Training behavior real-time analysis and feedback system for gas turbine power plant safety education

Abstract

The invention relates to the technical field of analysis and feedback, in particular to a gas turbine power plant safety education-oriented training behavior real-time analysis and feedback system, which comprises a task node monitoring module, a behavior stability description module, a task deviation judgment module, a path misunderstanding extraction module and a correction guide generation module. According to the method, behavior differences in different stages are analyzed, error zone path segments with path number repetition frequencies and response delay indexes higher than threshold values are screened, and node offset levels and action differences are integrated to generate an error zone aggregation segment group; according to the method, wrong operation nodes are positioned, a prompt set with sequence suggestions, instruction content and operation duration correction information is formed, a feedback list for training field interaction prompt is constructed based on the content, the traceability of a behavior misunderstanding and the timeliness of correction intervention are enhanced, the path normalization and task execution accuracy in the training process are improved, and the training efficiency is improved. The curing risk of path deviation is reduced, and the forming efficiency of a training effect closed loop is improved.

Description

Technical Field

[0001] This invention relates to the field of analysis and feedback technology, and in particular to a real-time analysis and feedback system for training behaviors in safety education for gas turbine power plants. Background Technology

[0002] The field of analysis and feedback technology involves the real-time acquisition, processing, modeling, and feedback control of behavioral data, with the core objective of achieving efficient evaluation and dynamic intervention of human behavioral processes. This field integrates multiple technical approaches, including rule-based behavior modeling, machine learning-driven behavior recognition algorithms, data feedback mechanisms in human-computer interaction, behavior deviation detection technology, and response mechanisms that dynamically adjust content or strategies based on feedback results. The systems typically rely on behavior monitoring terminals, data analysis modules, and feedback presentation components. They can continuously track, analyze deviations, and evaluate results in various scenarios such as training, education, production operations, medical rehabilitation, and traffic management, and achieve model-driven personalized feedback to improve the accuracy of behavior execution and the achievement rate of system goals. The technical approaches emphasize the combination of data-driven approaches and human factors engineering, requiring the system to be real-time, accurate, and operable.

[0003] The real-time analysis and feedback system for safety education in gas turbine power plants aims to improve the behavioral standardization and knowledge mastery of gas turbine power plant operators during safety training. Specifically, it uses behavioral monitoring, such as eye tracking, operation records, and interactive Q&A, to analyze in real time the operational paths, decision-making, and knowledge responses of operators during training. Based on the analysis results, it provides immediate feedback, such as error warnings, reinforcement prompts, and operational optimization suggestions, forming a quantifiable, assessable, and traceable training effectiveness closed loop. This enhances the relevance and effectiveness of the training process, reduces the risk of human error, and improves the overall operational safety level of the power plant.

[0004] Traditional analysis systems have shortcomings in identifying and intervening in recurring errors in work paths. When the same operator repeatedly triggers similar erroneous paths in multiple rounds, traditional systems only record surface-level deviation information. They lack aggregate analysis of node number offsets and reaction time distributions within the path segments, failing to accurately extract high-frequency error segments in the path and establish corresponding differences with task objectives. This results in feedback that is biased towards static prompts, making it impossible to achieve dynamic correction and precise intervention. For example, in load switching tasks, if the operation sequence deviates multiple times but is misjudged as an isolated event, it will lead to a lack of clear judgment on path error patterns, delaying the opportunity for behavioral correction, affecting training quality, and potentially creating operational risks. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a real-time analysis and feedback system for safety education in gas turbine power plants.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a real-time analysis and feedback system for safety education in gas turbine power plants, the system comprising:

[0007] The task node monitoring module acquires the operation process information of safety education and assessment, records the click number of each operator's instruction button on the operation panel, operation duration and task status switching sequence, and categorizes them by task cycle to generate operation node sequence behavior records.

[0008] The behavior stability characterization module calculates the stability score of the operation based on the operation node sequence behavior record, determines the stable region and the unstable region according to the stability evaluation interval, and generates task node behavior stability distribution information.

[0009] The task deviation judgment module obtains the path sequence number and result value offset level of the node operation performed by the operator in each stage based on the stable distribution information of the task node behavior, and establishes a comparison relationship based on the target correspondence table in the training instruction library to generate a list of node path and result deviation records.

[0010] The path error extraction module calls the list of node path and result deviation records, extracts the reaction delay time, operation object code and repetitive operation cycle of the action trigger in the path segment, aggregates similar paths and sorts them according to the repetitive trigger trend, and generates a group of operation behavior path error clusters.

[0011] As a further aspect of the present invention, the operation node sequence behavior record includes task cycle number, state switching sequence, click number set, duration distribution and panel response identifier. The stable distribution information of task node behavior specifically includes rhythm-stable node partition, rhythm-fluctuating node partition, node number and corresponding stability mapping table and rhythm-abnormal node index set. The node path and result deviation record list includes offset path number table, target number difference item, deviation level identifier group, behavior path misjudgment label and task target response offset record. The operation behavior path misconception aggregation segment group specifically refers to the misconception starting node set, repeated trigger path template, path aggregation number group, action instruction change mapping and misconception segment sorting index.

[0012] As a further aspect of the present invention, the task node monitoring module includes:

[0013] The task information extraction submodule obtains the operation process information of safety education and assessment. Based on the task process information of gas turbine combustion adjustment task, cold start task and load switching task, it collects the number index, instruction trigger position number and corresponding task start and end node number of each function button in the task operation panel. It establishes a correspondence between the function buttons and status numbers in each type of task and generates task node index structure information.

[0014] The task record organization submodule records the click number of the operation panel button and the task status switch number of the operator based on the task node index structure information. It extracts the duration value of the corresponding operation segment and the task instruction response flag bit according to the time sequence of each round of task operation process. It merges the instruction execution duration and status number to generate task operation behavior data information.

[0015] The behavior sequence classification submodule extracts the operation number, click location number, and status switch sequence number of each operator within the task cycle based on the operation behavior data information. It then organizes and divides the operation sequences within continuous task segments into cycles, classifies the operation sequences within a cycle by number, assigns task stage labels, and generates operation node sequence behavior records.

[0016] As a further aspect of the present invention, the behavior stability characterization module includes:

[0017] The duration difference calculation submodule, based on the operation node sequence behavior record, groups the duration sequence corresponding to each node according to the task round, according to the dwell duration sequence of each task node, calculates the dwell duration variance value within each node group, and obtains the node dwell difference information.

[0018] The switching mean extraction submodule extracts the task state switching time period value between adjacent nodes based on the node dwell difference information, groups the switching time corresponding to each type of node in the same round, calculates the mean of each group of time period values, and generates node switching mean data.

[0019] The stable distribution assessment submodule calls the node switching mean data, combines it with the variance value corresponding to each node in the node dwell difference value, calculates and obtains the node rhythm fluctuation score, sorts them in ascending order by node number, sets a rhythm stability assessment interval threshold, determines whether the score value falls within the interval range, marks nodes that exceed the assessment interval threshold as fluctuating nodes, and marks the rest as stable nodes, establishes a node stability and instability identification table, and generates stable distribution information of task node behavior.

[0020] As a further aspect of the present invention, the formula for calculating and obtaining the node rhythm fluctuation score is specifically as follows:

[0021]

[0022] Among them, S n V represents the rhythm fluctuation score of the nth task node. n M represents the variance of the dwell time at the nth node. n Z represents the average handover time of the nth node, α is the handover time offset correction constant, and Z i This represents the start time of the node's operation in the i-th round. The value represents the arithmetic mean of the start times of all rounds of operations for a node, where N represents the number of rounds the node participates in, β is the behavioral complexity adjustment factor, and C... n This represents the number of path branches of the nth node in the current task flow.

[0023] As a further aspect of the present invention, the task deviation judgment module includes:

[0024] The path number extraction submodule extracts the task node numbers marked as unstable areas based on the stable distribution information of the task node behavior, obtains the operation path sequence numbers recorded by the operator when performing the corresponding node operations in multiple training stages, and combines and marks them according to the node number and stage number to generate node path number combination information.

[0025] The parameter difference calculation submodule calls the node path number combination information, extracts the trigger time point and target parameter number of each instruction in the path, calculates the difference sequence for the instruction number and parameter number in adjacent stage paths under the same node, performs in-group averaging on the difference sequence, and obtains path instruction and parameter offset information.

[0026] The comparison result generation submodule, based on the path instructions and parameter offset information, calls the target comparison table composed of the corresponding task target number and path instruction number in the training instruction library, extracts the records with a difference greater than the set path error threshold and adds an offset level label, establishes a mapping structure between the corresponding node number and the offset level, and generates a list of node path and result deviation records.

[0027] As a further aspect of the present invention, the path error extraction module includes:

[0028] The high-frequency path filtering submodule calls the list of node paths and result deviation records, filters records with path number repetition rate greater than three times, detects whether the action offset level is higher than the upper limit of the allowable offset level of the task target, selects the path number and node number combination that meet the conditions, and generates high offset repetitive path information.

[0029] The behavior parameter extraction submodule extracts the trigger time, operation object code and repeated call cycle of the object in the task phase for each action node in the path segment based on the high offset repeated path information, and aligns the action execution records of multiple time periods under the same node to generate a set of repeated behavior parameters.

[0030] The path aggregation and sorting submodule calls the set of repeated behavior parameters to classify the set of action nodes with the same operation object code and path segment number. It calculates the number of repeated calls and the time series fluctuation value of the trigger interval for each type of path node, and sorts them in descending order with the number of repeated triggers as the main order. It establishes a matching table between the clustered path segments and the sorting number, and generates a group of error clusters in the operation behavior path.

[0031] As a further aspect of the present invention, the system further includes:

[0032] The correction guidance generation module gathers paragraph groups based on the errors in the work behavior path, calls the correct operation path corresponding to the node in the standard task flow, filters out erroneous operation nodes that do not match the task objective, generates a set of instruction content for each erroneous node with execution order suggestions, objective-corresponding prompt information and operation duration correction prompts, establishes feedback push content for training site prompts, and generates a list of training task correction prompt instructions;

[0033] The training task correction prompt instruction list includes a chain of correction action numbers, sequence adjustment prompts, duration correction markers, task node target reminders, and interactive push information formats.

[0034] As a further aspect of the present invention, the correction guide generation module includes:

[0035] The path information extraction submodule gathers paragraph groups based on the operational behavior path errors, extracts the trigger action number of the first operation node in the paragraph, the corresponding operation feedback delay time value, and the repeated trigger time interval sequence in multiple rounds of training, archives and classifies each parameter item according to the combination of path number and node number, and generates node behavior path attribute information.

[0036] The node deviation filtering submodule calls the node behavior path attribute information to obtain the correct click number, preset operation duration and task panel position number in the standard task path corresponding to each node. It compares these with the click number, operation time and interface position coordinates of the current node, calculates the node execution error degree value, sets the deviation identification threshold, filters nodes with scores greater than the deviation identification threshold as deviation nodes, establishes a node number and deviation label comparison table, and generates error task node discrimination information.

[0037] The feedback instruction construction submodule, based on the error task node identification information, extracts the occurrence position of the node in each round according to the original task path order based on the selected node number, and combines the operation number with the target task parameter correspondence table to generate the execution order suggestion, parameter comparison prompt and operation time correction statement set required for each node. These are integrated into a structured push content set, establishing a mapping relationship between the task path and the push content, and generating a training task correction prompt instruction list.

[0038] As a further aspect of the present invention, the formula for calculating the node execution error level is specifically as follows:

[0039]

[0040] Among them, E j B represents the execution error level of the j-th task node. j Indicates the current clicked number, R j Indicates the standard click number, T j This represents the duration of the click on the j-th node. P represents the average click time for all rounds of the j-th node, γ is the adjustment factor for duration error, and P j σ represents the difference in index between the current clicked position of the j-th node and the standard panel position, δ is the adjustment factor for panel layout deviation, and σ is the index difference. j This represents the standard deviation of the click duration of the j-th node.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0042] In this invention, by simultaneously extracting instruction numbers, durations, and state switching sequences from the task sequence during the recording of the operation process, a cross-round behavioral node stability evaluation standard is established. By analyzing behavioral differences in different stages, a node deviation trajectory comparison system is established by combining the number difference and trigger time point. Error path segments with path number repetition frequency and reaction delay index exceeding the threshold are screened. Node offset level and action difference are integrated to generate error cluster segment groups. Based on the number click and time sequence information of the standard process, erroneous operation nodes are located and a prompt set with sequence suggestions, instruction content, and operation duration correction information is formed. Based on the content, a feedback list for interactive prompts in the training session is constructed, enhancing the ability to trace the source of behavioral errors and the timeliness of corrective intervention, improving the path standardization and task execution accuracy in the training process, reducing the risk of path deviation becoming entrenched, and improving the efficiency of forming a closed loop of training effectiveness. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a system flowchart of the present invention;

[0045] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0046] Figure 3 This is a flowchart of the task node monitoring module of the present invention;

[0047] Figure 4 This is a flowchart of the behavior stability characterization module of the present invention;

[0048] Figure 5 This is a flowchart of the task deviation judgment module of the present invention;

[0049] Figure 6 This is a flowchart of the path error extraction module of the present invention;

[0050] Figure 7 This is a flowchart of the correction guide generation module of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0052] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0053] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0054] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0055] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0056] Please see Figure 1 A real-time analysis and feedback system for training behaviors in safety education for gas turbine power plants. The system includes a task node monitoring module, a behavior stability characterization module, a task deviation judgment module, a path error extraction module, and a correction guidance generation module.

[0057] The task node monitoring module acquires the operation process information of safety education and assessment, including gas turbine combustion adjustment tasks, cold start tasks and load switching tasks. It records the click number of each operator's instruction button on the operation panel, operation duration and task status switching sequence, and classifies them by task cycle to generate operation node sequence behavior records.

[0058] The behavior stability characterization module is based on the operation node sequence behavior record. It calls the dwell time sequence and switching time value recorded in each task node, calculates the variance of dwell time and the mean of switching interval for each type of node according to the training round, calculates the operation stability score, sorts and marks the nodes by node number, and determines the stable area and unstable area according to the stability assessment interval, generating task node behavior stability distribution information.

[0059] The variance of dwell time indicates the dispersion of the duration of an operator's operation at a certain task status node; the mean of switching interval indicates the average time taken to switch from the current task status node to the next node; the ratio of the above two indicators can reflect the stability of the switching rhythm and is often used in behavioral rhythm analysis to evaluate the continuity and stability of operation.

[0060] The task deviation judgment module extracts the node numbers marked as unstable areas based on the stable distribution information of task node behavior, obtains the path sequence number and result value offset level of the node operation performed by the operator in each stage, records the difference between the instruction trigger time point in the path and the target parameter number of the executed action, and establishes a comparison relationship based on the target correspondence table in the training instruction library to generate a list of node path and result deviation records.

[0061] The result value offset level is a classification level of the deviation between the standard task completion parameters and the actual operation results, which is commonly found in the graded evaluation mechanism of training and assessment systems; the target parameter number difference refers to the difference between the parameter number selected by the operator and the training target parameter number, which is used to judge the accuracy of target achievement.

[0062] The path error extraction module calls the list of node path and result deviation records, filters records with path number repetition rate more than three times and action deviation level at the upper limit of the task target allowable range, extracts the reaction delay time, operation object code and repetition operation cycle of the action in the path segment, aggregates similar paths and sorts them according to the repetition trigger trend, and generates operation behavior path error cluster segment group.

[0063] Path number repetition rate refers to the frequency with which the same erroneous path is repeatedly executed in multiple rounds of training, and is a measure of the degree of solidification of behavioral patterns; reaction delay time comes from the time interval between instruction input and system feedback, and is often used for operational response evaluation.

[0064] The correction and guidance generation module aggregates paragraph groups based on the errors in the work behavior path, extracts the trigger action number, operation feedback delay value and repeated trigger interval sequence of the starting node in each path, calls the correct operation path corresponding to the node in the standard task flow, compares the difference of the click instruction number, the operation duration sequence and the distance of the task panel position number of each node in turn, filters out the erroneous operation nodes that do not match the task goal, generates a set of instruction content for each erroneous node with execution order suggestions, target corresponding prompt information and operation duration correction prompts, establishes feedback push content for training site prompts, and generates a list of training task correction prompt instructions;

[0065] The click command number difference is used to measure the degree of operation deviation and is related to the difference between the actual click command and the correct command number; the position number distance is the spatial distribution difference of control elements in the task panel and is often used to evaluate the cognitive burden in operation guidance strategies and interface layout design.

[0066] The operation node sequence behavior record includes task cycle number, state switching sequence, click number set, duration distribution, and panel response identifier. The stable distribution information of task node behavior specifically includes rhythm-stable node partitions, rhythm-fluctuating node partitions, node number and corresponding stability mapping table, and rhythm-abnormal node index set. The node path and result deviation record list includes offset path number table, target number difference item, deviation level identifier group, behavior path misjudgment label, and task target response offset record. The operation behavior path misconception cluster segment group specifically refers to the misconception starting node set, repeated trigger path template, path aggregation number group, action instruction change mapping, and misconception segment sorting index. The training task correction prompt instruction list includes correction action number chain, sequence adjustment prompt statement, duration correction mark, task node target reminder content, and interactive push information format.

[0067] Please see Figure 2 and Figure 3 The task node monitoring module includes a task information extraction submodule, a job record organization submodule, and a behavior sequence classification submodule.

[0068] The task information extraction submodule obtains the operation process information of safety education and assessment. Based on the task process information of gas turbine combustion adjustment task, cold start task and load switching task, it collects the number index, instruction trigger position number and corresponding task start and end node number of each function button in the task operation panel. It establishes a correspondence between the function buttons and status numbers in each type of task and generates task node index structure information.

[0069] For the task flow information of gas turbine combustion adjustment, cold start, and load switching tasks, the node flowchart numbers in the task document structure definition section, the status identifiers in the task stage description table, and the instruction name indexes in the function action description list are extracted. Specifically, the starting node for the combustion adjustment task is set as "Ignition Pre-detection," and the ending node is "Combustion Stability Confirmation," with function button numbers F001 to F010. The starting node for the cold start task is "Hydraulic Pressure Initialization," and the ending node is "Main Shaft Temperature Rise Confirmation," with corresponding function button numbers F011 to F020. The starting node for the load switching task is "Load Connection Preparation," and the ending node is "Load Stability Adjustment," with function button numbers F021 to F030. When collecting the function button numbers in the above task operation panel, the sequence number, function logic number, and trigger response number of the buttons in the operation interface are cross-referenced. A mapping table is established between the task stage name corresponding to each function button and its internal system node number. For the collection of instruction trigger position numbers, the pixel coordinates of the instruction input events in the operation log are read and... It is normalized to interface partition numbers, for example, (210, 450) is mapped to segment A3, and (880, 122) is mapped to segment D1. Further, based on the node structure marked in the task flow diagram, the button numbers and status numbers are compared in position. The mapping relationship between function number F004 and status number S2 is recorded in the structure table. During the above operations, a two-dimensional mapping array is established with the task number as the index field, and the fields "Task Function Index" and "Task Status Number" are established in the structure table to correspond, so that each type of task can be displayed on the operation panel. A clear set of button-to-status number mappings is formed. In the threshold field setting, the completeness evaluation threshold of the task function mapping relationship is set to θ = 90%. When the success rate η of the function button mapping in the task satisfies η ≥ θ, the task process is considered to have the conditions for index structure extraction. This value comes from the comparison result of the balance between the number of task buttons (10, 10, 10) and the number of task status numbers (12, 11, 13) in each of the three task modules. In the actual test, the average mapping success rate of each task is about 92%, which is higher than the set value. Finally, the task node index structure information is obtained.

[0070] The task record organization submodule records the click numbers of the operation panel buttons and the task status switch numbers of the operators based on the task node index structure information. It extracts the duration value of the corresponding operation segment and the task instruction response flag bit according to the time sequence of each round of task operation process, merges the instruction execution duration and status number, and generates task operation behavior data information.

[0071] Based on the task node index structure information, the operation logs of the operators in the system's automatic logs are extracted. Using the record number as the primary key, the button click number and status switch number fields are read sequentially. The click timestamps are sorted in ascending order according to the task round number. The difference between consecutive button click timestamps is extracted as the instruction operation duration. If the recorded time sequence in a task round is [10.2s, 12.5s, 14.1s, 17.0s], then its operation durations are 2.3s, 1.6s, and 2.9s respectively, corresponding to the duration sequence [2.3, 1.6, 2.9]. This sequence serves as the time difference data source for the operation segments in each task round. Then, the above time sequences are matched sequentially according to the status switch numbers identified in the task status change records. After classifying the status numbers of each operation event, the operation durations under the same status number are averaged. For example, the instruction execution durations corresponding to status number S2 are 2.3s, 2.4s, and 2.2s, then its average value is (2). (3 + 2.4 + 2.2) / 3 = 2.3s. Further, the system's confirmation response flag after each command input is extracted from the operation response record. If the response flag is 1, it indicates confirmation of receipt; if it is 0, it indicates no response. Items with no response are removed and an anomaly is recorded. The response stability benchmark value λ is set to 0.95, based on the task response success rate φ = N1 / N, where N1 is the number of successful responses and N is the total number of operations. A task is considered valid only if φ ≥ λ. This setting value comes from the fact that the average success rate of operation response in the gas turbine task test record fluctuates between 94% and 99%. After rounding the sample mean, it is set to 95%. Finally, the four parameters of the operator in each round of task—status number, command click number, operation duration, and response flag—are integrated into an integrated operation behavior data table according to the task number. The task number is established as the main index, and the fields include "click number," "status number," "duration," and "response flag." Finally, the operation behavior data information is generated.

[0072] The behavior sequence classification submodule extracts the operation number, click location number, and status switch sequence number of each operator within the task cycle based on the operation behavior data information. It then organizes and divides the operation sequences within continuous task segments by number and assigns task stage labels to the operation sequences within the cycle, generating operation node sequence behavior records.

[0073] Based on the operational behavior data, the operation number field and click location number field of each operator within their task cycle are extracted. The operation numbers are then numbered and organized according to the time sequence recorded in the task cycle. If an operator's operation numbers in task cycle 1 are F003, F006, F009, and F011, they are organized into the sequence [3, 6, 9, 11]. The state transition sequence number field is then extracted, and a mapping list is established for the state transition numbers according to the time sequence. Interval judgment is performed to classify the task node sequences between two adjacent state numbers into operation sequences within the same task segment. This method completes the task segment division of the operation sequence. For each task operation sequence segment, a stage label is assigned. For example, stage T1 represents the "pre-start segment," stage T2 is the "loading and adjustment segment," and stage T3 is the "stability maintenance segment." The stage label value is determined according to the task segment. In the definition of the point index structure, if the state switching sequence of a certain operator is S1→S3→S5→S7, its operation sequence number is [3,6] as T1, [9] as T2, and

[11] as T3. Add labels respectively to complete the classification and marking of the operation sequence within the task cycle. Bind the operation number sequence, click position number sequence and stage label information corresponding to each task stage. After integration, establish a task operation stage sequence record table. In order to prevent the stage division granularity from being too fine or too coarse and causing label mismatch, set the stage label recognition stability coefficient κ=0.85. When the stage label consecutive repetition consistency ratio μ satisfies μ≥κ, it is considered that the stage classification is stable. If it is lower than this value, adjacent label intervals need to be merged or re-segmented. This value is set based on the average consistency ratio of 84.7% in the stage label recognition sample and rounded upward. Finally, the operation node sequence behavior record is generated.

[0074] Please see Figure 2 and Figure 4 The behavioral stability characterization module includes a duration difference calculation submodule, a switching mean extraction submodule, and a stable distribution evaluation submodule.

[0075] The duration difference calculation submodule is based on the operation node sequence behavior record. According to the dwell time sequence of each task node, the duration sequence corresponding to each node is grouped according to the task round, and the dwell time variance value within each node group is calculated to obtain the node dwell time difference information.

[0076] The duration difference calculation submodule obtains the dwell time sequence of each task node in the operation node sequence behavior record and processes it by task round. First, it extracts the entry time and exit timestamp of each task node in the record, using node number j as an index, and then calculates the start time of each task round. With end time Subtracting the two gives the duration of stay. For example, for node j=2, its entry and exit times in the five rounds of tasks are (101.2, 106.4), (98.0, 102.5), (100.0, 104.7), (103.1, 107.8), and (99.5, 104.2), respectively, corresponding to a dwell time sequence of [5.2, 4.5, 4.7, 4.7, 4.7]. This sequence is assigned to task node number 2. Next, for this type of sequence for each task node, it is aggregated by node number and subgrouped by round dimension to further calculate the degree of fluctuation. The standard deviation calculation formula is used: Where D j,i Let be the dwell time of node j in the i-th round. Let N be the average dwell time across all task rounds at this node, and N be the number of task rounds. Taking the data above as an example, the average dwell time for node 3 is... The standard deviation is:

[0077]

[0078] To determine the impact of variance on rhythm fluctuations, a threshold θ for identifying dwell time differences is set. v =0.08, when V j ≥θ v Time nodes will be considered to have significant dwell instability. This threshold is based on test results where the average dwell time of the task fluctuates within the range of [0.02, 0.12] and the 80th percentile is used as the judgment criterion, which has sufficient empirical support. Finally, the dwell time series of all nodes are processed by variance calculation and organized into a unified structure to generate node dwell difference information.

[0079] The switching mean extraction submodule extracts the task status switching time period value between adjacent nodes based on the node dwell difference information, groups the switching time corresponding to each type of node in the same round, calculates the mean of the time period value of each group, and generates node switching mean data.

[0080] The mean extraction submodule extracts the task state transition time interval between adjacent nodes based on the node dwell difference information. In each task round, for each pair of consecutive nodes j and j+1, the departure timestamp of node j is obtained. The entry timestamp of node j+1 Calculate its switching interval as

[0081]

[0082] The switching intervals of all rounds for the same node pair are summarized into a time period sequence, and each Δ j The average switching time M for each type of node is obtained by averaging the values ​​of i. jTaking node pair (3,4) as an example, its switching times in the 5 rounds of tasks are [1.8, 2.1, 2.0, 1.9, 1.7], and the average value is calculated as follows:

[0083]

[0084] To identify nodes where the switching rhythm exhibits abnormal lag, a baseline value λ for the rhythm average is set. m =2.5, when M j >λ m It was determined that the node had a potential switching delay. The baseline value was determined by taking the median of 1.85 seconds and adding a 35% safety margin, based on the longest switching interval of the system task cycle being 3.2 seconds and the shortest being 1.1 seconds. The average switching time of all nodes was then compiled and summarized to generate the node switching mean data.

[0085] The stable distribution assessment submodule calls the node switching mean data, combines it with the variance value corresponding to each node in the node dwell difference value, calculates and obtains the node rhythm fluctuation score, sorts them in ascending order by node number, sets the rhythm stability assessment interval threshold, judges whether the score value falls within the interval range, marks the nodes that exceed the assessment interval threshold as fluctuating nodes, and marks the rest as stable nodes, establishes a node stability and instability identification table, and generates task node behavior stable distribution information.

[0086] The specific formula for calculating and obtaining the node rhythm fluctuation score is as follows:

[0087]

[0088] Among them, S n V represents the rhythm fluctuation score of the nth task node. n M represents the variance of the dwell time at the nth node. n Z represents the average handover time of the nth node, α is the handover time offset correction constant, and Z i This represents the start time of the node's operation in the i-th round. The value represents the arithmetic mean of the start times of all rounds of operations for a node, where N represents the number of rounds the node participates in, β is the behavioral complexity adjustment factor, and C... n This represents the number of path branches of the nth node in the current task flow;

[0089] This formula aims to quantify the degree of fluctuation in the operational rhythm of each task node in a multi-round training task. Its core lies in unifying the measurement of the temporal discreteness of node behavior, node switching efficiency, and task structural complexity.

[0090] The first calculation item is... Where V nM represents the variance of the dwell time of a node in different training rounds, reflecting the degree of fluctuation in operation time. n This represents the average time taken for node state transitions, serving as the base frequency for characterizing task rhythm. By combining these two metrics and correcting for potential amplification of anomalies caused by low-latency switching nodes with a constant α, the results in this section reflect the temporal instability of nodes under unit task switching efficiency.

[0091] The second calculation item is... Used to measure the dispersion of a node's startup time across multiple training rounds. Z i This refers to the start time of this node in each round. This is the average value across all rounds. This part characterizes the consistency deviation in time rhythm using the square root of the standard mean absolute deviation; the larger the value, the stronger the fluctuation in behavior over time.

[0092] The third term is ln(1+β·C) n In ), C n This represents the number of branch paths that this node appears in the task path structure, and β is a complexity adjustment factor. The larger this value is, the higher the degree of freedom in path selection of this node, the higher its behavioral complexity, and the higher the tolerance should be given to its rhythm fluctuations. Therefore, a logarithmic function is introduced to appropriately scale it down.

[0093] The entire scoring formula constructs a comprehensive measure of rhythm fluctuation score by multiplying three segments, which makes it possible to clearly identify unstable nodes caused by behavioral instability, inconsistent time rhythm, or complex structure in the analysis of rhythm changes in node tasks.

[0094] The stable distribution assessment submodule calls the node switching mean data and combines it with the variance value corresponding to each node in the node dwell difference data to calculate the rhythm fluctuation. For this purpose, the following formula is constructed:

[0095]

[0096] In the formula, S j V represents the rhythm fluctuation score of the j-th task node, a comprehensive value measuring its behavioral stability. j The variance of node dwell time is derived from the fluctuation of operator's operation time at that node across multiple task rounds, and is expressed in seconds. To eliminate the influence of units, this embodiment normalizes it to a percentage of the maximum variance value. M j Z represents the average task state switching time of this node, in seconds. Its normalized form is the ratio to the average time of the node group. α is the switching time offset correction constant, set to 0.25. This value is derived from measured data of interface response delay during task switching, measured as the minimum stable offset delay recorded by the actual button response on the gas turbine control console.i This represents the start time of the operation at this node in the i-th round, in seconds, and is taken from the system timestamp of the operator's first entry into the node's task phase during each training round. This represents the arithmetic mean of the start time of this node across all rounds, N represents the number of sampling rounds (5 in this example), β represents the behavior complexity adjustment factor, used to adjust the interference of complexity on the score in multi-path tasks (set to 0.3 in this example), and is derived by fitting based on the principle of minimizing the regression residual between the number of path branches and the accuracy of user operations. j This indicates the number of path branches of the j-th node in the current task flow. This parameter is obtained by counting the number of branches of the nodes in the tree structure and has a value range of 1–5, indicating that there are several possible execution path directions for this node.

[0097] The formula calculation logic is as follows:

[0098] Part One It is used to reflect the relative ratio between dwell fluctuation and switching rhythm. If the dwell fluctuation is large and the switching rhythm is small, the ratio is relatively high.

[0099] Part Two This is used to measure the consistency of the node's entry time during the training cycle; if the fluctuation is large, the stability is weak.

[0100] Part 3 ln(1+β·C) j This is a complexity adjustment term; if there are many path branches, the instability needs to be appropriately increased in weight.

[0101] The entire formula embodies a complex logical structure of "behavioral differences × entry volatility × complexity index".

[0102] To demonstrate the complete calculation process, the following provides the actual collected data and calculation examples for each parameter:

[0103] Table 1 Calculation Parameters for Node Rhythm Fluctuation

[0104]

[0105] Taking node j=1 as an example:

[0106] α=0.25, β=0.3, N=5;

[0107] The first item is:

[0108] The second item is:

[0109] The third term is: ln(1+0.3·2)=ln(1.6)≈0.470;

[0110] Comprehensive calculation yields:

[0111] S1=0.248×0.707×0.470≈0.082;

[0112] The results indicate that node 1 exhibits operational fluctuations across multiple training rounds, but remains within the edge of the rhythm stability evaluation range [0.05, 0.10] set by the system. If a slight upward fluctuation occurs subsequently, it will be marked as a fluctuating node. This value will then be used to establish a table of stable and unstable nodes, providing a calculation basis for classifying and judging the stable distribution information of node behavior in the next stage of the task.

[0113] The advantage of the formula is that it allows for the modification of the path complexity parameter C. j Using exponential logarithmic multiplication can enhance the sensitivity of scoring when node complexity increases abruptly. At the same time, by introducing normalized offset differences and adjustable coefficients α and β, it can flexibly adapt to the error adjustment needs under different training task conditions, thus providing a quantifiable and interpretable indicator system for behavioral stability assessment.

[0114] Please see Figure 2 and Figure 5 The task deviation judgment module includes a path number extraction submodule, a parameter difference calculation submodule, and a comparison result generation submodule;

[0115] The path number extraction submodule extracts the task node numbers marked as unstable areas based on the stable distribution information of task node behavior, obtains the operation path sequence numbers recorded by the operators when performing corresponding node operations in multiple training stages, and combines and marks them according to the node number and stage number to generate node path number combination information.

[0116] The path number extraction submodule, based on the stable distribution information of task node behavior, first extracts the node numbers of all nodes marked as unstable. For example, nodes numbered 5, 7, and 9 have stability scores of 3.2, 2.8, and 3.5 respectively, exceeding the set stability interval threshold [0.5, 2.5]. Therefore, they are included in the unstable node set. The submodule then obtains the path records of operators performing operations on these nodes during multiple training phases, recording the order in which each operator executes nodes 5, 7, and 9 in each training phase, and assigning it a path sequence number P. j,kWhere j is the node number and k is the stage number, for example, in stage 1, the operation sequence number of worker A on nodes 5, 7, and 9 is 2, 3, and 5, and in stage 2 it is 1, 2, and 4 respectively. The system uses a combination of "node number + stage number", such as "5-1", "5-2", "7-1", etc., to group and organize all path number information, and establish a standard format data table to uniformly identify the path sequence number corresponding to the node and stage. The setting of the stable interval threshold is based on the median value of the stable node score distribution in the rhythm fluctuation score, which is 1.8. Combined with the stable node variance range of 1.0, the threshold is adjusted by 0.3 and 0.7 respectively to set the lower limit of 0.5 and the upper limit of 2.5, forming a judgment interval. The boundary value can be dynamically adjusted within a 0.3 times interval as the complexity of the task process increases, and finally the node path number combination information is generated.

[0117] The parameter difference calculation submodule calls the node path number combination information, extracts the trigger time point and target parameter number of each instruction in the path, calculates the difference sequence for the instruction number and parameter number in adjacent stage paths under the same node, performs intra-group averaging on the difference sequence, and obtains the path instruction and parameter offset information.

[0118] The parameter difference calculation submodule calls the node path number combination information to extract the job instruction trigger time point T in each operation path under the combination identifier. p With the corresponding target parameter number Q p For example, under the combination identifier "5-1", the trigger times in the path are [12.3, 14.7, 17.9], and the corresponding target parameter numbers are [201, 204, 210]. In combination "5-2", they are [13.0, 15.1, 18.5], and the corresponding target parameter numbers are [202, 205, 213]. The difference between the corresponding instruction number and parameter number at the same node in both stages of the path is calculated, and the difference sequence is as follows:

[0119] ΔP=[|2-1|,|3-2|,|5-4|]=[1,1,1];

[0120] ΔQ=[|201-202|,|204-205|,|210-213|]=[1,1,3];

[0121] Calculate the average of each set of difference sequences to obtain the average difference of the path operation numbers. The average difference of the target parameter number is To reasonably determine path consistency offset, a parameter offset identification threshold δ is set. q=2.0, and its setting is based on the following: The system statistics show that the maximum number span of the same node across stages in all operation paths of three typical tasks is 5. Considering the average frequency of abnormal instruction tolerance and normal adjustment behavior is 1.4, the difference between its mean of 1.4 and the maximum value of 5 is calculated to be 3.6. After multiplying by a coefficient of 0.5 and correcting downwards, it becomes 1.8. The approximate value of 2.0 is taken as the threshold for judging parameter number offset. The threshold value can be adjusted down as the accuracy requirement of parameter number increases, but it should not be lower than 1.5. This is used to judge whether there are any over-limit offset records. Finally, the average difference information corresponding to all node path combinations is summarized to generate path instruction and parameter offset information.

[0122] The comparison result generation submodule calls the target comparison table composed of the corresponding task target number and path instruction number in the training instruction library according to the path instruction and parameter offset information. It extracts the records with a difference greater than the set path error threshold and adds an offset level label, establishes a mapping structure between the corresponding node number and the offset level, and generates a list of node path and result deviation records.

[0123] The comparison result generation submodule, based on the path instructions and parameter offset information, calls the preset target comparison table in the training instruction library and extracts the target task number G. t With path instruction number P t The standard correspondence is established, such as the standard instruction sequence for the path of target task number 305 being [101, 102, 103]. The acquired instruction number offset values ​​are compared one by one with the target sequence to determine whether they exceed the allowable error range. For example, a path error threshold θ can be set. p =2. This value is set based on the maximum allowable frequency of erroneous operations of 3 times in the three types of tasks: gas turbine start-up, load switching, and combustion adjustment, and the average task instruction span of 5. Taking the product of these two values ​​(15) and 0.13 of that, we get 1.95, rounded to the nearest integer 2 as the error threshold. It is stipulated that stages with high task switching frequency can allow an upward adjustment of 0.5, with the lower limit maintained at 2.0. If the path number offset value of a record is 3, it is greater than the threshold, and it is considered that an operational misalignment has occurred on that path under the node-stage combination. The system adds an offset level label to this record. The offset level is divided by the difference range, where 0-1 is low offset, 2 is medium offset, and 3 and above is high offset. For example, the number "5-2" is... The time marker is set as the high offset level. After traversing all the combined records in sequence, a mapping structure is established between the node number and the offset level label, and finally a list of node paths and result deviation records is generated.

[0124] Please see Figure 2 and Figure 6 The path error extraction module includes a high-frequency path filtering submodule, a behavior parameter extraction submodule, and a path aggregation and sorting submodule.

[0125] The high-frequency path filtering submodule calls the list of node paths and result deviation records, filters records with path number repetition rate greater than three times, checks whether the action offset level is higher than the upper limit of the allowable offset level of the task target, selects the path number and node number combination that meet the conditions, and generates high offset repetitive path information.

[0126] After calling the list of node path and result deviation records, the system reads the path number of each record and the number of times it appears in multiple training tasks, aggregates each record entry by path number, counts the repetition frequency of each path number, sets the cumulative value of the path number through a counter, and considers it to meet the high repetition standard when the cumulative value is greater than 3. The system selects the path number and extracts its associated node number information, reads the offset level of each node number, and compares it with the upper limit of the allowable offset level of the task target. If the two are equal, the system retains the combination of the path number and the node number as a valid screening object. The upper limit of the allowable offset level of the task target is set to 5, based on the fault tolerance setting of high-risk misoperation nodes in the gas turbine operation behavior assessment specification. The value of 5 corresponds to the boundary case where the offset between the operation instruction and the target instruction in the execution parameters does not exceed 2 coding units. If the offset level exceeds 5, it will be forcibly marked as a dangerous path. The setting process refers to the offset data of 3 typical tasks, and sets the standard offset level sequence to 1 to 7. With 5 as the upper limit, if the maximum offset value in the path is 5, it can be included in the screening range of this sub-module, and finally, high offset repetition path information is generated.

[0127] The behavior parameter extraction submodule extracts the trigger time, operation object code and repeated call cycle of the object in the task phase for each action node in the path segment based on the high offset repeated path information. It also aligns the action execution records of multiple time periods under the same node by number and generates a set of repeated behavior parameters.

[0128] Based on the selected path number and node number combination in the high offset repetitive path information, the record data of each action node is extracted item by item. The trigger timestamp information is obtained through the operation record field of the corresponding node in the task log file. The target object code in each record is read to identify the number of calls and call cycle in the complete task stage. The action records are sorted by node number and timestamp. The position number alignment processing is performed on the records in different time periods. For example, if the node number is 12, the trigger time in stage 1 is 11:02:34 and the trigger time in stage 2 is 11:03:22, then the repetitive record number alignment mapping of node number 12 is established. The call cycle difference of the operation object in each stage is further read. The cycle value is obtained by dividing the call duration of each stage by the number of triggers. For example, if the operation object X is triggered 3 times in stage 1, the trigger intervals are 20 seconds, 25 seconds and 30 seconds respectively, and the average cycle is 25 seconds in stage 1 and 26 seconds in stage 2. After grouping by node number, the records are integrated to finally generate a set of repetitive behavior parameters.

[0129] The path aggregation and sorting submodule calls the set of repeated behavior parameters, classifies the set of action nodes with the same operation object code and path segment number, calculates the number of repeated calls and the time series fluctuation value of the trigger interval of nodes under each type of path, and sorts them in descending order with the number of repeated triggers as the main order, establishes a matching table between clustered path segments and sorting numbers, and generates a group of error clusters in the operation behavior path.

[0130] After calling the repetitive behavior parameter set, the action records of each node contained therein are categorized according to the combination of operation object code and path segment number. All records in each group have the same operation object and path number. For all action nodes in each group, the number of repetitions is counted and the interval fluctuation value is calculated. The number of repetitions is obtained by counting all trigger record entries. The interval fluctuation value is calculated using the following standard deviation method: if the interval time sequence in the record is 12s, 14s, 13s, and its mean is 13s, then the fluctuation value is... All path segments are sorted in descending order of the number of repeated calls, with the fluctuation value used as a secondary sorting factor. After sorting, a matching table between the clustered path segment numbers and the sorted numbers is established, and the error clustering segment groups of the operation behavior path are output. The threshold for the number of repeated calls is set to 3, based on the fact that if the behavior redundancy or misuse tendency in a path segment is more than 3 times, it can be regarded as a behavior solidification pattern. This setting comes from the statistics of 10 sets of training data, in which 98% of the repeated misoperation paths have a trigger number of no less than 3 times, and the numerical setting has significant distinguishability.

[0131] Please see Figure 2 and Figure 7The correction guidance generation module includes a path information extraction submodule, a node deviation filtering submodule, and a feedback instruction construction submodule;

[0132] The path information extraction submodule gathers paragraph groups based on the errors in the operation behavior path, extracts the trigger action number of the first operation node in the paragraph, the corresponding operation feedback delay time value, and the repeated trigger time interval sequence in multiple rounds of training, archives and classifies each parameter item according to the combination of path number and node number, and generates node behavior path attribute information.

[0133] After obtaining the error clusters in the task behavior path, the trigger action number of the first node in each segment is extracted. The initial operation information corresponding to the node in each round of the task can be mapped and transformed. By reading the timestamp information contained in the task behavior log, the time difference between the moment the action node is triggered and the time when the response system feedback takes effect is recorded, and the feedback delay value is calculated. This feedback delay value can be obtained by subtracting the trigger time from the operation response time. In the multi-round training records, the response operation of a certain node is timestamped, the time periods in which it appears multiple times in different rounds are selected, and the time interval sequence between its triggers is recorded. For example, when a certain node appears in rounds 3, 6, and 10, the trigger interval is the sequence of the trigger time differences in the corresponding rounds. Then, a two-dimensional key-value table is constructed according to the combination of path number and node number. The extracted action number, feedback delay value, and interval sequence are labeled and archived as structured data, thereby generating node behavior path attribute information.

[0134] The node deviation filtering submodule calls the node behavior path attribute information to obtain the correct click number, preset operation duration and task panel position number in the standard task path corresponding to each node. It compares these with the click number, operation time and interface position coordinates of the current node, calculates the node execution error degree value, sets the deviation identification threshold, filters nodes with scores greater than the deviation identification threshold as deviation nodes, establishes a node number and deviation label comparison table, and generates error task node identification information.

[0135] The specific formula for calculating the degree of node execution error is as follows:

[0136]

[0137] Among them, E j B represents the execution error level of the j-th task node. j Indicates the current clicked number, R j Indicates the standard click number, T j This represents the duration of the click on the j-th node. P represents the average click time for all rounds of the j-th node, γ is the adjustment factor for duration error, and P jσ represents the difference in index between the current clicked position of the j-th node and the standard panel position, δ is the adjustment factor for panel layout deviation, and σ is the index difference. j This represents the standard deviation of the click duration of the j-th node;

[0138] This formula is used to measure the degree of execution deviation between each task node and the standard task path during actual training operations. Its core logic is to generate a unified error evaluation index by fusing the numerical values ​​of four dimensions: operation accuracy, rhythm consistency, interface space deviation and behavior volatility.

[0139] Molecular part In the middle, the first item |B j -R j The first term represents the difference between the current click command and the standard command number, indicating whether the operator has selected the wrong command. The larger the difference, the more the operation deviates from the task objective. The second term is a relative deviation measure consisting of the difference between the click duration and its historical average, reflecting the degree of deviation between the current node's operation rhythm and its own stable rhythm. γ is its influence weight adjustment factor. The overall square of this term aims to increase the scoring effect of significant deviation behaviors, ensuring that errors or unstable behaviors are not masked.

[0140] The denominator is Where P j The difference in number between the current operating position and the standard panel position is used to enhance its response strength in spatial deviation by taking the square root of the square plus 1. The second part is δ·ln(1+σ). j In ), σ j The standard deviation of the click duration is used to measure the time stability of the operator at this node during the training process, and is used in conjunction with the adjustment factor δ to control its influence on the final error.

[0141] This overall structure expresses the operation offset, time offset, position offset, and volatility using a ratio structure, allowing the execution behavior of each node to be expressed through a unified error index E. j The system performs sorting, filtering, and discrimination. A higher value indicates a greater deviation between the node's operation and standard requirements, serving as a basis for triggering training feedback, misconception alerts, and corrective suggestions.

[0142] Retrieve the node behavior path attribute information and extract the click number B of the current task node. j Duration T j Interface location number P j And match the standard click number R of the corresponding node in the standard task path. j The average click time for this node across all training rounds. In addition to the standard interface location number, the above data parameters are organized under the same node number, and error evaluation is performed on all nodes, specifically using the formula for calculating the degree of execution error:

[0143]

[0144] Among them, |B j -R j | represents the modulus of the difference between the current click number and the standard click number, used to reflect the numbering deviation; To normalize the relative difference in click time, a coefficient γ (duration error adjustment factor) is added to form a weighted adjustment term. γ is typically set to 0.6, and this value is derived from the normalized results of the click error fluctuation sensitivity statistics for each type of task node in the path behavior test group; the denominator is... Reflecting the difference in interface position, δ·ln(1+σ j This is used to penalize situations where node click time stability is insufficient, where δ = 0.5 represents the panel offset tolerance setting, which is half the standard deviation (4 pixels) of the click area control width in the experiment. The value is reasonable and does not fluctuate with task type. σ j The standard deviation of node click times is obtained from the standard deviation of all click times of the node during training.

[0145] For ease of understanding, let's assign the click number B to a certain node. j =23, standard number R j =20, click duration is T j =5.6s, the average training time for this node is The difference in node interface numbers is P j =6, and the standard deviation of click duration is σ. j =0.75, substitute into the formula:

[0146]

[0147] Based on the node deviation identification requirements, a deviation identification threshold of 1.2 is set, derived from the task path accuracy control standard. This threshold is set to E for the average execution error range where the node number deviation is within ±2 and the click time fluctuation does not exceed 15%. threshold =1.2. The current node's score is 1.511, which is higher than the deviation identification threshold, therefore it is determined to be a deviation node.

[0148] The results indicate that the combined difference between the node click number and the click duration leads to a large bias score. Even after correction with location information, the score still does not fall within a reasonable error range. Therefore, the node should be marked in the discrimination information and a bias label should be generated.

[0149] Table 2 Example of Task Node Execution Error

[0150]

[0151] As shown in Table 2, node j=17 was identified as a deviation node due to severe offset. The advantage of the formula lies in the fact that, through the nonlinear weighted combination of number difference, time difference, and position difference, and by applying a logarithmic penalty to the volatility factor, it can comprehensively evaluate the stability and deviation of the operation behavior from multiple dimensions, effectively improving the sensitivity and robustness of deviation identification. The generated error task node discrimination information can accurately characterize the node error behavior.

[0152] The feedback instruction construction submodule identifies the error task nodes based on the error task node identification information, extracts the position of the nodes in each round according to the original task path order based on the selected node numbers, and combines the operation number with the target task parameter correspondence table to generate the execution order suggestions, parameter comparison prompts and operation time correction statement set required for each node. It integrates them into a structured push content set, establishes the mapping relationship between the task path and the push content, and generates a training task correction prompt instruction list.

[0153] Based on the identified erroneous task node identification information, for the labeled node number, according to the order in which the node number appears in the training log of each round, the relative order number of the node's operation execution in each round is extracted. Then, the click operation number recorded by the node in the corresponding round is obtained. The task target and parameter mapping table is queried to extract the target parameter number that the operation should achieve. Combined with the duration value recorded in the previous path information, the target parameter offset interval between the current operation and the recommended operation is calculated, and it is determined whether it exceeds the recommended operation interval. If it does, a corresponding prompt statement is generated. The prompt content includes three types of information: first, the relative position expression of the suggested execution order, such as "This operation is recommended to be executed in the 3rd position in the current path"; second, the target parameter matching prompt, such as "The operation target should point to the object with number B26"; and third, the click time suggestion, such as "It is recommended to control the click duration to be completed within 4.2 seconds". All prompt statements are uniformly collected by node number to construct JSON structure content, complete the structured integration, establish the mapping between path number and structured push content, and generate a training task correction prompt instruction list.

[0154] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0155] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0156] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0157] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0158] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0159] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0160] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0162] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A real-time analysis and feedback system for safety education in gas turbine power plants, characterized in that: The system includes: The task node monitoring module acquires the operation process information of safety education and assessment, records the click number of each operator's instruction button on the operation panel, operation duration and task status switching sequence, and categorizes them by task cycle to generate operation node sequence behavior records. The behavior stability characterization module calculates the stability score of the operation based on the operation node sequence behavior record, determines the stable region and the unstable region according to the stability evaluation interval, and generates task node behavior stability distribution information. The task deviation judgment module obtains the path sequence number and result value offset level of the node operation performed by the operator in each stage based on the stable distribution information of the task node behavior, and establishes a comparison relationship based on the target correspondence table in the training instruction library to generate a list of node path and result deviation records. The path error extraction module calls the list of node path and result deviation records, extracts the reaction delay time, operation object code and repetitive operation cycle of the action trigger in the path segment, aggregates similar paths and sorts them according to the repetitive trigger trend, and generates a group of operation behavior path error clusters.

2. The real-time analysis and feedback system for safety education in gas turbine power plants according to claim 1, characterized in that, The operation node sequence behavior record includes task cycle number, state switching sequence, click number set, duration distribution, and panel response identifier. The stable distribution information of task node behavior specifically includes rhythm-stable node partition, rhythm-fluctuating node partition, node number and corresponding stability mapping table, and rhythm-abnormal node index set. The node path and result deviation record list includes offset path number table, target number difference item, deviation level identifier group, behavior path misjudgment label, and task target response offset record. The operation behavior path misconception aggregation segment group specifically refers to the misconception starting node set, repeated trigger path template, path aggregation number group, action instruction change mapping, and misconception segment sorting index.

3. The real-time analysis and feedback system for safety education in gas turbine power plants according to claim 2, characterized in that, The task node monitoring module includes: The task information extraction submodule obtains the operation process information of safety education and assessment. Based on the task process information of gas turbine combustion adjustment task, cold start task and load switching task, it collects the number index, instruction trigger position number and corresponding task start and end node number of each function button in the task operation panel. It establishes a correspondence between the function buttons and status numbers in each type of task and generates task node index structure information. The task record organization submodule records the click number of the operation panel button and the task status switch number of the operator based on the task node index structure information. It extracts the duration value of the corresponding operation segment and the task instruction response flag bit according to the time sequence of each round of task operation process. It merges the instruction execution duration and status number to generate task operation behavior data information. The behavior sequence classification submodule extracts the operation number, click location number, and status switch sequence number of each operator within the task cycle based on the operation behavior data information. It then organizes and divides the operation sequences within continuous task segments into cycles, classifies the operation sequences within a cycle by number, assigns task stage labels, and generates operation node sequence behavior records.

4. The real-time analysis and feedback system for safety education in gas turbine power plants according to claim 3, characterized in that, The behavior stability characterization module includes: The duration difference calculation submodule, based on the operation node sequence behavior record, groups the duration sequence corresponding to each node according to the task round, according to the dwell duration sequence of each task node, calculates the dwell duration variance value within each node group, and obtains the node dwell difference information. The switching mean extraction submodule extracts the task state switching time period value between adjacent nodes based on the node dwell difference information, groups the switching time corresponding to each type of node in the same round, calculates the mean of each group of time period values, and generates node switching mean data. The stable distribution assessment submodule calls the node switching mean data, combines it with the variance value corresponding to each node in the node dwell difference value, calculates and obtains the node rhythm fluctuation score, sorts them in ascending order by node number, sets a rhythm stability assessment interval threshold, determines whether the score value falls within the interval range, marks nodes that exceed the assessment interval threshold as fluctuating nodes, and marks the rest as stable nodes, establishes a node stability and instability identification table, and generates stable distribution information of task node behavior.

5. The real-time analysis and feedback system for training behavior in safety education for gas turbine power plants according to claim 4, characterized in that, The specific formula for obtaining the node rhythm fluctuation score is as follows: Among them, S n V represents the rhythm fluctuation score of the nth task node. n M represents the variance of the dwell time at the nth node. n Z represents the average handover time of the nth node, α is the handover time offset correction constant, and Z i This represents the start time of the node's operation in the i-th round. The value represents the arithmetic mean of the start times of all rounds of operations for a node, where N represents the number of rounds the node participates in, β is the behavioral complexity adjustment factor, and C... n This represents the number of path branches of the nth node in the current task flow.

6. The real-time analysis and feedback system for training behavior in safety education for gas turbine power plants according to claim 5, characterized in that, The task deviation judgment module includes: The path number extraction submodule extracts the task node numbers marked as unstable areas based on the stable distribution information of the task node behavior, obtains the operation path sequence numbers recorded by the operator when performing the corresponding node operations in multiple training stages, and combines and marks them according to the node number and stage number to generate node path number combination information. The parameter difference calculation submodule calls the node path number combination information, extracts the trigger time point and target parameter number of each instruction in the path, calculates the difference sequence for the instruction number and parameter number in adjacent stage paths under the same node, performs in-group averaging on the difference sequence, and obtains path instruction and parameter offset information. The comparison result generation submodule, based on the path instructions and parameter offset information, calls the target comparison table composed of the corresponding task target number and path instruction number in the training instruction library, extracts the records with a difference greater than the set path error threshold and adds an offset level label, establishes a mapping structure between the corresponding node number and the offset level, and generates a list of node path and result deviation records.

7. The real-time analysis and feedback system for safety education in gas turbine power plants according to claim 6, characterized in that, The path error extraction module includes: The high-frequency path filtering submodule calls the list of node paths and result deviation records, filters records with path number repetition rate greater than three times, detects whether the action offset level is higher than the upper limit of the allowable offset level of the task target, selects the path number and node number combination that meet the conditions, and generates high offset repetitive path information. The behavior parameter extraction submodule extracts the trigger time, operation object code and repeated call cycle of the object in the task phase for each action node in the path segment based on the high offset repeated path information, and aligns the action execution records of multiple time periods under the same node to generate a set of repeated behavior parameters. The path aggregation and sorting submodule calls the set of repeated behavior parameters to classify the set of action nodes with the same operation object code and path segment number. It calculates the number of repeated calls and the time series fluctuation value of the trigger interval for each type of path node, and sorts them in descending order with the number of repeated triggers as the main order. It establishes a matching table between the clustered path segments and the sorting number, and generates a group of error clusters in the operation behavior path.

8. The real-time analysis and feedback system for training behavior in safety education for gas turbine power plants according to claim 7, characterized in that, The system also includes: The correction guidance generation module gathers paragraph groups based on the errors in the work behavior path, calls the correct operation path corresponding to the node in the standard task flow, filters out erroneous operation nodes that do not match the task objective, generates a set of instruction content for each erroneous node with execution order suggestions, objective-corresponding prompt information and operation duration correction prompts, establishes feedback push content for training site prompts, and generates a list of training task correction prompt instructions; The training task correction prompt instruction list includes a chain of correction action numbers, sequence adjustment prompts, duration correction markers, task node target reminders, and interactive push information formats.

9. The real-time analysis and feedback system for training behavior in safety education for gas turbine power plants according to claim 8, characterized in that, The correction guidance generation module includes: The path information extraction submodule gathers paragraph groups based on the operational behavior path errors, extracts the trigger action number of the first operation node in the paragraph, the corresponding operation feedback delay time value, and the repeated trigger time interval sequence in multiple rounds of training, archives and classifies each parameter item according to the combination of path number and node number, and generates node behavior path attribute information. The node deviation filtering submodule calls the node behavior path attribute information to obtain the correct click number, preset operation duration and task panel position number in the standard task path corresponding to each node. It compares these with the click number, operation time and interface position coordinates of the current node, calculates the node execution error degree value, sets the deviation identification threshold, filters nodes with scores greater than the deviation identification threshold as deviation nodes, establishes a node number and deviation label comparison table, and generates error task node discrimination information. The feedback instruction construction submodule, based on the error task node identification information, extracts the occurrence position of the node in each round according to the original task path order based on the selected node number, and combines the operation number with the target task parameter correspondence table to generate the execution order suggestion, parameter comparison prompt and operation time correction statement set required for each node. These are integrated into a structured push content set, establishing a mapping relationship between the task path and the push content, and generating a training task correction prompt instruction list.

10. The real-time analysis and feedback system for training behavior in safety education for gas turbine power plants according to claim 9, characterized in that, The formula for obtaining the node execution error level value is as follows: Among them, E j B represents the execution error level of the j-th task node. j Indicates the current clicked number, R j Indicates the standard click number, T j This represents the duration of the click on the j-th node. P represents the average click time for all rounds of the j-th node, γ is the adjustment factor for duration error, and P j σ represents the difference in index between the current clicked position of the j-th node and the standard panel position, δ is the adjustment factor for panel layout deviation, and σ is the index difference. j This represents the standard deviation of the click duration of the j-th node.

Images