parsing module parses the source text to generate candidate structured operation description units; The verification module performs pattern verification on candidate structured operation description units; The mapping completion module maps the validated structured operation description units to generate a state machine, completing the implicit content not explicitly stated in the source text; The binding disambiguation module retrieves and binds the action items in the state machine to control commands with writable attributes from the control command information base, and retrieves and binds the condition items to measurement points with readable attributes from the measurement point information base. The binding process includes multi-candidate disambiguation. Multi-candidate disambiguation involves matching and generating a set of candidates to be bound from the control command information database and the measurement point information database. After filtering based on preset hard filtering conditions, the scores are weighted according to preset scoring indicators and sorted in descending order; the binding is successful when the highest score is unique and greater than or equal to the preset binding threshold, and the difference between the highest score and the second highest score is greater than or equal to the score difference interval threshold. The labeling and degradation module labels nodes as executable based on the binding results. Nodes that trigger degradation conditions are downgraded to manual step nodes that retain the connection between the previous and subsequent migrations. The generation module generates configuration auxiliary description files and conversion reports.

Citation Information

Patent Citations

  • Model-based code generation method and related device

    CN120560623A