Virtual reality-based reciprocating compressor system operation simulation training method

Abstract

The invention discloses a reciprocating compressor system operation simulation training method based on virtual reality, and particularly relates to the technical field of industrial simulation training, and the method comprises the steps: constructing a three-dimensional simulation scene consistent with an equipment structure, and configuring a standard operation process and an interaction object; user interaction training is executed, and sight, path and response data are collected in real time; identifying a behavior offset and triggering an evaluation process; calculating a cognitive offset index and an interaction consistency index to reflect user behavior characteristics; inputting the two indexes into a pre-training model, evaluating the behavior quality, and when the result is lower than an accuracy threshold, dynamically adjusting the layout density of the target object to optimize the interaction experience; according to the method, the actual operation process is reproduced by constructing the simulation scene, the user behavior data is collected in real time to identify the offset, evaluation is carried out based on the cognitive offset index and the interaction consistency index, the target object layout density is automatically adjusted when the behavior is abnormal, and closed-loop optimization of the training effect is achieved.

Description

Technical Field

[0001] This invention relates to the field of industrial simulation training technology, and more specifically, to a virtual reality-based simulation training method for operating reciprocating compressor systems. Background Technology

[0002] Reciprocating compressors, also known as API compressors, are a type of high-energy-consuming, high-precision mechanical equipment commonly used in industries such as chemical, petroleum, natural gas, power, and pharmaceuticals. They primarily compress and transport gas through the reciprocating motion of a piston within a cylinder. API compressors are widely used in critical industrial scenarios where safety and stability requirements are extremely high, such as gas pipelines and natural gas purification and synthesis processes.

[0003] Reciprocating compressors typically operate in conjunction with a control system during normal operation, using components such as buttons, knobs, and displays to set parameters, monitor status, and respond to faults. In practical applications, due to the complex structure of these systems, the multi-step sequential operation process, and the extremely low tolerance for operational errors, highly skilled and responsive operators are required. Traditional training methods rely on hands-on equipment drills, which suffer from numerous problems, including limited equipment resources, high operational risks, and delayed training feedback.

[0004] In recent years, virtual reality (VR) technology has been increasingly applied to industrial training scenarios, using 3D modeling and interactive systems to virtualize physical equipment, thereby providing operators with an immersive and highly controllable training environment. However, most existing VR training systems only focus on reproducing operational procedures, lacking the ability to identify deviations in user behavior and failing to establish a feedback mechanism between behavioral quality and interface optimization. Therefore, this invention proposes a virtual reality-based simulation training method for reciprocating compressor systems to address the aforementioned problems. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: A virtual reality-based simulation training method for reciprocating compressor system operation includes the following steps: Constructing a 3D simulation scenario: Based on the actual structure and control logic of the reciprocating compressor system, a virtual interactive space is created, and standard operating procedures and interactive objects consistent with the actual training objectives are configured; Execute interactive training: Users complete preset tasks in a virtual space. During the process, the user's gaze trajectory, action path, operation response sequence and input characteristics are collected in real time to build behavioral data records. Identify Deviation Behavior: When a user's behavior is detected to deviate from the standard operation path beyond the preset tolerance, the interaction behavior evaluation process is triggered. This deviation includes, but is not limited to, path discontinuity, response lag, or target misjudgment. Calculate behavioral indices: Based on the collected data, construct two evaluation indices. The first is the cognitive bias index, which reflects the spatial mismatch between the user's focus and the target object. The second is the interaction consistency index, which represents the degree of consistency between the user's operation behavior and the standard operation path. Execution interaction optimization: The cognitive bias index and the interaction consistency index are jointly input into the pre-trained model to output the operation behavior evaluation result. When the result is lower than the set accuracy threshold, the layout density of the target object is dynamically adjusted.

[0006] In a preferred embodiment, the user operation behavior data collected during interactive training simultaneously records the task number, operation step sequence identifier, and timestamp information, and establishes a one-to-one mapping relationship with the name of the target interactive object in the 3D simulation scene to form a complete operation data sequence for time-series behavior analysis. This data sequence is used as an input data source for calculating the cognitive bias index and interaction consistency index after the interactive behavior evaluation process is triggered, and is also used as a subsequent training sample to supplement the training sample set for incremental model updates.

[0007] In a preferred embodiment, identifying deviation behavior includes making a comprehensive judgment based on the user's operation path, instruction call order, response latency, and target recognition behavior during the interactive training process. The instruction call sequence is compared with the standard task path in an ordered sequence. The longest common subsequence algorithm is used to calculate the instruction matching rate S. S is equal to the length of the longest common subsequence between the user's actual instruction call sequence and the standard instruction sequence divided by the length of the standard sequence. When S is less than a preset threshold, it is determined that the instruction order is abnormal. The response delay is evaluated using the difference Δσ = |σ-σ0| between the standard deviation of the response time for each operation step σ and the standard deviation of the standard operation time σ0. When Δσ exceeds the preset threshold two, it is determined to be an abnormal response delay. When any one of the following occurs simultaneously—an abnormal instruction sequence, an abnormal response delay, or a path similarity matching value below a preset threshold—it is determined that the current behavior meets the offset condition for triggering the interaction behavior evaluation process.

[0008] In a preferred embodiment, path similarity comparison employs a method based on the shortest edit distance, which performs structured sequence parsing on continuous action segments and then performs matching to obtain path similarity matching values.

[0009] In a preferred embodiment, the calculation of the cognitive shift index includes the following steps: The first step is to extract the gaze point sequence formed by the user during the interactive training process. The gaze point sequence consists of the coordinates of the end point of the user's gaze trajectory in the three-dimensional simulation scene and the corresponding timestamp, which is used to reflect the changes in gaze position during task execution. The second step is to construct the target position function, which represents the geometric center coordinates of the target interactive object in three-dimensional space corresponding to each operation step. The target position function is located by the task number and operation step sequence identifier to realize the matching relationship between the gaze point and the target object. The third step is to generate a gaze distance sequence. Each element of the gaze distance sequence represents the minimum Euclidean distance between the user's gaze point and the spatial coordinates of the corresponding target interactive object within a certain operation step time period, which is used to measure the actual gaze offset of the step. The fourth step is to standardize the gaze distance sequence by using the maximum spatial distance between all interactive objects in the simulation scene as the normalization benchmark to obtain the gaze offset value sequence, where each gaze offset value is a standardized spatial offset score value. Fifth step: Taking each operation step as the center, analyze and construct the gaze stability function. The gaze stability function calculates the autocorrelation of the gaze distance change trend between consecutive steps through a sliding window method, and generates a stability score that reflects the degree of gaze fluctuation. Step 6: Calculate the cognitive shift index. The cognitive shift index is the arithmetic mean of the product of each fixation shift value and the complementary value of the fixation stability function score for that step in all operation steps. The complementary value is defined as a minus the fixation stability function score, which is used to represent the degree of gaze dispersion. Finally, the average of the products of all steps is taken as the output result of the cognitive shift index.

[0010] In a preferred embodiment, the calculation of the interaction consistency index includes the following steps: Data metrics for determining behavioral deviations are extracted, including the command matching rate, response latency standard deviation difference, path similarity matching value, and behavioral sequence difference distance generated during user operations. The command matching rate represents the consistency between the user's command order and the standard task path. The response latency standard deviation difference is the difference between the standard deviation of the user's operation time series and the standard deviation of the standard response time. The path similarity matching value is the structural similarity between the user's operation path and the standard path. The behavioral sequence difference distance is the dynamic registration cost between the user's operation event sequence and the standard event sequence. The behavioral sequence difference distance is calculated based on the dynamic time warping algorithm. The difference in standard deviation of response delay and the distance of behavioral sequence difference are normalized to obtain the consistency score of operation rhythm and the consistency score of behavior structure. The consistency score of operation rhythm is one minus the difference in standard deviation of response delay, and the consistency score of behavior structure is one minus the distance of behavioral sequence difference. Both are used to measure the complementarity of various behavioral biases. The interaction consistency index is the fourth root of the product of instruction matching rate, operation rhythm consistency score, path similarity matching value and behavior structure consistency score. The four factors represent the interaction consistency characteristics of the four dimensions of sequence, duration, path and behavior structure, respectively. The index value is used to measure the degree of consistency between the overall user operation behavior and the standard behavior.

[0011] In a preferred embodiment, the pre-trained model is a multi-layer neural network structure built based on supervised learning methods. The training samples are derived from standard behavior datasets and non-standard behavior datasets with operational experience labels. The standard behavior datasets are derived from the behavior sequences recorded by actual operators when completing tasks in a 3D simulation scene. The non-standard behavior datasets include user data from beginners or users with operational errors. During training, the input is a joint feature vector composed of the cognitive bias index and the interaction consistency index, and the output is a numerical label of the operational behavior evaluation result, which represents the level of difference between the current behavior and the standard behavior.

[0012] In a preferred embodiment, during the interaction optimization process, when the evaluation result of the operation behavior is lower than the set accuracy threshold, the spatial distribution parameters of the interactive objects in the 3D simulation scene are adjusted. The adjustment method is to modify the average distribution density of the interactive objects in the unit visible area of ​​the scene interface. The adjustment ratio of the layout density is calculated by the cognitive offset index and the interaction consistency index. The calculation method is as follows: Let the cognitive offset index be D1 and the interaction consistency index be D2. Then the layout expansion factor K is calculated as K=1+α×D1+β×(1-D2), where α and β are preset empirical adjustment coefficients with values ​​ranging from 0 to 1. K is used as the magnification factor of the original minimum boundary spacing R to reset the boundary buffer R'=K×R of each interactive object. Finally, the arrangement distance between interactive objects is updated to realize the dynamic adjustment of the layout density of the target object.

[0013] The technical effects and advantages of this invention are as follows: This invention constructs a simulation training environment that closely matches the actual equipment structure and operating procedures, effectively enabling visualized training of user operating skills and standardized learning of procedures. By constructing a 3D simulation scene based on the actual structure and control logic of a reciprocating compressor system, and configuring standard operating procedures and interactive objects consistent with the actual training objectives within it, this invention provides a virtual interactive space highly consistent with the real operating process. This allows users to master operating skills through immersive training without touching the physical equipment. This method not only improves the safety and efficiency of training but also helps users establish correct operating paths and target recognition capabilities under the guidance of standardized procedures.

[0014] This invention achieves accurate recording and deviation judgment of the entire user operation process by real-time acquisition of key behavioral data such as gaze trajectory, movement path, and operation response. During user interactive training, this invention collects behavioral data in real time, covering gaze movement trajectory, operation path, and the response timing and input characteristics of each step. This multi-dimensional behavioral data collection provides detailed evidence for subsequent judgment of whether the user deviates from the standard operation path. When the system detects that the user's behavior deviates from the standard path beyond the preset tolerance, such as path interruption, response lag, or target misjudgment, it can promptly trigger the interactive behavior evaluation process, thereby achieving immediate response and intervention for abnormal behavior.

[0015] This invention introduces a dual-exponential behavior evaluation and adaptive layout adjustment mechanism to construct a closed-loop feedback mechanism for training effects, thereby achieving personalized optimization of the user's training path and dynamic adjustment of operational burden. After behavior data collection and deviation identification, this invention constructs a cognitive deviation index and an interaction consistency index to measure the spatial deviation of the user's focus and the consistency between the operational behavior and the standard path, respectively. These two indices are jointly input into the pre-trained model, outputting a quantitative evaluation of the current operational behavior. When the evaluation result is lower than the accuracy threshold set by the system, the system dynamically adjusts the layout density of target objects in the current interactive interface, thereby reducing operational complexity, guiding the user to focus more on key target areas, and forming a closed-loop matching between training content and behavioral capabilities. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of the virtual reality-based reciprocating compressor system operation simulation training method of the present invention. Detailed Implementation

[0017] 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.

[0018] Reference Figure 1 The following examples were obtained: Example 1: A virtual reality-based simulation training method for reciprocating compressor system operation, comprising the following steps: Constructing a 3D Simulation Scene: In this invention, a model is first built based on the actual physical structure and control flow of the reciprocating compressor (API) system, constructing an interactive 3D virtual simulation scene. This scene not only reproduces the external outline and component layout of the compressor cabinet, but also integrates the logical functions of common operating components such as control buttons, status indicators, and panel switching areas, enabling users to complete the entire process of system operation in virtual space. The simulation environment also presets standard operating paths, namely a set of typical operating task sequences set according to industrial specifications, including processes such as startup, parameter setting, anomaly handling, and shutdown. This scene serves as the basic environment for interactive training, providing a standard space and action benchmark for subsequent operation behavior acquisition, offset recognition, and performance evaluation.

[0019] Interactive Training: After the 3D simulation scene is constructed, the user enters the interactive training phase guided by the system. During training, the user needs to complete a series of preset task steps in the virtual space. These task steps are derived from the actual system operation process and have industrial practical significance. While the user completes the operation tasks, the system collects their operation behavior data in real time with high temporal resolution, including eye trajectory, hand operation path, button response time, instruction execution sequence, and input action characteristics. This data constitutes the original set of user behavior data and is linked to task number, step identifier, and interaction object mapping relationship in the data structure, providing a data foundation for subsequent behavior deviation judgment and behavior quality index calculation.

[0020] Identifying Deviations: During user interaction training, to ensure that behavior quality remains consistent with the standard path, the system continuously monitors the user's real-time behavior and dynamically compares it with the standard operation path. When a deviation exceeding a preset tolerance is detected in the user's behavior compared to the standard path, the interaction behavior evaluation process is triggered. Deviation identification serves as a trigger point to determine whether a more refined behavior index evaluation stage is needed to quantify the current operation quality.

[0021] Calculating Behavioral Indices: Once the behavior assessment process begins, the system will structure the previously collected user action data and construct two behavior quality indices: the Cognitive Deviation Index and the Interaction Consistency Index. The Cognitive Deviation Index is primarily based on the spatial shift between the user's gaze point and the target object in each task step, as well as gaze stability characteristics, reflecting the degree of spatial misalignment in the user's attention. The Interaction Consistency Index, on the other hand, quantifies the consistency between user behavior and the standard path from multiple dimensions, including the sequence of operation steps, the rhythm of operation response, path structure, and behavioral event registration. Together, these two indices form a comprehensive description of operational deviations, providing indicator support for behavior optimization.

[0022] Interaction optimization is performed by combining the cognitive bias index and the interaction consistency index into a feature vector, which is then input into a pre-trained behavior evaluation model. The model outputs a quality score for the current operation. When this score is lower than the system's set accuracy threshold, it indicates that the user has significant cognitive bias or execution inconsistency in the current task, and the system initiates the interaction optimization mechanism. During optimization, the focus is on areas where users have difficulty identifying targets. The layout density of interactive objects in the 3D simulation scene is dynamically adjusted, specifically the relative distance and arrangement density between interactive objects in the spatial interface, making them easier to distinguish visually and in terms of operational behavior. This reduces the user's error rate in subsequent training and improves cognitive matching and execution efficiency.

[0023] In the interactive training process of this invention, to support the quality assessment and adaptive optimization of subsequent operational behaviors, the system performs full-process data collection and structured recording of user operational behaviors. The collected user operational behavior data not only includes process information of the user performing interactive actions in the 3D simulation scene, but also includes task-level and step-level positioning identifiers, which are used to realize the binding relationship between temporal behaviors and spatial objects, and build a high-quality behavioral data foundation.

[0024] The task number is a unique identifier assigned by the system in each round of training, representing a complete interactive task unit. This number is used to classify and track different operation sessions, ensuring the integrity and independence of the data during later processing. Secondly, the operation step sequence identifier refers to the sequential encoding of each sub-step under the task number. This encoding clarifies the specific location of the user's operation within the task flow, allowing the data to be mapped to the corresponding steps in the standard operation process. Thirdly, the timestamp information refers to the specific time point of each user operation event. The system records these time points using high-frequency sampling to reflect characteristics such as the temporal continuity, response latency, and step rhythm of user behavior.

[0025] In the 3D simulation scene, the system has configured a unique object name for each interactive element, representing a specific virtual device, button, indicator area, or other human-computer interaction component. To structurally map the user's actual actions to the interaction targets in the standard process, this invention establishes a one-to-one mapping relationship between action data and target interaction object names. The mechanism for establishing this mapping relationship is as follows: when a user completes a valid operation (e.g., click, rotate, swipe, etc.), the system spatially matches the user's interaction position coordinates with the standard target object name indicated by the current task step, and records the target name as the object identifier for this operation. Through this mapping mechanism, the system can reconstruct a complete "time-step-target" ternary action trajectory. An integrated data set containing task number, operation step sequence identifier, timestamp information, and target interaction object name is uniformly constructed into a complete action data sequence. This action data sequence not only possesses spatial accuracy and temporal continuity but also supports fine-grained reconstruction of behavioral features, and can be directly used in subsequent behavior evaluation processes.

[0026] When the system detects that a user's action has triggered evaluation conditions (such as path anomalies or response delays) while identifying deviation behavior, it will automatically retrieve the previously recorded complete sequence of operation data as the input data source to calculate the cognitive offset index and the interaction consistency index. Specifically, the cognitive offset index is calculated by analyzing the spatial offset trend and stability characteristics between the gaze point and the target object; the interaction consistency index comprehensively evaluates the consistency of operation behavior through factors such as the order of instruction invocation, response rhythm, and behavior registration distance. The calculation of both indices is highly dependent on the timestamps and step sequence structure in the operation data sequence.

[0027] This invention supports a mechanism for continuous model optimization during interactive training. After the aforementioned operational data sequences have been used for evaluation, they will be added to the training sample set as training samples for incremental model updates. This training sample set includes labeled standard and non-standard behavioral data. The system incorporates new behavioral samples into the training set after classifying them according to their scoring results, and fine-tunes the structure of the original evaluation model through supervised learning. This allows the model to better adapt to changes in user skill levels, task complexity, and operational habits, improving the model's generalization ability and evaluation stability.

[0028] In the operation simulation training method of this invention, to improve the accuracy of judging user operation behavior and enhance the controllability of behavior quality assessment, the system sets up a "behavior deviation recognition" mechanism. This mechanism monitors the user's operation behavior in real time during the interactive training task, performing multi-dimensional feature monitoring, and makes a comprehensive judgment based on the difference between the operation behavior data and the standard task path. When the judgment result shows that the deviation exceeds the set tolerance, the subsequent interactive behavior assessment process is triggered.

[0029] Behavioral deviation identification includes, but is not limited to, the following four dimensions: operation path, instruction call order, response latency, and target recognition behavior. The system constructs an evaluation process based on these four dimensions to accurately determine whether a user exhibits non-standard operational behavior during a particular step or task.

[0030] In terms of instruction call order, the system extracts all operation instructions executed by the user in the current interactive task and constructs a user instruction sequence in chronological order. This user instruction sequence is then compared with the standard instruction sequence defined in the standard task path. To quantify the structural similarity between the two sequences, this invention uses the Longest Common Subsequence (LCS) algorithm for comparison. Specifically, if the length of the standard sequence is N, and the length of the longest common subsequence between the user instruction sequence and the standard instruction sequence is L, then the instruction matching rate S is calculated, defined as S = L / N. The value of S ranges from 0 to 1; the closer the value is to 1, the more consistent the instruction order. When S is less than a preset threshold, the system determines that there is an instruction order anomaly.

[0031] In the response latency dimension, the system analyzes the fluctuations in response time when users interact with objects at each step. The response time for each step is defined as Ti, the standard deviation of the operation time for the entire task sequence is σ, and the standard deviation of the response time for each step in the standard task flow is σ0. The difference between the two, Δσ, is calculated as Δσ = |σ - σ0|. When this difference Δσ is greater than a system-defined threshold, it indicates that the user exhibits significant fluctuations in timing during this training session, and the system determines that there is an anomaly in response latency.

[0032] Further comparison of the completeness and structural order of the user's execution path is used to determine whether the spatial behavior execution conforms to the task path standard. Path comparison is based on the time series of continuous operation actions in the user behavior data, extracting the interaction action type of each step (such as click, rotate, drag, etc.) and constructing these actions into a symbol sequence. Then, it is transformed into a continuous action segment sequence through structured parsing. This sequence is structurally aligned with the corresponding action segment sequence in the standard path. The system uses the shortest edit distance algorithm (LevenshteinDistance) to calculate the edit distance between the two and maps this distance to a path similarity matching value, which also ranges from 0 to 1. The higher the value, the more consistent the path execution. When the path similarity matching value is lower than the set threshold of three, it is considered that there is a structural deviation in the path.

[0033] Based on the comprehensive judgment criteria, if any one of the above three indicators—abnormal instruction sequence (i.e., S is below threshold one) or abnormal response delay (i.e., Δσ exceeds threshold two)—is met, and the path similarity matching value is also below threshold three, then the system determines that the user's behavior constitutes a "significant deviation," meeting the behavior deviation triggering condition, and immediately initiates the interaction behavior evaluation process. In subsequent steps, this evaluation process will calculate the cognitive deviation index and interaction consistency index based on the collected operation data sequence, serving as important basic inputs for judging the quality of user operation behavior and further driving the interaction optimization module. For example, when a user performs a system operation training task, the standard path includes: "Power button → Parameter setting knob → Confirm start → Read status indicator → Execute shutdown process," and the system records this standard sequence as 5 steps. If a user's operation path omits "Read status indicator" and repeatedly performs "Parameter setting," the instruction matching rate S drops to 0.6, and due to repeated operations, the response time standard deviation Δσ is significantly higher than the normal value. Simultaneously, the edit distance calculation shows a path similarity matching value of only 0.68. Therefore, these three indicators meet the deviation judgment logic, and the system enters the evaluation process accordingly.

[0034] The calculation of the cognitive shift index includes the following steps: The first step is to extract the gaze point sequence formed by the user during the interactive training process. The gaze point sequence consists of the coordinates of the end point of the user's gaze trajectory in the three-dimensional simulation scene and the corresponding timestamp, which is used to reflect the changes in gaze position during task execution. The second step is to construct the target position function, which represents the geometric center coordinates of the target interactive object in three-dimensional space corresponding to each operation step. The target position function is located by the task number and operation step sequence identifier to realize the matching relationship between the gaze point and the target object. The third step is to generate a gaze distance sequence. Each element of the gaze distance sequence represents the minimum Euclidean distance between the user's gaze point and the spatial coordinates of the corresponding target interactive object within a certain operation step time period, which is used to measure the actual gaze offset of the step. The fourth step is to standardize the gaze distance sequence by using the maximum spatial distance between all interactive objects in the simulation scene as the normalization benchmark to obtain the gaze offset value sequence, where each gaze offset value is a standardized spatial offset score value. Fifth step: Taking each operation step as the center, analyze and construct the gaze stability function. The gaze stability function calculates the autocorrelation of the gaze distance change trend between consecutive steps through a sliding window method, and generates a stability score that reflects the degree of gaze fluctuation. Step 6: Calculate the cognitive shift index. The cognitive shift index is the arithmetic mean of the product of each fixation shift value and the complementary value of the fixation stability function score for that step in all operation steps. The complementary value is defined as a minus the fixation stability function score, which is used to represent the degree of gaze dispersion. Finally, the average of the products of all steps is taken as the output result of the cognitive shift index.

[0035] Specifically, based on user operation behavior data, a gaze point sequence is extracted. The gaze point sequence is a set of coordinates of the endpoint of the gaze trajectory and the corresponding timestamp recorded during the interactive training process, denoted as F(t), where t represents the time point in the task, and F(t) is the three-dimensional gaze coordinate at that moment. Construct a target position function P(i), where P(i) is the spatial center coordinate of the target interactive object that the user should look at in step i, defined as P(i)=(xᵢ,yᵢ,zᵢ), where i is the task operation step number and the three-dimensional coordinate is the geometric center point of the target object in the three-dimensional simulation scene for that step; Construct a gaze distance sequence D(i), where D(i) represents the minimum Euclidean distance between the user's gaze point and the target object in step i, calculated as follows: ; Indicates the time in step i 3D gaze coordinates Indicates time The time range between step i and step i+1. The three-dimensional Euclidean distance between the spatial center coordinates of the target interactive object that should be viewed at different times in step i and during the execution phase of step i. Generate the gaze stability function R(i,k), where k is the sliding window width, and R(i,k) is defined as the autocorrelation coefficient of D(i) over k time intervals near step i. The calculation method is as follows: ; Indicates a time index. Within the sliding window The average value; The stability of the gaze distance variation is represented by the analysis of gaze behavior changes in this method; The cognitive shift index is finally calculated using the following method: n is the total number of steps. The formula represents the maximum spatial distance between all interactive objects in the simulation scene. The first term measures the degree of offset, and the second term measures gaze stability. Together, they form the cognitive offset assessment value. .

[0036] The calculation of the interaction consistency index includes the following steps: Data metrics for determining behavioral deviations are extracted, including the command matching rate, response latency standard deviation difference, path similarity matching value, and behavioral sequence difference distance generated during user operations. The command matching rate represents the consistency between the user's command order and the standard task path. The response latency standard deviation difference is the difference between the standard deviation of the user's operation time series and the standard deviation of the standard response time. The path similarity matching value is the structural similarity between the user's operation path and the standard path. The behavioral sequence difference distance is the dynamic registration cost between the user's operation event sequence and the standard event sequence. The behavioral sequence difference distance is calculated based on the dynamic time warping algorithm. The difference in standard deviation of response delay and the distance of behavioral sequence difference are normalized to obtain the consistency score of operation rhythm and the consistency score of behavior structure. The consistency score of operation rhythm is one minus the difference in standard deviation of response delay, and the consistency score of behavior structure is one minus the distance of behavioral sequence difference. Both are used to measure the complementarity of various behavioral biases. The interaction consistency index is the fourth root of the product of instruction matching rate, operation rhythm consistency score, path similarity matching value and behavior structure consistency score. The four factors represent the interaction consistency characteristics of the four dimensions of sequence, duration, path and behavior structure, respectively. The index value is used to measure the degree of consistency between the overall user operation behavior and the standard behavior.

[0037] Specifically, the process unfolds from four dimensions, each derived from: Instruction sequence matching rate S, which assesses the consistency between the user's execution steps and the standard task path (S ranges from 0 to 1, with higher values ​​indicating higher sequence consistency); Response delay standard deviation difference Δσ, the difference between the standard deviation of the user's operation response time sequence and the standard deviation of the standard operation time, representing the degree of fluctuation in operation rhythm; Path similarity matching value MatchSim, calculated using the shortest edit distance to determine the similarity between the user's path and the standard path, reflecting the consistency of spatial behavior execution; Constructing user behavior sequences and standard behavior sequences: Each user action is abstracted into a symbol sequence (e.g., clicking A, rotating B, dragging C), forming a time-stamped behavior trajectory sequence. A dynamic time warping algorithm is used to flexibly align these two sequences: The original algorithm is used for minimum-cost registration of non-equal-length sequences; in this method, it is modified by adding a behavior event type constraint factor, allowing only registration between behavior events of the same type and prohibiting alignment between semantically different operation steps; Generating a "behavior difference distance" D_diff, representing the overall difference in user operation behavior in terms of event type and timing.

[0038] To ensure consistency in the dimensions and direction of different indicators, the following processing is performed: The instruction matching rate S remains unchanged, with a value range of 0-1; the response latency difference Δσ, after normalization, becomes T_norm=1-Δσ / Δσ_max; the path similarity matching value MatchSim remains unchanged, with a value range of 0-1; the behavior difference distance D_diff, after normalization, becomes B_norm=1-D_diff / D_max; T_norm and B_norm represent the consistency of operation time and the consistency of behavior structure, respectively. The higher the value, the stronger the consistency. Δσ_max and D_max are different preset normalization benchmark values. The final interaction consistency index D2 is calculated as: D2=(S×T_norm×MatchSim×B_norm)^(1 / 4). Using the idea of ​​weighted nonlinear averaging, a "balanced product model" is applied, and the fourth root is used to smooth out multidimensional fluctuations, avoiding a single indicator dominating the overall output and adapting to the complex characteristics of user interaction behavior.

[0039] In the operational behavior evaluation mechanism of this invention, the final determination of behavior quality relies on a pre-trained model constructed through supervised learning. This model intelligently discriminates the behavioral data generated by the user during interactive training and outputs quantitative evaluation results. The model structure is based on a multi-layer neural network architecture commonly used in artificial intelligence, recognizing complex behavioral patterns through hierarchical abstraction and feature transformation capabilities. The model is constructed using supervised learning, meaning that during training, the model's parameters are optimized using a clearly labeled training dataset to establish a mapping relationship between the input and the desired output. To achieve this goal, this invention constructs two independent datasets for supervised learning: a standard behavior dataset and a non-standard behavior dataset.

[0040] Standard behavior datasets refer to collections of data consisting of behavioral sequences generated by experienced personnel (such as professionally trained operators) during the completion of target tasks in a 3D simulation scenario. This type of data primarily features behaviors that closely approximate standard operating procedures, including multi-dimensional features such as task number, operation sequence, time response, and target object selection, and is labeled with "high consistency" and "high accuracy" by professionals. In contrast, non-standard behavior datasets include data generated by beginners or users who exhibit cognitive errors, operational delays, or misaligned steps during task execution. This type of data is labeled as "behavioral deviation" samples by the system or evaluators, reflecting a certain difference between the behavior and the standard path.

[0041] In the specific model training process, the system uses two key quantitative indicators of user operation behavior in a complete training task as input: the cognitive bias index and the interaction consistency index. These two indices are calculated by the evaluation module mentioned above. The former reflects the degree of spatial mismatch between the user's gaze and the target object, while the latter measures the consistency between the user's step sequence, response time, and behavioral path in the operation process and the standard path. The two-dimensional vector formed by these two indices is combined to form a joint feature vector, which serves as the input to the neural network.

[0042] After training, the neural network model performs forward propagation on the joint feature vector and outputs a numerical label. This label can be a continuous real value or a hierarchical discrete value, used to represent the degree of difference between the current user's operation and the standard behavior. For example, an output label of 0 indicates that the behavior fully conforms to the standard process, a label of 1 indicates that the behavior has a small deviation, and a label of 2 or above indicates that there is a significant cognitive mismatch or operational confusion.

[0043] To further illustrate, the following is a typical implementation example: Suppose a user performs a standard task with 10 steps in a simulation scenario. After collecting behavior data, the system calculates the following: a cognitive bias index of 0.42 (indicating some lack of focus) and an interaction consistency index of 0.65 (indicating a difference in the operation sequence and rhythm compared to the standard). This joint feature vector [0.42, 0.65] is input into a pre-trained multi-layer neural network, and the model outputs a numerical label of 1.7. According to the system definition, this score falls within the "moderate deviation" range, indicating that the user exhibits significant cognitive load and path error in the current task. This will trigger an interaction optimization mechanism to adjust the interface layout parameters.

[0044] In the virtual simulation training method of this invention, to improve the adaptability of the training system to deviations in user operation behavior, the system sets up an adaptive adjustment mechanism for the interactive interface driven by evaluation results, i.e., an interaction optimization process. This optimization mechanism is based on the output results of the operation behavior evaluation model. When the quality of user behavior is determined to be below the acceptable threshold set by the system, the system proactively adjusts the spatial layout of the current interactive training environment to enhance the user's clarity of understanding of the interaction objective and the feasibility of the operation path.

[0045] In the specific implementation process, the system first judges the results output by the evaluation model. When the evaluation result of the operation behavior is lower than the set accuracy threshold, that is, the model judges that the current behavior deviates significantly from the standard, the system initiates the layout adjustment logic. At this time, the optimization target focuses on all interactive components in the 3D simulation scene, that is, the interface objects that the user can operate. These objects are collectively referred to as interactive objects in this invention, and their scope includes virtual elements that can be perceived and acted upon simultaneously by the user's line of sight and hand movements, such as virtual device control buttons, knobs, switches, status display components, and functional blocks. In the training scenario, the target object refers to the specific interactive object that the user needs to complete the operation for the current task step, and is the focus of the current step in the set of interactive objects.

[0046] To enhance the recognizability and distinguishability of these interactive objects, the system adaptively adjusts their spatial distribution parameters within the 3D scene interface. These spatial distribution parameters specifically refer to the space occupied by each interactive object within the user's field of view and its screen occupancy rate. By controlling these parameters, the system dynamically changes the average distribution density within a unit of visible area. This means that while maintaining the interface proportions, the system appropriately increases or decreases the distance between interactive objects, thereby avoiding visual clustering, overlapping interactions, or accidental touches.

[0047] This invention introduces two evaluation metrics during layout adjustment: the cognitive shift index (D1) and the interaction consistency index (D2). The cognitive shift index reflects the degree of matching between the user's visual focus and the actual spatial location of the target object; a higher value indicates more scattered attention. The interaction consistency index measures the consistency of the user's operation sequence, timing, and behavioral path with the standard task; a higher value indicates stronger operational stability and standardization. The system inputs both D1 and D2 into the calculation formula to construct a variable used to control the degree of layout adjustment, called the layout expansion factor K. The calculation formula for this factor is: K = 1 + α × D1 + β × (1 - D2), where α and β are empirical adjustment coefficients used to control the contribution weights of D1 and D2 in the K value, and their values ​​range from 0 to 1. When D1 is high (large cognitive bias), the value of α × D1 increases, pushing K upward; when D2 is low (poor operational consistency), the value of (1 - D2) increases, and β × (1 - D2) also pushes K to grow; the final result K serves as the amplification factor for the arrangement parameters of each interactive object in the current scene. To achieve actual adjustment, the system sets a boundary buffer for each interactive object. This buffer refers to the minimum safe distance R reserved between objects and adjacent objects in the 3D interface (used to avoid interface operation interference or overlap). The calculated K value is multiplied by the original boundary distance R as an amplification factor to obtain the updated distance value R': R' = K × R, thereby controlling the interface interaction density by increasing the spatial interval without changing the object itself.

[0048] For example, during training, a user's evaluation results are D1=0.35, D2=0.60. Taking α=0.5 and β=0.4, then K=1+0.5×0.35+0.4×(1-0.6)=1+0.175+0.16=1.335. If the original minimum boundary spacing R=20 pixels, then after adjustment R′≈26.7 pixels. The system will spread out the interactive objects in the scene, reducing clustering and helping users more clearly identify target objects, thus improving the accuracy of subsequent operations.

[0049] This mechanism establishes a closed-loop mechanism between cognitive features and interactive interface adaptation by rationally utilizing two behavioral index-driven parameter adjustments. This allows feedback information from operational behaviors to directly guide the dynamic evolution of the interface, forming a behavior-driven "cognition-operation-interface" adaptive system. The updated spacing value R' is then mapped back to a complete set of spatial distribution parameters through a pre-defined mapping mechanism. Input layer: Calculated layout expansion factor K; Intermediate layer: Multiplied by the original spacing R to obtain R′; Mapping layer: R′ determines the minimum spacing between objects; the system rearranges object coordinates accordingly; then updates the "spatial distribution parameter set" such as object centroid distribution matrix, screen occupancy, and interaction hotspot distance; Output layer: After spatial reconstruction, a new average distribution density is formed (the system can further evaluate operability based on this). For example, in the initial state, a task page displays 10 interactive objects, the original buffer spacing R is 20 pixels, and the current unit area density is approximately 1.2 objects per 1000px². After behavior evaluation feedback, the system calculates K=1.5, obtaining R′=30 pixels, then the system expands the spacing of all objects to 1.5 times the original. After rearrangement, only an average of 0.9 objects are displayed per unit area, the average distribution density decreases, and the accidental touch rate decreases accordingly. Furthermore, during the layout density adjustment process for interactive optimization, the display area and boundary position of all interactive objects in the 3D simulation scene are obtained by spatial coordinate mapping rules. The adjusted layout parameters take effect immediately after the update and are simultaneously displayed in the user view through transparency gradient animation and smooth displacement animation, avoiding abrupt switching during interface changes that may affect user attention. The layout update process also caches and retains the task execution status to ensure that user interaction is not interrupted during interface adjustment, completing a continuous process of evaluation → optimization → continued operation.

[0050] 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.

[0051] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes 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 this application.

[0052] 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 implementation should not be considered beyond the scope of this application.

[0053] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0054] 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. A virtual reality-based simulation training method for reciprocating compressor system operation, characterized in that, Includes the following steps: Constructing a 3D simulation scenario: Based on the actual structure and control logic of the reciprocating compressor system, a virtual interactive space is created, and standard operating procedures and interactive objects consistent with the actual training objectives are configured; Execute interactive training: Users complete preset tasks in a virtual space. During the process, the user's gaze trajectory, action path, operation response sequence and input characteristics are collected in real time to build behavioral data records. Identify Deviation Behavior: When a user's behavior is detected to deviate from the standard operation path beyond the preset tolerance, the interaction behavior evaluation process is triggered. Calculate behavioral indices: Based on the collected data, construct two evaluation indices. The first is the cognitive bias index, which reflects the spatial mismatch between the user's focus and the target object. The second is the interaction consistency index, which represents the degree of consistency between the user's operation behavior and the standard operation path. Execution interaction optimization: The cognitive bias index and the interaction consistency index are jointly input into the pre-trained model to output the operation behavior evaluation result. When the result is lower than the set accuracy threshold, the layout density of the target object is dynamically adjusted.

2. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 1, characterized in that, The user operation behavior data collected during interactive training simultaneously records the task number, operation step sequence identifier, and timestamp information, and establishes a one-to-one mapping relationship with the name of the target interactive object in the 3D simulation scene, forming a complete operation data sequence for time-series behavior analysis. This data sequence is used as an input data source to calculate the cognitive bias index and interaction consistency index after the interactive behavior evaluation process is triggered, and is also used as a supplement to the training sample set for incremental model updates.

3. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 2, characterized in that, Identifying deviation behavior involves a comprehensive judgment based on the user's operation path, instruction call order, response latency, and target recognition behavior during interactive training. The instruction call sequence is compared with the standard task path in an ordered sequence. The longest common subsequence algorithm is used to calculate the instruction matching rate S. S is equal to the length of the longest common subsequence between the user's actual instruction call sequence and the standard instruction sequence divided by the length of the standard sequence. When S is less than a preset threshold, it is determined that the instruction order is abnormal. The response delay is evaluated using the difference Δσ = |σ-σ0| between the standard deviation of the response time for each operation step σ and the standard deviation of the standard operation time σ0. When Δσ exceeds the preset threshold two, it is determined to be an abnormal response delay. When any one of the following occurs simultaneously—an abnormal instruction sequence, an abnormal response delay, or a path similarity matching value below a preset threshold—it is determined that the current behavior meets the offset condition for triggering the interaction behavior evaluation process.

4. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 3, characterized in that, Path similarity comparison uses a method based on the shortest edit distance. After performing structured sequence parsing on continuous action segments, matching is performed to obtain path similarity matching values.

5. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 4, characterized in that, The calculation of the cognitive shift index includes the following steps: The first step is to extract the gaze point sequence formed by the user during the interactive training process. The gaze point sequence consists of the coordinates of the end point of the user's gaze trajectory in the three-dimensional simulation scene and the corresponding timestamp, which is used to reflect the changes in gaze position during task execution. The second step is to construct the target position function, which represents the geometric center coordinates of the target interactive object in three-dimensional space for each operation step. The third step is to generate a gaze distance sequence. Each element of the gaze distance sequence represents the minimum Euclidean distance between the user's gaze point and the spatial coordinates of the corresponding target interactive object within a certain operation step time period, which is used to measure the actual gaze offset of the step. The fourth step is to standardize the gaze distance sequence by using the maximum spatial distance between all interactive objects in the simulation scene as the normalization benchmark to obtain the gaze offset value sequence, where each gaze offset value is a standardized spatial offset score value. Fifth step: Taking each operation step as the center, analyze and construct the gaze stability function. The gaze stability function calculates the autocorrelation of the gaze distance change trend between consecutive steps through a sliding window method, and generates a stability score that reflects the degree of gaze fluctuation. Step 6: Calculate the cognitive shift index. The cognitive shift index is the arithmetic mean of the product of each fixation shift value and the complementary value of the fixation stability function score for that step in all operation steps. The complementary value is defined as a minus the fixation stability function score, which is used to represent the degree of gaze dispersion. Finally, the average of the products of all steps is taken as the output result of the cognitive shift index.

6. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 5, characterized in that, The calculation of the interaction consistency index includes the following steps: Data metrics for determining behavioral deviations are extracted, including the command matching rate generated during user operations, the difference in standard deviation of response latency, the path similarity matching value, and the behavioral sequence difference distance. The behavioral sequence difference distance is the dynamic registration cost between the user operation event sequence and the standard event sequence. The calculation of the behavioral sequence difference distance is based on the dynamic time warping algorithm. The difference in standard deviation of response delay and the distance of behavioral sequence difference are normalized to obtain the consistency score of operation rhythm and the consistency score of behavior structure. The consistency score of operation rhythm is one minus the difference in standard deviation of response delay after normalization, and the consistency score of behavior structure is one minus the distance of behavioral sequence difference after normalization. The interaction consistency index is the fourth root of the product of instruction matching rate, operation rhythm consistency score, path similarity matching value and behavior structure consistency score.

7. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 6, characterized in that, The pre-trained model is a multi-layer neural network structure built based on supervised learning methods. The training samples come from standard behavior datasets and non-standard behavior datasets with operational experience labels. The standard behavior datasets come from the behavior sequences recorded by actual operators when completing tasks in a 3D simulation scene. The non-standard behavior datasets include user data from beginners or users with operational errors. During training, the input is a joint feature vector composed of the cognitive bias index and the interaction consistency index, and the output is a numerical label of the operational behavior evaluation result, which represents the level of difference between the current behavior and the standard behavior.

8. The virtual reality-based reciprocating compressor system operation simulation training method according to claim 7, characterized in that, During the interaction optimization process, when the evaluation result of the operation behavior is lower than the set accuracy threshold, the spatial distribution parameters of the interactive objects in the 3D simulation scene are adjusted. The adjustment method is to modify the average distribution density of the interactive objects in the unit visible area of ​​the scene interface. The adjustment ratio of this layout density is calculated by the cognitive offset index and the interaction consistency index. The calculation method is as follows: Let the cognitive offset index be D1 and the interaction consistency index be D2. Then the layout expansion factor K is calculated as K=1+α×D1+β×(1-D2), where α and β are preset empirical adjustment coefficients with values ​​ranging from 0 to 1. K is used as the magnification factor of the original minimum boundary spacing R to reset the boundary buffer R'=K×R of each interactive object, and finally update the arrangement distance between interactive objects.

Images