Interactive control method for power operation and maintenance virtual simulation training

Abstract

The application discloses an interactive control method for power operation and maintenance virtual simulation training, and particularly relates to the technical field of power operation and maintenance simulation, which comprises the following steps: obtaining operation behavior data of students, using a multi-dimensional capability evaluation model to divide the real-time capability grades of the students, constructing a difference adaptation parameter model in combination with the differences between virtual equipment and target actual equipment, matching training contents with corresponding difficulty from a preset training task library, constructing a knowledge graph of the students, and generating a personalized training path through a path recommendation engine, iteratively updating the training difficulty and the recommended path according to newly generated behavior data and the performance of the students in the difference adaptation capability, constructing the difference adaptation parameter model, incorporating the differences into the adjustment of the training tasks and the capability evaluation mechanism, dynamically adjusting the training task parameters to fit the complexity of the actual equipment, and breaking through the migration barriers between virtual training and real operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power operation and maintenance simulation technology, and more specifically, to an interactive control method for virtual simulation training in power operation and maintenance. Background Technology

[0002] With the rapid development of augmented reality (AR) and virtual reality (VR) technologies, power operation and maintenance training based on virtual simulation environments has shown broad application prospects in improving training efficiency and reducing practical training risks. By constructing three-dimensional models of typical power equipment and operation and maintenance scenarios, trainees can complete various operation exercises and fault handling simulations in an immersive environment, which has gradually become a powerful supplement to traditional power training.

[0003] However, existing virtual simulation training systems still have the following technical bottlenecks and shortcomings in practical applications:

[0004] Most current simulation training systems use fixed training content and processes, ignoring the differences among learners in terms of knowledge base, operating habits, and cognitive level. This results in a lack of flexibility and pertinence in the teaching process. In terms of operation difficulty, task path, and prompt intensity, they cannot be adjusted in real time according to the learners' learning outcomes, which affects training efficiency and the depth of skill mastery.

[0005] Virtual simulation models of power equipment typically differ from actual equipment in terms of structural details, operational layout, and human-machine interaction feedback. These differences are mainly reflected in the following aspects:

[0006] Structural differences include the device's external dimensions, component layout, and type of operating interface. Due to performance optimization or interaction simplification considerations, virtual modeling often cannot fully reproduce the detailed structure of the real device.

[0007] Differences in response behavior: Virtual devices lack real physical feedback, such as button resistance, switch delay, system linkage response, etc., and their operating logic is mostly a preset process, lacking dynamism and complexity;

[0008] Due to the aforementioned differences, the operating habits that trainees develop in the virtual environment are difficult to fully transfer to actual equipment, resulting in strong maladaptation in high-pressure scenarios such as fault handling and emergency response, thus limiting the effectiveness and practical value of simulation training.

[0009] To address the aforementioned shortcomings, a technical solution is proposed. Summary of the Invention

[0010] The purpose of this invention is to provide an interactive control method for virtual simulation training in power operation and maintenance, in order to solve the problems mentioned in the background art.

[0011] To achieve the above objectives, the present invention provides the following technical solution: an interactive control method for virtual simulation training of power operation and maintenance, the specific steps of which include acquiring the trainee's operational behavior data, including task completion time, frequency of misoperation, accuracy rate of operation sequence, and training completion rate;

[0012] Based on the operational behavior data, a multi-dimensional ability assessment model is used to classify trainees' ability levels in real time.

[0013] By combining the structural and response behavior differences between virtual devices and target physical devices, a differential adaptation parameter model is constructed.

[0014] Based on the trainee's current ability level and the aforementioned differential adaptation parameter model, the system matches training content of corresponding difficulty from the preset training task library and automatically adjusts the parameter configuration of the simulation training task.

[0015] Based on the student's historical training path and device adaptability performance, a knowledge graph of the student is constructed, and a personalized training path is generated through a path recommendation engine.

[0016] After each round of training, the trainees' abilities are reassessed based on the newly generated behavioral data and their performance in terms of differential adaptability. The training difficulty and recommended path are then iteratively updated to achieve an adaptive and real-world transferable simulation training process.

[0017] Preferably, the method for obtaining student operational behavior data is as follows:

[0018] The task completion time is calibrated as Tct. The start time is recorded as Tst when the virtual simulation task begins and the completion time is recorded as Ten when the virtual simulation task ends. The expression for calculating the task completion time is Tct = Ten - Tst. For interrupted tasks and tasks that exit midway, the task duration of the current stage is recorded.

[0019] Set up a task flowchart, which includes the correct operation steps and the operation objects. Match each student's operation with the task flowchart and record any mismatched operation as a misoperation. Define the misoperation frequency as Mif, then Mif = number of misoperations / total number of operations.

[0020] Based on the longest common subsequence algorithm, the actual operation sequence executed by the trainee is compared with the preset standard operation sequence. The standard operation sequence is labeled as S = {s1, s2, ..., s}. m}, of length m, the actual operation sequence of the trainees is U={u1,u2,…,u n}, of length n, where each element s i ∈{s1,s2,…,s m} and u j ∈{u1,u2,…,un} represents a discrete power operation and maintenance step;

[0021] Let L(S,U) denote the longest common subsequence of sequences S and U, which satisfies and The order of elements in the longest common subsequence remains consistent, i.e., if L(S,U)={x1,x2,…,x…} k If k is the length of L(S,U), then there exists a set of indices that satisfy... and Similarly, for sequence U, k = |L(S,U)| is the longest of all common subsequences that satisfy the above conditions;

[0022] Constructing a two-dimensional matrix using dynamic programming Where D is a two-dimensional array of size (m+1)×(n+1), m is the length of the standard operation sequence, n is the length of the actual operation sequence, and each element D in the two-dimensional matrix D is a variable. i,j D represents the length of the longest common subsequence between the first i steps of the standard operation sequence and the first j steps of the student's actual operation sequence. i,j The recursive formula is Finally, we obtain the bottom right element D of the two-dimensional matrix D. m,n Let l be the length of the longest common subsequence of sequences S and U, denoted as l = |L(S,(U)|;

[0023] The correctness rate of the operation sequence is then calculated. Where m = |S| is the length of the standard operation sequence S;

[0024] The training task is divided into multiple subtasks. When each subtask is completed, a labeling signal is set to mark the node as completed. The training completion rate is defined as Trc, where Trc = number of completed nodes / total number of nodes. If the task fails or is interrupted, the completion rate is output as a percentage other than 100%.

[0025] The preferred method for classifying trainees' abilities in real time is as follows:

[0026] Define the ability status vector of the trainee as Xcs = [Tct, Mif, Sco, Trc]. A multi-dimensional ability evaluation model is constructed based on the ability status vector through a fuzzy membership function, and the ability levels are defined. Define the ability score function Asf of the multi-dimensional ability evaluation model. The calculation expression of Asf is Asf = α1*f1(Tct) + α2*f2(Mif) + ɑ3*f3(Sco) + α4*f4(Trc). In the formula, α1, α2, α3, and α4 are the weight coefficients of the parameters of each dimension in the ability status vector, and α1, α2, α3, and α4 are all positive numbers. f1(·), f2(·), f3(·), and f4(·) are the fuzzy membership functions of the parameters of each dimension in the ability status vector, and Asf represents the comprehensive ability level of the trainee.

[0027] Set the score thresholds Ath1 and Ath2, where 0 ≤ Ath1 < Ath2 ≤ 1. According to the value range of the trainee ability score function Asf, the trainees are divided into the following ability levels:

[0028] When Asf is less than or equal to Ath1, define the trainee ability level as L1 level;

[0029] When Asf is greater than Ath1 and less than Ath2, define the trainee ability level as L2 level;

[0030] When Ath2 < Asf, define the trainee ability level as L3 level;

[0031] Set up a real-time update mechanism to automatically collect the operation behavior data of the trainee in this round of tasks at the end of the training task, calculate the trainee's ability score function Asf in real time, re-determine the trainee's current ability. If the trainee's ability level changes compared with the previous round of evaluation results, trigger the personalized training path adjustment mechanism to recommend the task type and difficulty corresponding to the trainee's ability level.

[0032] Preferably, the method for constructing the differential adaptation parameter model is as follows:

[0033] Define the differential parameter dimension to divide the difference into the structural difference parameter ΔS and the response behavior difference parameter ΔR. The structural difference parameter ΔS includes the button position offset ΔS1 and the change in the complexity of the operation path branch ΔS2. The response behavior difference parameter ΔR includes the feedback delay difference ΔR1 and the action-response mismatch degree ΔR2. After normalizing the above four difference parameters, a difference vector Δ = [ΔS1, ΔS2, ΔR1, ΔR2] is formed;

[0034] Define the adaptation correction function Adf(Δ) for the trainees whose ability levels have been divided. The expression of the adaptation correction function Adf(Δ) is Adf(Δ):L c → Adapt the training adjustment parameters for each trainee ability level Lc Where c = {1, 2, 3}, and given the known difference parameter Δ, output a training task parameter configuration scheme that matches the trainee's ability level and equipment differences;

[0035] The logic of the training task parameter configuration scheme based on matching the output of the adaptive correction function with the trainee's ability level and equipment differences is as follows:

[0036] Based on the trainees' ability level and the difference vector Δ, matching task groups are searched from the preset training task library. The matching rules include task fitness threshold, operation complexity index and transfer difficulty assessment, and tasks that can improve practical operation ability under the influence of differences are matched.

[0037] For the matched training task, the adaptive training adjustment parameters output by the adaptive correction function Adf(Δ) are automatically configured to adjust the degree of operation guidance, the degree of response speed simulation, the tolerance range of misoperation, and the logic of displaying and hiding prompts, so as to realize the adaptive enhancement and personalized adjustment of the training task to the differences of the device.

[0038] Preferably, the method for obtaining the button position offset ΔS1 is as follows:

[0039] Extract the 2D coordinates of the same button in both the virtual and physical interfaces. The 2D coordinates of the same button in the virtual interface are (v1, v2), and the 2D coordinates of the same button in the physical interface are (a1, a2). Calculate the Euclidean distance between the two coordinates as the button position offset ΔS1. The calculation expression is as follows:

[0040] The method for obtaining the change in complexity ΔS2 of the operation path branch is as follows:

[0041] By performing directed graph analysis, the task flow of virtual and physical devices is modeled as an operation path graph G. v and G r In the operation path graph, nodes represent operation steps, and edges represent operation flows. The expression for calculating the average branching degree Bv of the virtual device is: The expression for calculating the path length Br of the actual device is as follows: Where the number of nodes v is the operation path graph G v The total number of nodes, i.e., the total number of operation steps, and the ∑ outgoing edge number p, represent the operation path graph G. v The number of directed edges originating from node Vp represents the number of possible next operation steps that this step can lead to. Node Vp is part of the operation path graph G. v The p-th node in the operation path graph G has a node number r. r The set of all possible valid operation paths, where ∑ outgoing edges Vq is the operation path graph G. rThe number of nodes in path Vq represents the number of operation steps required to complete the task. Path Vq is the operation path graph G. r The q-th path in the middle;

[0042] The logic for obtaining the feedback delay difference ΔR1 is as follows:

[0043] The system logs record the response time of each interactive operation. The response time of an interactive operation in the virtual device is denoted as tv, and the response time in the actual device is denoted as tr. The average latency of multiple interactive operations in both the virtual and actual devices is calculated as follows: and Among them, t v,a Let t be the feedback time after the a-th operation in the virtual device. r,a Let z be the feedback time after the a-th operation in the actual device, z be the total number of interactive operations, and a be the sequence number of the interactive operation. Then, the expression for calculating the feedback delay difference ΔR1 is:

[0044] The logic for obtaining the action-response mismatch degree ΔR2 is as follows:

[0045] Construct action-response mappings, which include action-response mappings (Mapv) for virtual devices and action-response mappings (Mapr) for physical devices. The expression for the action-response mapping (Mapv) of the virtual device is: The expression for the actual device's action-response mapping (Mapr) is: Among them, O a For the a-th interactive operation, For virtual devices to O a Feedback For virtual devices to O a Feedback;

[0046] Calculate the similarity between the virtual device and the actual device for each action response, and obtain a weighted average similarity Sre based on the similarity. The calculation expression is as follows: In the formula, Sre is the weighted average similarity, and β a Let β be the response weight for the a-th interaction, and let β be the response weight for the a-th interaction. a Rsi is a positive number. a Let Rsi be the response similarity of the a-th interaction, and Rsi a If the value is positive, then the expression for calculating the action-response mismatch ΔR2 is ΔR2=1-Sre.

[0047] The preferred method for calculating similarity is as follows:

[0048] Similarity is obtained through structural similarity index (SSIM) and Mel-frequency cepstral coefficient similarity. For operation responses involving visual feedback, the SSIM is used to compare the virtual device response image with the actual device response image. The method for obtaining the SSIM is as follows:

[0049] The formula for calculating the structural similarity index (SSIM) is as follows: x and y represent the pixel matrices of the virtual response image and the actual response image, respectively, μ x μ y These represent the average pixel values ​​of the image at x and y, respectively. Let σ represent the variances of the images x and y, respectively. xy C1 and C2 represent the covariance of the images x and y. C1 and C2 are small constants introduced to avoid the denominator being zero, and both C1 and C2 are positive numbers.

[0050] The image window comprises Top pixels, and the pixel value sequences of image x and image y are denoted as Pix = {Se1, Se2, ..., Se...}. Top} and Piy={Qu1,Qu2,…,Qu Top}, Se hp ∈{Se1,Se2,...,Se Top And Qu hp ∈{Qu1,Qu2,…,Qu Top}, then μ x The calculation expression is as follows μ y The calculation expression is as follows

[0051] variance of image x The calculation expression is as follows variance of image y The calculation expression is as follows Covariance σ of images x and y xy The calculation expression is

[0052] For operational responses involving auditory feedback, the Mel-frequency cepstral coefficient similarity method is used to calculate the similarity between the virtual and actual response speech signals. The method for obtaining the Mel-frequency cepstral coefficient similarity is as follows:

[0053] For a speech signal s(t), after pre-emphasis, framing, windowing, fast Fourier transform, and Mel filter bank processing, the Mel frequency cepstral coefficient vector Sfvx of the virtual response and the Mel frequency cepstral coefficient vector Sfvy of the actual response are obtained. The Mel frequency cepstral coefficient similarity between Sfvx and Sfvy is then calculated. The calculation expression is ∥Sfvx∥ and ∥Sfvy∥ represent the magnitudes of vectors Sfvx and Sfvy, respectively;

[0054] Then for interactive operation O a Response similarity Rsi a The calculation expression is Where γ is the weighting coefficient and γ is a positive number.

[0055] The preferred method for constructing the student's knowledge graph is as follows:

[0056] The knowledge graph is labeled Glk = (V, E), where the node set V is represented as V = {v1, v2, ..., v...}. b}, for each node v d ∈V, including task type, ability level, equipment involved, and training completion degree, and the edge set E is represented as E={e df =(v d ,v f ,Ret df )}, e df Ret represents the associated path from node d to node f. df This indicates the type of relationship from node d to node f.

[0057] The preferred method for generating personalized training paths is as follows:

[0058] Using a sliding window method and a trend-weighted scoring model, based on the learner's most recent Nota operation behavior data and device adaptation performance, the potential ability trend value of the learner is evaluated. A time series is generated based on the learner's most recent Nota task completion time, error frequency, operation sequence accuracy, training completion rate, and average response similarity of interactive operations. t is the task number corresponding to the current time point, Nota is the sliding window length, tas is the index of the task number, representing the time from task t-Nota+1 to task t, and Rsi tas Let be the average response similarity of the interactive operations within the 'tas'th task, calculated as follows: Where zis is the total number of interactive operations in the tas-th task, and a is the sequence number of the interactive operation;

[0059] Define a positive score function fsf for each dimension parameter jnd (tas), and assign a time decay weight ω to each training task. tas The formula for calculating the time decay weight is: Where atf is the decay factor used to highlight recent training performance, e is the base of the natural logarithm, and jnd is the subscript variable used for normalizing the summation of the denominator, normalizing the parameters of each dimension, and providing a positive scoring function for task completion time. Where T max T min These represent the maximum and minimum task completion times, T. tas The time spent by the student in the tas-th task has a positive scoring function fsf2(tas) = ​​1 - Mif for the frequency of incorrect operations. tas The function fsf3(tas) = ​​Sco has a positive score for the accuracy of the operation sequence. tas The training completion rate has a positive score function fsf4(tas) = ​​Trc tas The average response similarity of interactive operations has a positive score function fsf5(tas) = ​​Rsi tas The formula for calculating the potential ability trend value Pctv is: In the formula, λ jnd Let λ be the weighting coefficient for each dimension parameter, and λ be the weighting coefficient for each dimension parameter. jnd It is a positive number;

[0060] Set the potential ability threshold as Pth, compare the calculated potential ability trend value Pctv with the potential ability threshold Pth, and if the potential ability trend value Pctv is greater than or equal to the potential ability threshold Pth, then remove nodes in the knowledge graph whose ability level is lower than the student's current ability level.

[0061] If the potential ability trend value Pctv is less than the potential ability threshold Pth, then nodes with ability levels lower than the student's current ability level will be retained in the knowledge graph.

[0062] Using the currently completed task node as the student's starting point, traverse all valid paths originating from the starting point node that are connected and satisfy dependencies. For each valid path Pa... kin ={v1→v2→…→v fin} Marked as a candidate path, v fin For the effective path Pa kin The final task node calculates the comprehensive path score (Cps) for each candidate path. kin =δ1·Dep kin +δ2·Rhtc kin Among them, Cps kin For the comprehensive path score, δ1 and δ2 are respectively Dep kin Rhtc kin The weighting coefficients, where δ1 and δ2 are both positive, Dep kinThis represents the overall difficulty increase value for the path, Rhtc kin The correlation between the path and the trainee's historical training content;

[0063] Let each task v d ∈Pa kin The simulation parameters are Sp d The overall path difficulty increase value Dep kin The calculation expression is as follows

[0064] Let the set of tasks in the path be Ptc = {v1, v2, ..., v...} fin Each task is associated with a set of knowledge points Kno(v) d The set of tasks completed by the trainees is Com = {ct1, ct2, ..., ct...} al The completed tasks involve the following set of knowledge points: Where gt is the sequence number of the task completed by the student, and al is the total number of tasks completed by the student, then the correlation between the path and the student's historical training content is Rhtc kin The calculation expression is as follows

[0065] The comprehensive path scores of all candidate paths are compared, and the path Pa with the highest comprehensive path score is selected. hig =argmaxCps kin This is pushed to the student's AR or VR terminal as a recommended path for simulation training.

[0066] Preferably, the logic for achieving an adaptive simulation training process with real-world transferability is as follows:

[0067] The system acquires operational behavior data and device adaptability performance generated during the current training task and appends them to the student's record in a time series. It also constructs a new training sample vector and re-evaluates the student's performance across multiple capability dimensions based on the updated capability state vector. The system calculates the student's current capability level and compares it with the capability level of the previous stage. Based on the response differences between virtual tasks and actual devices during this training round, the system calculates new differential adaptation parameters and updates the differential adaptation parameter model. According to the new differential adaptation parameter model, the system selects training task nodes from the training task library that match the student's capabilities and removes task nodes that the student's current capability has not reached. Based on the differential parameters, the system adjusts the task detail parameters, performs knowledge graph traversal, path scoring, and optimal path selection, and pushes the recommended path for the next round.

[0068] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0069] Employing a multidimensional ability assessment model and a real-time classification mechanism, the system standardizes various original operational behavior indicators by constructing fuzzy membership functions. Based on each trainee's actual operational performance, it dynamically and personalized assesses their current ability level, forming standardized ability indicator values. Combined with the set multidimensional weight parameters, the system uses a score-weighted model to calculate the trainee's comprehensive ability score, achieving real-time dynamic ability level classification and effectively supporting subsequent training path recommendations and difficulty adjustments.

[0070] By constructing a differential adaptation parameter model, the differences between virtual and real devices are quantified, and these differences are incorporated into the adjustment of training tasks and the capability assessment mechanism. This allows for dynamic adjustment of training task parameters to match the complexity of real devices. Based on the learners' differential adaptation capabilities, training content is pushed forward step by step, improving the transfer effect of simulation training results in real device operation and breaking down the transfer barriers between virtual training and real operation and maintenance.

[0071] A knowledge graph is constructed based on trainees' historical training paths and equipment adaptability performance. Personalized training paths are generated through a path recommendation engine. This method continuously records trainees' training behavior trajectories and ability development dynamics, models the correlation between operational skills, knowledge mastery, and equipment adaptability, and uses graph traversal and path optimization algorithms to push task paths that match abilities, have good difference adaptability, and high growth potential. It integrates an equipment difference adaptability assessment mechanism to optimize the possibility of task migration to real operation scenarios, fully explores the factors of trainees' training history and equipment differences, and improves the scientific nature and dynamic adaptability of recommended paths. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0073] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0075] Example 1: Please refer to Figure 1As shown, the present invention is an interactive control method for virtual simulation training of power operation and maintenance. The specific steps include acquiring the trainee's operation behavior data, including task completion time, frequency of misoperation, accuracy of operation sequence, and training completion rate.

[0076] Based on the operational behavior data, a multi-dimensional ability assessment model is used to classify trainees' ability levels in real time.

[0077] By combining the structural and response behavior differences between virtual devices and target physical devices, a differential adaptation parameter model is constructed.

[0078] Based on the trainee's current ability level and the aforementioned differential adaptation parameter model, the system matches training content of corresponding difficulty from the preset training task library and automatically adjusts the parameter configuration of the simulation training task.

[0079] Based on the student's historical training path and device adaptability performance, a knowledge graph of the student is constructed, and a personalized training path is generated through a path recommendation engine.

[0080] After each round of training, the trainees' abilities are reassessed based on the newly generated behavioral data and their performance in terms of differential adaptability. The training difficulty and recommended path are then iteratively updated to achieve an adaptive and real-world transferable simulation training process.

[0081] The method for obtaining student action data is as follows:

[0082] The task completion time is calibrated as Tct. The start time is recorded as Tst when the virtual simulation task begins and the completion time is recorded as Ten when the virtual simulation task ends. The expression for calculating the task completion time is Tct = Ten - Tst. For interrupted tasks and tasks that exit midway, the task duration of the current stage is recorded.

[0083] Set up a task flowchart, which includes the correct operation steps and the operation objects. Match each student's operation with the task flowchart and record any mismatched operation as a misoperation. Define the misoperation frequency as Mif, then Mif = number of misoperations / total number of operations.

[0084] Based on the longest common subsequence algorithm, the actual operation sequence executed by the trainee is compared with the preset standard operation sequence. The standard operation sequence is labeled as S = {s1, s2, ..., s}. m}, of length m, the actual operation sequence of the trainees is U={u1,u2,…,u n}, of length n, where each element s i ∈{s1,s2,…,s m} and u j ∈{u1,u2,…,u n} represents a discrete power operation and maintenance step;

[0085] Let L(S,U) denote the longest common subsequence of sequences S and U, which satisfies and The order of elements in the longest common subsequence remains consistent, i.e., if L(S,U)={x1,x2,…,x…} k If k is the length of L(S,U), then there exists a set of indices that satisfy... and Similarly, for sequence U, k = |L(S,U)| is the longest of all common subsequences that satisfy the above conditions;

[0086] Constructing a two-dimensional matrix using dynamic programming Where D is a two-dimensional array of size (m+1)×(n+1), m is the length of the standard operation sequence, n is the length of the actual operation sequence, and each element D in the two-dimensional matrix D is a variable. i,j D represents the length of the longest common subsequence between the first i steps of the standard operation sequence and the first j steps of the student's actual operation sequence. i,j The recursive formula is Finally, we obtain the bottom right element D of the two-dimensional matrix D. m,n Let l be the length of the longest common subsequence of sequences S and U, denoted as l = |L(S,U)|;

[0087] The correctness rate of the operation sequence is then calculated. Where m = |S| is the length of the standard operation sequence S;

[0088] The training task is divided into multiple subtasks. When each subtask is completed, a labeling signal is set to mark the node as completed. The training completion rate is defined as Trc, where Trc = number of completed nodes / total number of nodes. If the task fails or is interrupted, the completion rate is output as a percentage other than 100%.

[0089] Define a multi-dimensional representation structure of the trainee's ability based on the trainee's operation behavior data. The multi-dimensional representation structure includes the following dimensions: operation efficiency dimension, operation accuracy dimension, process compliance ability dimension, and stable completion ability dimension. Define the ability status vector of the trainee as Xcs = [Tct, Mif, Sco, Trc] according to the four types of dimensions. Construct a multi-dimensional ability evaluation model based on the ability status vector through a fuzzy membership function,划定 the ability level, and define the ability score function Asf of the multi-dimensional ability evaluation model. The calculation expression of Asf is Asf = α1*f1(Tct) + α2*f2(Mif) + α3*f3(Sco) + α4*f4(Trc), where α1, α2, α3, and α4 are the weight coefficients of the parameters of each dimension in the ability status vector, and α1, α2, α3, and α4 are all positive numbers. f1(·), f2(·), f3(·), and f4(·) are the fuzzy membership functions of the parameters of each dimension in the ability status vector, and Asf represents the comprehensive ability level of the trainee.

[0090] Set the score thresholds Ath1 and Ath2, where 0 ≤ Ath1 < Ath2 ≤ 1. According to the value range of the trainee's ability score function Asf, divide the trainees into the following ability levels:

[0091] When Asf is less than or equal to Ath1, define the trainee's ability level as L1 level;

[0092] When Asf is greater than Ath1 and less than Ath2, define the trainee's ability level as L2 level;

[0093] When Ath2 < Asf, define the trainee's ability level as L3 level;

[0094] Set up a real-time update mechanism to automatically collect the trainee's operation behavior data in this round of tasks at the end of the training task, calculate the trainee's ability score function Asf in real time, re-determine the trainee's current ability. If the trainee's ability level changes compared with the previous round of evaluation results, trigger the personalized training path adjustment mechanism to recommend the task type and difficulty corresponding to the trainee's ability level.

[0095] Fuzzy membership functions are essentially standardized transformation functions for each indicator, used to map specific dimensional indicators to a value between 0 and 1, representing that the value belongs to a certain evaluation level. They achieve degree-distinguishing expressions through triangular or trapezoidal functions, standardizing raw operational data with different dimensions and distributions into a quantitative representation of a certain ability to a certain extent, connecting raw behavioral data with comprehensive ability levels. Different dimensional indicators have different weight coefficients, which are set by professionals in the field based on historical data or entropy weighting. Level L1 is the beginner level, suitable for guided training; Level L2 is the level with some operational experience; and Level L3 is the advanced operator level, suitable for complex simulation training.

[0096] By employing the aforementioned multidimensional competency assessment model and real-time classification mechanism, various original operational behavior indicators are standardized through the construction of fuzzy membership functions. Based on each trainee's actual operational performance, their current competency level is dynamically and individually assessed, forming standardized competency indicator values. Combined with the set multidimensional weight parameters, a score-weighted model is used to calculate the trainee's comprehensive competency score, achieving real-time dynamic competency level classification. This effectively supports subsequent training path recommendations and difficulty adjustments, thereby supporting targeted training content recommendations and process adaptation, and improving the adaptability and practicality of virtual simulation training.

[0097] By combining the structural and response behavior differences between virtual devices and target physical devices, a differential adaptation parameter model is constructed.

[0098] The differential adaptation parameter model is a mathematical model used to quantify the degree of difference between virtual simulation devices and target actual devices in terms of structural layout, interaction methods, dynamic response, etc. By numerically modeling the dimensions of structure and response behavior, it is used to adjust training content and interaction strategies to enhance the adaptability and transferability of simulation training to real-world scenarios.

[0099] The method for constructing a differential adaptation parameter model is as follows:

[0100] The difference parameters are defined into structural difference parameters ΔS and response behavior difference parameters ΔR. Structural difference parameters ΔS include button position offset ΔS1 and operation path branch complexity change ΔS2. Response behavior difference parameters ΔR include feedback delay difference ΔR1 and action-response mismatch degree ΔR2. After normalizing the above four difference parameters, a difference vector Δ=[ΔS1,ΔS2,ΔR1,ΔR2] is formed.

[0101] For trainees whose ability levels have been assigned, an adaptive correction function Adf(Δ) is defined. The expression for the adaptive correction function Adf(Δ) is Adf(Δ):L c →Adapt training parameters to suit the ability level L of each student.c Where c = {1, 2, 3}, given the difference parameter Δ, output a training task parameter configuration scheme that matches the trainee's ability level and equipment differences.

[0102] The logic of the training task parameter configuration scheme based on matching the output of the adaptive correction function with the trainee's ability level and equipment differences is as follows:

[0103] Based on the trainees' ability level and the difference vector Δ, matching task groups are searched from the preset training task library. The matching rules include task fitness threshold, operation complexity index and transfer difficulty assessment, and tasks that can improve practical operation ability under the influence of differences are matched.

[0104] For the matched training task, the adaptive training adjustment parameters output by the adaptive correction function Adf(Δ) are automatically configured to adjust the degree of operation guidance, the degree of response speed simulation, the tolerance range of misoperation, and the logic of displaying and hiding prompts, so as to realize the adaptive enhancement and personalized adjustment of the training task to the differences of the device.

[0105] The button position offset ΔS1 is the geometric offset of the corresponding position of the same function button in the virtual device interface and the actual device interface. It is obtained through interface layout alignment comparison. The two-dimensional coordinates of the same button in the virtual interface and the actual interface are extracted. The two-dimensional coordinates of the same button in the virtual interface are (v1, v2), and the two-dimensional coordinates of the same button in the actual interface are (a1, a2). The Euclidean distance between the two coordinates is calculated as the button position offset ΔS1. The calculation expression is as follows:

[0106] The change in operation path branch complexity ΔS2 represents the difference in interaction branch complexity experienced by the trainee during task execution between the virtual device and the actual device. Interaction branches include menu levels and sub-operation steps, obtained through directed graph analysis. The task flow of the virtual and actual devices is modeled as an operation path graph G. v and G r In the operation path graph, nodes represent operation steps, and edges represent operation flows. The expression for calculating the average branch degree BvBv of the virtual device is: The expression for calculating the path length Br of the actual device is as follows: Where the number of nodes v is the operation path graph G v The total number of nodes, i.e., the total number of operation steps, and the ∑ outgoing edge number p, represent the operation path graph G. v The number of directed edges originating from node Vp represents the number of possible next operation steps that this step can lead to. Node Vp is part of the operation path graph G. v The p-th node in the operation path graph G has a node number r. rThe set of all possible valid operation paths, where ∑ outgoing edges Vq is the operation path graph G. r The number of nodes in path Vq represents the number of operation steps required to complete the task. Path Vq is the operation path graph G. r The q-th path in the middle;

[0107] The feedback latency difference ΔR1 is the time difference between when a student completes an interactive operation and when the terminal VR or AR device returns visual, auditory, or status feedback. It represents the latency difference between the virtual and physical devices. The system log records the response time of each interactive operation. The response time of an interactive operation in the virtual device is denoted as tv, and the response time in the physical device is denoted as tr. The average latency of multiple interactive operations in both the virtual and physical devices is calculated as follows: and Among them, t v,a Let t be the feedback time after the a-th operation in the virtual device. r,a Let z be the feedback time after the a-th operation in the actual device, z be the total number of interactive operations, and a be the sequence number of the interactive operation. Then, the expression for calculating the feedback delay difference ΔR1 is:

[0108] Action-response mismatch ΔR² is a measure of the difference between the way, degree, or content of the feedback from the virtual device and the actual device after a student performs an interactive operation. An action-response map is constructed, and the similarity between the responses of each interactive operation is calculated. The action-response mismatch is obtained by defining a weighted average similarity. The action-response map includes the virtual device's action-response map Mapv and the actual device's action-response map Mapr. The expression for the virtual device's action-response map Mapv is: The expression for the actual device's action-response mapping (Mapr) is: Among them, O a For the a-th interactive operation, For virtual devices to O a Feedback For virtual devices to O a Feedback;

[0109] The logic for obtaining action-response mismatch by defining a weighted average similarity is as follows:

[0110] Calculate the similarity between the virtual device and the actual device for each action response, and obtain a weighted average similarity Sre based on the similarity. The calculation expression is as follows: In the formula, Sre is the weighted average similarity, and β a Let β be the response weight for the a-th interaction, and let β be the response weight for the a-th interaction. a Rsi is a positive number. a Let Rsi be the response similarity of the a-th interaction, and Rsi aIf the value is positive, then the expression for calculating the action-response mismatch ΔR2 is ΔR2=1-Sre;

[0111] The similarity of interactive operations is obtained through structural similarity index (SSIM) and Mel-frequency cepstral coefficient similarity. For operation responses involving visual feedback, the SSIM is used to compare the virtual device response image with the actual device response image. The method for obtaining the SSIM is as follows:

[0112] The formula for calculating the structural similarity index (SSIM) is as follows: x and y represent the pixel matrices of the virtual response image and the actual response image, respectively, μ x μ y These represent the average pixel values ​​of the image at x and y, respectively. Let σ represent the variances of the images x and y, respectively. xy C1 and C2 represent the covariance of the images x and y. C1 and C2 are small constants introduced to avoid the denominator being zero, and both C1 and C2 are positive numbers.

[0113] The image window comprises Top pixels, and the pixel value sequences of image x and image y are denoted as Pix = {Se1, Se2, ..., Se...}. Top} and Piy={Qu1,Qu2,…,Qu Top}, Se hp ∈{Se1,Se2,...,Se Top And Qu hp ∈{Qu1,Qu2,…,Qu Top}, then μ x The calculation expression is as follows μ y The calculation expression is as follows

[0114] variance of image x The calculation expression is as follows variance of image y The calculation expression is as follows Covariance σ of images x and y xy The calculation expression is

[0115] For operational responses involving auditory feedback, the Mel-frequency cepstral coefficient similarity method is used to calculate the similarity between the virtual and actual response speech signals. The method for obtaining the Mel-frequency cepstral coefficient similarity is as follows:

[0116] For a speech signal s(t), after pre-emphasis, framing, windowing, fast Fourier transform, and Mel filter bank processing, the Mel frequency cepstral coefficient vector Sfvx of the virtual response and the Mel frequency cepstral coefficient vector Sfvy of the actual response are obtained. The Mel frequency cepstral coefficient similarity between Sfvx and Sfvy is then calculated. The calculation expression is ∥Sfvx∥ and ∥Sfvy∥ represent the magnitudes of vectors Sfvx and Sfvy, respectively;

[0117] Then for interactive operation O a Response similarity Rsi a The calculation expression is Where γ is the weighting coefficient and γ is a positive number.

[0118] Traditional virtual simulation training systems for power operation and maintenance are often based on virtual equipment modeling. While their interface structure, interactive buttons, and feedback responses visually and logically simulate real equipment, structural and behavioral differences still exist. Structural differences include different button positions, simplified panel layouts, and simplified interaction path logic; behavioral differences include different system feedback delays, limited feedback types, and inaccurate voice and image feedback. These differences directly affect the transfer of training habits to real-world scenarios, leading to the problem that while trainees may be proficient in simulation, they cannot accurately replicate the results in reality. This invention constructs a difference adaptation parameter model to quantify the differences between virtual and real equipment and incorporates these differences into the adjustment of training tasks and capability assessment mechanisms. This allows for dynamic adjustment of training task parameters to match the complexity of actual equipment. Based on the trainees' ability to adapt to these differences, training content is progressively pushed, improving the transfer effect of simulation training results in real equipment operation and breaking down the transfer barriers between virtual training and real operation and maintenance.

[0119] Based on the student's historical training path and device adaptability performance, a knowledge graph of the student is constructed, and a personalized training path is generated through a path recommendation engine.

[0120] The student's knowledge graph is a dynamically updated graph structure model, labeled as Glk = (V, E), where the node set V is represented as V = {v1, v2, ..., v...}. b}, for each node v d ∈V, including task type, ability level, equipment involved, and training completion degree, and the edge set E is represented as E={e df =(v d ,v f ,Ret df )}, e df Ret represents the associated path from node d to node f. df Indicates the type of relationship from node d to node f;

[0121] It should be noted that the task types, such as high-voltage switchgear maintenance and transformer inspection, include L1, L2 and L3 capability levels, and involve equipment such as circuit breakers, terminals, and protection relays. The relationship types include priority repair relationship, skill support and performance relationship.

[0122] Using a sliding window method and a trend-weighted scoring model, based on the learner's most recent Nota operation behavior data and device adaptation performance, the potential ability trend value of the learner is evaluated. A time series is generated based on the learner's most recent Nota task completion time, error frequency, operation sequence accuracy, training completion rate, and average response similarity of interactive operations. t is the task number corresponding to the current time point, Nota is the sliding window length, tas is the index of the task number, representing the time from task t-Nota+1 to task t, and Rsi tas Let be the average response similarity of the interactive operations within the 'tas'th task, calculated as follows: Where zis is the total number of interactive operations in the tas-th task, and a is the sequence number of the interactive operation;

[0123] Define a positive score function fsf for each dimension parameter jnd (tas), and assign a time decay weight ω to each training task. tas The formula for calculating the time decay weight is: Where atf is the decay factor used to highlight recent training performance, e is the base of the natural logarithm, and jnd is the subscript variable used for normalizing the summation of the denominator, normalizing the parameters of each dimension, and providing a positive scoring function for task completion time. Where T max T min These represent the maximum and minimum task completion times, T. tas The time spent by the student in the tas-th task has a positive scoring function fsf2(tas) = ​​1 - Mif for the frequency of incorrect operations. tas The function fsf3(tas) = ​​Sco has a positive score for the accuracy of the operation sequence. tas The training completion rate has a positive score function fsf4(tas) = ​​Trc tas The average response similarity of interactive operations has a positive score function fsf5(tas) = ​​Rsi tas The formula for calculating the potential ability trend value Pctv is: In the formula, λ jnd Let λ be the weighting coefficient for each dimension parameter, and λ be the weighting coefficient for each dimension parameter. jndIt is a positive number;

[0124] Set the potential ability threshold as Pth, compare the calculated potential ability trend value Pctv with the potential ability threshold Pth, and if the potential ability trend value Pctv is greater than or equal to the potential ability threshold Pth, then remove nodes in the knowledge graph whose ability level is lower than the student's current ability level.

[0125] If the potential ability trend value Pctv is less than the potential ability threshold Pth, then nodes with ability levels lower than the student's current ability level will be retained in the knowledge graph.

[0126] Using the currently completed task node as the student's starting point, traverse all valid paths originating from the starting point node that are connected and satisfy dependencies. For each valid path Pa... kin ={v1→v2→…→v fin} is marked as a candidate path, v fin For the effective path Pa kin The final task node calculates the comprehensive path score (Cps) for each candidate path. kin =δ1·Dep kin +δ2·Rhtc kin Among them, Cps kin For the comprehensive path score, δ1 and δ2 are respectively Dep kin Rhtc kin The weighting coefficients, where δ1 and δ2 are both positive, Dep kin This represents the overall difficulty increase value for the path, Rhtc kin The correlation between the path and the trainee's historical training content;

[0127] Let each task v d ∈Pa kin The simulation parameters are Sp d The overall path difficulty increase value Dep kin The calculation expression is as follows

[0128] Let the set of tasks in the path be Ptc = {v1, v2, ..., v...} fin Each task is associated with a set of knowledge points Kno(v) d The set of tasks completed by the trainees is Com = {ct1, ct2, ..., ct...} al The completed tasks involve the following set of knowledge points: Where gt is the sequence number of the task completed by the student, and al is the total number of tasks completed by the student, then the correlation between the path and the student's historical training content is Rhtc kin The calculation expression is as follows

[0129] The comprehensive path scores of all candidate paths are compared, and the path Pa with the highest comprehensive path score is selected. hig =argmaxCps kin This is pushed to the student's AR or VR terminal as a recommended path for simulation training.

[0130] The path recommendation engine selects suitable training task paths from the knowledge graph based on the student's ability level, training history, and device adaptation performance information, ensuring that the difficulty progression of the paths is reasonable and transferable. At the same time, it automatically updates and recommends paths, including building a knowledge graph, filtering task nodes, generating candidate paths using graph traversal algorithms, implementing path evaluation strategies, and pushing the optimal path.

[0131] Traditional virtual simulation training for power operation and maintenance typically employs fixed training paths or simple hierarchical push strategies based on task module difficulty. This neglects trainees' skill development trends and training preferences, failing to dynamically adapt to equipment differences. This approach addresses this by constructing a knowledge graph based on trainees' historical training paths and equipment adaptability, and generating personalized training paths through a path recommendation engine. It continuously records trainees' training behavior and skill development, models the correlation between operational skills, knowledge mastery, and equipment adaptability, and uses graph traversal and path optimization algorithms to push task paths that match skills, have excellent adaptability, and high growth potential. By integrating an equipment adaptability assessment mechanism, it optimizes the possibility of migrating tasks to real-world operational scenarios, fully exploring trainees' training history and equipment differences, and improving the scientific rigor and dynamic adaptability of recommended paths.

[0132] After each round of simulation training tasks is completed, the trainee’s current ability level is dynamically evaluated based on the newly generated operational behavior data and equipment adaptability. The training tasks and paths for the next stage are then reconfigured based on the updated ability level, thus achieving a training closed loop with real-world transferability and personalized adaptability.

[0133] The system acquires operational behavior data and device adaptability performance generated during the current training task and appends them to the student's record in a time series. It also constructs a new training sample vector and re-evaluates the student's performance across multiple capability dimensions based on the updated capability state vector. The system calculates the student's current capability level and compares it with the capability level of the previous stage. Based on the response differences between virtual tasks and actual devices during this training round, the system calculates new differential adaptation parameters and updates the differential adaptation parameter model. According to the new differential adaptation parameter model, the system selects training task nodes from the training task library that match the student's capabilities and removes task nodes that the student's current capability has not reached. Based on the differential parameters, the system adjusts the task detail parameters, performs knowledge graph traversal, path scoring, and optimal path selection, and pushes the recommended path for the next round.

[0134] This invention employs a multidimensional ability assessment model and a real-time classification mechanism. By constructing fuzzy membership functions, it standardizes various original operational behavior indicators. Based on each student's actual operational performance, it dynamically and personalized assesses their current ability level, forming standardized ability indicator values. Combined with set multidimensional weight parameters, it uses a score-weighted model to calculate the student's comprehensive ability score, achieving real-time dynamic ability level classification and effectively supporting subsequent training path recommendations and difficulty adjustments.

[0135] By constructing a differential adaptation parameter model, the differences between virtual and real devices are quantified, and these differences are incorporated into the adjustment of training tasks and the capability assessment mechanism. This allows for dynamic adjustment of training task parameters to match the complexity of real devices. Based on the learners' differential adaptation capabilities, training content is pushed forward step by step, improving the transfer effect of simulation training results in real device operation and breaking down the transfer barriers between virtual training and real operation and maintenance.

[0136] A knowledge graph is constructed based on trainees' historical training paths and equipment adaptability performance. Personalized training paths are generated through a path recommendation engine. This method continuously records trainees' training behavior trajectories and ability development dynamics, models the correlation between operational skills, knowledge mastery, and equipment adaptability, and uses graph traversal and path optimization algorithms to push task paths that match abilities, have good difference adaptability, and high growth potential. It integrates an equipment difference adaptability assessment mechanism to optimize the possibility of task migration to real operation scenarios, fully explores the factors of trainees' training history and equipment differences, and improves the scientific nature and dynamic adaptability of recommended paths.

[0137] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0138] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0139] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An interactive control method for virtual simulation training in power operation and maintenance, characterized in that, The specific steps include acquiring the trainees' operational behavior data, including task completion time, frequency of misoperations, accuracy of operation sequence, and training completion rate; Based on the operational behavior data, a multi-dimensional ability assessment model is used to classify trainees' ability levels in real time. By combining the structural and response behavior differences between virtual devices and target physical devices, a differential adaptation parameter model is constructed. Based on the trainee's current ability level and the aforementioned differential adaptation parameter model, the system matches training content of corresponding difficulty from the preset training task library and automatically adjusts the parameter configuration of the simulation training task. Based on the student's historical training path and device adaptability performance, a knowledge graph of the student is constructed, and a personalized training path is generated through a path recommendation engine. After each round of training, the trainees’ abilities are reassessed based on the newly generated behavioral data and their performance in terms of differential adaptability. The training difficulty and recommended path are then iteratively updated to achieve an adaptive and realistically transferable simulation training process. The knowledge graph construction method for trainees is as follows: The knowledge graph is labeled Glk=(V,E), where the node set V is represented as... For each node It includes task type, capability level, equipment involved, and training completion rate, and the edge set E is represented as... , This represents the associated path from node d to node f. Indicates the type of relationship from node d to node f; The method for generating personalized training paths is as follows: Using a sliding window method and a trend-weighted scoring model, based on the learner's most recent Nota operation behavior data and device adaptation performance, the potential ability trend value of the learner is evaluated. A time series is generated based on the learner's most recent Nota task completion time, error frequency, operation sequence accuracy, training completion rate, and average response similarity of interactive operations. t is the task number corresponding to the current time point, Nota is the sliding window length, and tas is the index of the task number, indicating the starting point from the current time. From task t to task t. Let be the average response similarity of the interactive operations within the 'tas'th task, calculated as follows: , where zis is the total number of interactive operations in the tas-th task, and a is the sequence number of the interactive operation; Define a positive scoring function for each dimension parameter. And assign time decay weights to each training task. The formula for calculating the time decay weight is: Where atf is the decay factor used to highlight recent training performance, e is the base of the natural logarithm, and jnd is the subscript variable used for normalizing the summation of the denominator, normalizing the parameters of each dimension, and providing a positive scoring function for task completion time. ,in , These are the maximum and minimum times for task completion, respectively. The time spent by the student in the tas-th task has a positive scoring function for the frequency of errors. It has a positive scoring function for the correctness of the operation sequence. It has a positive scoring function for training completion. The average response similarity of interactive operations has a positive scoring function. The formula for calculating the potential ability trend value Pctv is: In the formula, These are the weighting coefficients for the parameters of each dimension, and It is a positive number; Set the potential ability threshold as Pth, compare the calculated potential ability trend value Pctv with the potential ability threshold Pth, and if the potential ability trend value Pctv is greater than or equal to the potential ability threshold Pth, then remove nodes in the knowledge graph whose ability level is lower than the student's current ability level. If the potential ability trend value Pctv is less than the potential ability threshold Pth, then nodes with ability levels lower than the student's current ability level will be retained in the knowledge graph. Using the currently completed task node as the student's starting point, traverse all valid paths originating from the starting point, connected, and conforming to dependencies. For each valid path... Marked as a candidate path, For an effective path The final task node calculates the comprehensive path score for each candidate path. ,in, To achieve a comprehensive path score, They are respectively , The weighting coefficients, and All are positive numbers. This represents the overall increase in path difficulty. The correlation between the path and the trainee's historical training content; Make each task The simulation parameters are The overall difficulty increase value of the path The calculation expression is as follows ; Let the set of tasks in the path be... Each task is associated with a set of knowledge points. The set of tasks completed by the trainees is The completed tasks involve the following set of knowledge points: Where gt is the sequence number of the task completed by the student. Given the total number of tasks completed by the student, the correlation between the path and the student's historical training content is calculated. The calculation expression is as follows ; The overall path scores of all candidate paths are compared, and the path with the highest overall path score is selected. This is pushed to the student's AR or VR terminal as a recommended path for simulation training.

2. The interactive control method for virtual simulation training of power operation and maintenance according to claim 1, characterized in that, The method for obtaining student action data is as follows: The task completion time is calibrated as Tct. The start time of the virtual simulation task is recorded as Tst, and the completion time is recorded as Ten. The calculation expression for the task completion time is Tct = Ten - Tst. For interrupted tasks and tasks that exit midway, the task duration of the current stage is recorded. Set up a task flowchart, which includes the correct operation steps and the operation objects. Match each student's operation with the task flowchart and record any mismatched operation as a misoperation. Define the misoperation frequency as Mif, then Mif = number of misoperations / total number of operations. Based on the longest common subsequence algorithm, the actual operation sequence executed by the trainee is compared with the preset standard operation sequence, and the standard operation sequence is marked as... The length is m, and the student's actual operation sequence is: A denoted , of length n, wherein each element and This represents a discrete power operation and maintenance step; set up Let S be the longest common subsequence of sequences S and U, which satisfies and The order of elements in the longest common subsequence remains consistent, that is, if k is If the length is such that there exists a set of indices satisfying the following condition: and Similarly, the same applies to sequence U. The longest of all common subsequences that satisfy the above conditions; Constructing a two-dimensional matrix using dynamic programming Where D is a value of size A two-dimensional array, where m is the length of the standard operation sequence and n is the length of the actual operation sequence, and each element in the two-dimensional matrix D... Let represent the length of the longest common subsequence between the first i steps of the standard operation sequence and the first j steps of the student's actual operation sequence. The recursive formula is Finally, the bottom right element of the two-dimensional matrix D is obtained. Let be the length of the longest common subsequence of sequences S and U, denoted as . ; The correctness rate of the operation sequence is then calculated. ,in The length of the standard operation sequence S; The training task is divided into multiple subtasks. When each subtask is completed, a labeling signal is set to mark the node as completed. The training completion rate is defined as Trc, where Trc = number of completed nodes / total number of nodes. If the task fails or is interrupted, the completion rate is output as a percentage other than 100%.

3. The interactive control method for virtual simulation training of power operation and maintenance according to claim 2, characterized in that, The method for classifying trainees' real-time ability levels is as follows: Define the trainee's ability state vector as Xcs=[Tct,Mif,Sco,Trc]. Construct a multidimensional ability assessment model based on the ability state vector using fuzzy membership functions, define ability levels, and define the ability score function Asf for the multidimensional ability assessment model. The calculation expression for Asf is: In the formula, and These are the weight coefficients of the parameters in each dimension of the capability state vector, and and All are positive numbers. and These are the fuzzy membership functions for each dimension parameter in the capability state vector. This indicates the trainee's overall ability level; Set score thresholds Ath1 and Ath2, where Based on the range of values ​​for the student ability score function Asf, students are divided into the following ability levels: When Asf is less than or equal to Ath1, the student's ability level is defined as L1. When Asf is greater than Ath1 and Asf is less than Ath2, the student's ability level is defined as L2. When Ath2 is less than Asf, the student's ability level is defined as L3. A real-time update mechanism is set up to automatically collect the trainees' operational behavior data in that round of tasks at the end of the training task, calculate the trainees' ability score function Asf in real time, re-evaluate the trainees' current ability, and trigger a personalized training path adjustment mechanism if the trainees' ability level changes from the previous round of evaluation results, recommending task types and difficulties corresponding to the trainees' ability level.

4. The interactive control method for virtual simulation training of power operation and maintenance according to claim 1, characterized in that, The method for constructing a differential adaptation parameter model is as follows: Define the dimensions of the difference parameters to divide the differences into structural difference parameters. and response behavior difference parameters Structural difference parameters Including button position offset Changes in the complexity of operation path branches Response behavior difference parameters Including feedback delay difference Action-response mismatch After normalizing the above four difference parameters, a difference vector is formed. ; Define an adaptive adjustment function Adf( for trainees whose ability levels have been classified) ), Adaptive correction function Adf( The expression for ) is For each type of student's ability level ,in Given the difference parameters Under the premise of matching the trainees' ability level and equipment differences, output a training task parameter configuration scheme. The logic of the training task parameter configuration scheme based on matching the output of the adaptive correction function with the trainee's ability level and equipment differences is as follows: Based on the trainees' ability levels and variance vectors The system searches for matching task groups from a pre-defined training task library. The matching rules include task fitness threshold, operation complexity index and transfer difficulty assessment. The system matches tasks that can improve actual operation ability under the influence of differences. For the matched training task, based on the adaptive correction function Adf( The output of adaptation training adjustment parameters automatically configures the task, adjusting the degree of operation guidance, the degree of response speed simulation, the tolerance range of misoperation, and the logic of displaying and hiding prompts, so as to realize the adaptation of training tasks to equipment differences and personalized adjustment.

5. The interactive control method for virtual simulation training of power operation and maintenance according to claim 4, characterized in that, Button position offset The method for obtaining it is as follows: Extract the 2D coordinates of the same button in both the virtual and physical interfaces. The 2D coordinates of the same button in the virtual interface are (v1, v2), and the 2D coordinates of the same button in the physical interface are (a1, a2). Calculate the Euclidean distance between the two coordinates as the button position offset. The calculation expression is: ; Changes in the complexity of operation path branches The method for obtaining it is as follows: By performing directed graph analysis, the task flows of virtual and physical devices are modeled as operation path graphs. and In the operation path graph, nodes represent operation steps, and edges represent operation flows. The expression for calculating the average branching degree Bv of the virtual device is: The expression for calculating the actual path length Br of the device is as follows: Where the number of nodes v is the operation path graph The total number of nodes, i.e., the total number of operation steps. The number of outgoing edges p is the operation path graph The number of directed edges originating from node Vp represents the number of possible next operation steps that this step can lead to. Node Vp is part of the operation path graph. The p-th node in the graph, where the number of nodes r is the operation path graph. The set of all possible valid operation paths. The number of outgoing edges, Vq, is the operation path graph. The number of nodes in path Vq represents the number of operation steps required to complete the task; path Vq is the operation path graph. The q-th path in the middle; Poor feedback delay The acquisition logic is as follows: The system logs record the response time of each interactive operation. The response time of an interactive operation in the virtual device is denoted as tv, and the response time in the actual device is denoted as tr. The average latency of multiple interactive operations in both the virtual and actual devices is calculated as follows: and ,in, Let be the feedback time after the a-th operation in the virtual device. Let z be the feedback time after the a-th operation in the actual device, z be the total number of interactive operations, and a be the sequence number of the interactive operation. Then the feedback delay difference is... The calculation expression is as follows ; Action-response mismatch The acquisition logic is as follows: Construct action-response mappings, which include action-response mappings (Mapv) for virtual devices and action-response mappings (Mapr) for physical devices. The expression for the action-response mapping (Mapv) of the virtual device is: The expression for the actual device's action-response mapping (Mapr) is: ,in, For the a-th interactive operation, For virtual devices Feedback For virtual devices Feedback; Calculate the similarity between the virtual device and the actual device for each action response, and obtain a weighted average similarity Sre based on the similarity. The calculation expression is as follows: In the formula, Sre is the weighted average similarity. Let be the response weight of the a-th interaction, and It is a positive number. Let be the response similarity of the a-th interaction, and If it is a positive number, then the action-response mismatch degree is... The calculation expression is as follows .

6. The interactive control method for virtual simulation training of power operation and maintenance according to claim 5, characterized in that, The similarity is calculated as follows: Similarity is obtained through structural similarity index (SSIM) and Mel-frequency cepstral coefficient similarity. For operation responses involving visual feedback, the SSIM is used to compare the virtual device response image with the actual device response image. The method for obtaining the SSIM is as follows: The formula for calculating the structural similarity index (SSIM) is as follows: x and y represent the pixel matrices of the virtual response image and the actual response image, respectively. , These represent the average pixel values ​​of the image at x and y, respectively. , Let x and y represent the variances of the image, respectively. This represents the covariance of the images x and y. , To avoid the small constant introduced by a denominator of zero, and , All are positive numbers; The image window comprises Top pixels, and the pixel value sequences of image x and image y are denoted as follows: and , and ,but The calculation expression is as follows , The calculation expression is as follows ; variance of image x The calculation expression is as follows The variance of the image y The calculation expression is as follows The covariance of the image x and y The calculation expression is ; For operational responses involving auditory feedback, the Mel-frequency cepstral coefficient similarity method is used to calculate the similarity between the virtual and actual response speech signals. The method for obtaining the Mel-frequency cepstral coefficient similarity is as follows: For a speech signal s(t), after pre-emphasis, framing, windowing, fast Fourier transform, and Mel filter bank processing, the Mel frequency cepstral coefficient vector Sfvx of the virtual response and the Mel frequency cepstral coefficient vector Sfvy of the actual response are obtained. The Mel frequency cepstral coefficient similarity between Sfvx and Sfvy is then calculated. The calculation expression is: , and Let Sfvx and Sfvy represent the magnitudes of the vectors, respectively. Then for interactive operations Response similarity The calculation expression is ,in, The weighting coefficients and It is a positive number.

7. The interactive control method for virtual simulation training of power operation and maintenance according to claim 1, characterized in that, The logic for achieving an adaptive simulation training process with real-world transferability is as follows: The system acquires operational behavior data and device adaptability performance generated during the current training task and appends them to the student's record in a time series. It also constructs a new training sample vector and re-evaluates the student's performance across multiple capability dimensions based on the updated capability state vector. The system calculates the student's current capability level and compares it with the capability level of the previous stage. Based on the response differences between virtual tasks and actual devices during this training round, the system calculates new differential adaptation parameters and updates the differential adaptation parameter model. According to the new differential adaptation parameter model, the system selects training task nodes from the training task library that match the student's capabilities and removes task nodes that the student's current capability has not reached. Based on the differential parameters, the system adjusts the task detail parameters, performs knowledge graph traversal, path scoring, and optimal path selection, and pushes the recommended path for the next round.

Images