CN118135878A

CN118135878A - Three-dimensional simulation training system for power grid automation equipment

Info

Publication number: CN118135878A
Application number: CN202410546565.3A
Authority: CN
Inventors: 宋新新; 荆辉; 王志强; 周博曦; 王仕韬; 李经纬; 徐英杰
Original assignee: Pingyuan Power Supply Co Of State Grid Shandong Electric Power Co; Jiangsu Wanju Technology Co ltd; State Grid of China Technology College
Current assignee: Pingyuan Power Supply Co Of State Grid Shandong Electric Power Co; Jiangsu Wanju Technology Co ltd; State Grid of China Technology College
Priority date: 2024-05-06
Filing date: 2024-05-06
Publication date: 2024-06-04
Anticipated expiration: 2044-05-06
Also published as: CN118135878B

Abstract

The invention discloses a three-dimensional simulation training system of power grid automation equipment, and aims to provide a high-efficiency, real and personalized learning platform; firstly, a three-dimensional simulation module creates and renders a three-dimensional model of power grid automation equipment by using a graphic processing unit and a physical engine, and provides visual reality and operational sense for a user; secondly, the intelligent teaching module automatically adjusts teaching contents and difficulty by analyzing interaction behaviors and learning progress of the user so as to match the actual operation level of the user; finally, the user interface is used as a platform for interaction between the user and the system, provides an interactive operation interface, allows the user to execute simulation operation, and adjusts teaching contents according to the guidance of the intelligent teaching module; the system provides an innovative training tool for professionals and learners in the field of power grid automation through integrating graphic rendering technology, physical simulation and artificial intelligence teaching, and aims to enhance understanding, improve operation skills and promote learning efficiency.

Description

Three-dimensional simulation training system for power grid automation equipment

Technical Field

The invention relates to the technical field of computers, in particular to a three-dimensional simulation training system for power grid automation equipment.

Background

In the field of grid automation, operator training is critical to ensure safe, reliable and efficient operation of the power system. With the progress of technology, three-dimensional simulation technology has been widely applied to various training systems, providing a more intuitive and interactive learning experience than traditional teaching methods. However, existing grid automation equipment training systems often rely on outdated two-dimensional graphics and limited interactive functionality, which limits the user's overall understanding and in-depth learning of the complex grid automation equipment operating environment.

In addition, conventional training methods rarely provide instant feedback and personalized teaching plans, resulting in low learning efficiency and difficulty in adapting to specific learning needs of different users. In these systems, the authenticity of physical interactions and device operations is often neglected, which further increases the difficulty of transitioning from a learning environment to an actual operating environment.

Therefore, it is necessary to develop a new three-dimensional simulation training system for power grid automation equipment.

Disclosure of Invention

The application provides a three-dimensional simulation training system for power grid automation equipment, which is used for improving training effect.

The application provides a three-dimensional simulation training system of power grid automation equipment, which comprises the following components:

the three-dimensional simulation module comprises a graphic processing unit and a physical engine, wherein the graphic processing unit is used for creating and rendering a three-dimensional model of the power grid automation equipment; the physical engine is used for simulating real physical interaction and feedback of the operation of the power grid automation equipment;

The intelligent teaching module is used for determining a teaching plan and a difficulty level matched with the actual operation level of the user according to the interactive behavior and the learning progress of the user and the three-dimensional simulation module through the user interface; according to the determined teaching plan and difficulty level, a teaching adjustment instruction is sent to a user;

a user interface for presenting to a user the grid automation device and environment rendered by the three-dimensional simulation module through the graphics processing unit; providing an interactive interface to enable a user to perform a simulation operation on a three-dimensional model, wherein the simulation operation performs feedback simulation through a physical engine; and receiving a teaching adjustment instruction from the intelligent teaching module, and adjusting the displayed teaching content and difficulty according to the received teaching adjustment instruction.

Furthermore, the graphic processing unit is used for executing a ray tracing algorithm to simulate ray behaviors under complex illumination conditions in the power grid automation environment and interaction of the ray and the surface material of the power grid automation equipment, wherein the ray behaviors comprise scattering, reflection and refraction of the ray; the ray tracing algorithm comprises the following steps:

Setting a scene, including setting the type of a light source, the position of the light source, the light intensity of the light source and the geometric shape and material properties of the power grid automation equipment in the scene; wherein the light source type comprises a point light source and a directional light source, and the material property comprises reflectivity and refractive index;

Emitting a virtual ray from a light source and tracking a propagation path of the virtual ray in a scene, wherein the propagation path comprises scattering, reflection and refraction of the ray when the ray meets the surface of the equipment;

For the interaction of the virtual light in the propagation path and the power grid automation equipment, calculating the illumination intensity and the color of an interaction point, wherein the calculation formulas are respectively shown in the following formulas 1 and 2:

；

Wherein, Representing the total illumination intensity of the interaction point; /(I)Representing the intensity of ambient light; /(I)Representing the brightness of the light source; /(I)Is the normal vector of the interaction point surface; /(I)Is a unit vector from the interaction point to the light source; /(I)Is a reflection vector; /(I)Is the vector from the surface point to the observer; /(I)Is the diffuse reflection coefficient of the surface; /(I)Is the specular reflection coefficient; /(I)Is the glossiness of the material; /(I)Representing the total color of the interaction points; /(I)Representing the effect of ambient light on the color of the object, as a constant value representing the minimum amount of color reflected by the object in the absence of direct illumination; /(I)Is the color of the light source, and represents the color characteristic of the light source; /(I)Is the color of the object itself.

Still further, the graphics processing unit is configured to perform a texture mapping algorithm to optimize surface details of the grid automation device; the texture mapping algorithm uses the following equation 3 for texture adjustment:

；

Wherein, Representing the adjusted texture color; /(I)Is the original texture color; /(I)Is the surface brightness by total illumination intensity/>, of the interaction points on the surfaceSumming to obtain; /(I)Is the light reflectivity of the material.

Further, the intelligent module adopts a trained hybrid neural network model to dynamically determine teaching plans and difficulty levels matched with actual operation levels of users; the hybrid neural network model comprises a decision tree network, a behavior pattern recognition network and a comprehensive evaluation network;

The decision tree network is realized by adopting a decision tree model and is used for processing direct operation data of a user to obtain a preliminary classification result of learning progress and skill level of the user, wherein the direct operation data comprises an operation success rate, average operation time, error times and help request frequency;

the behavior pattern recognition network is realized by adopting a convolutional neural network and is used for processing operation behavior pattern data of a user to obtain a learning disorder analysis result of the user, wherein the operation behavior pattern data comprises gesture recognition and time distribution of an operation sequence;

The comprehensive evaluation network is realized by adopting a fully-connected neural network and is used for processing the preliminary classification result and the learning disability analysis result so as to obtain a teaching plan and a difficulty level matched with the actual operation level of the user.

Still further, the decision tree network includes a dynamic feature selection mechanism that automatically identifies and selects features most relevant to user learning progress and skill level based on machine learning algorithms; wherein the most relevant features include speed variation of user operation, accuracy of operation, and improved speed of user on operation task.

Further, the behavior pattern recognition network adopts a mixed structure of a convolutional neural network and a cyclic neural network.

Further, the comprehensive evaluation network comprises an adaptive learning module based on user feedback, and the adaptive learning module dynamically adjusts the weight and parameters of the comprehensive evaluation network by using direct feedback from the user on the satisfaction of the teaching contents.

Still further, the user interface includes a dashboard for dynamically showing the user's learning progress, the completed tutorial units and upcoming tutorials.

Still further, the user interface includes an interactive question-and-answer function that enables a user to submit questions associated with the three-dimensional simulation module directly through the user interface and to receive customized answers generated by the intelligent teaching module.

Still further, the user interface includes a mode switching function that allows a user to switch between different views and modes of operation according to personal preferences.

The application has the following beneficial technical effects:

(1) Providing a highly realistic operating environment: by combining an advanced graphic processing unit and a physical engine, the system can create and render a very real three-dimensional model of the power grid automation equipment, and provide almost real visual experience and operation feedback for users. The highly real simulation environment can remarkably improve the learning efficiency and operation accuracy of users, and particularly in the aspects of complex equipment operation and fault handling training.

(2) The intelligent teaching module can dynamically adjust teaching plans and difficulty levels according to interactive behaviors and learning progress of users. The personalized learning path not only can adapt to the requirements of users with different levels, but also can provide customized training aiming at the weaknesses of the users, thereby promoting the improvement of the skills of the users more effectively.

(3) The user interface provides an interactive operation platform, so that a user can directly perform simulation operation on the three-dimensional model and receive feedback simulation from the physical engine in real time, and the interactivity greatly enhances the immersion sense and the experience quality of learning. Through the instant feedback of the simulation operation, the user can immediately know whether the operation is correct or not, and the understanding and the memorization are deepened.

(4) Timely updating and personalizing the adjusted teaching content: the system can receive and execute teaching adjustment instructions from the intelligent teaching module, and timely update and individuate adjustment of teaching contents according to the learning progress and operation performance of the user. The timeliness and adaptability of the teaching plan are guaranteed, and the user is helped to achieve the best learning effect in the shortest time.

Drawings

Fig. 1 is a schematic diagram of a three-dimensional simulation training system for power grid automation equipment according to a first embodiment of the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.

The application provides a three-dimensional simulation training system for power grid automation equipment. Referring to fig. 1, a schematic diagram of a first embodiment of the present application is shown. A first embodiment of the present application is described in detail below with reference to fig. 1, which provides a three-dimensional simulation training system for an electrical grid automation device.

The three-dimensional simulation training system of the power grid automation equipment comprises a three-dimensional simulation module 101, an intelligent teaching module 102 and a user interface 103.

A three-dimensional simulation module 101 comprising a graphics processing unit and a physics engine, the graphics processing unit being configured to create and render a three-dimensional model of the grid automation device, providing visual realism to a user; the physical engine is used for simulating actual physical interaction and feedback of the operation of the power grid automation equipment and providing operational sense of reality for users.

In the three-dimensional simulation training system for the power grid automation equipment provided by the embodiment, the three-dimensional simulation module 101 is a core of the system, and provides a highly real and interactive learning environment for users by integrating a graphic processing unit and a physical engine.

The graphics processing unit is responsible for generating and rendering a three-dimensional model of the grid automation device in this module. It uses advanced graphics rendering techniques such as ray tracing and texture mapping to create impressive three-dimensional images of visual effects. The techniques can ensure that the model is accurate in geometric shape and extremely real in light and shadow effect and material property, thereby providing a user with an experience visually appearing as if the user were in a real device operating environment. To achieve this, the graphics processing unit utilizes high performance graphics processing hardware and optimized software algorithms to support real-time rendering of complex scenes.

Working in parallel with the graphics processing unit is a physics engine responsible for simulating the physical interactions and operational feedback of the grid automation device in a simulation environment. This includes, but is not limited to, start-up, shut-down of the device, dynamic response in the event of load changes, and performance in the event of device failure. The physical engine provides a sense of realism in operation by accurately calculating the effects of device operation on system state, and then feeding these effects back to the user in visual and non-visual (e.g., audible or tactile feedback) form. To achieve this effect, the physics engine employs state-of-the-art physics modeling techniques such as rigid body dynamics, fluid dynamics, and electromagnetic field modeling, ensuring that each detail of the simulation is as close as possible to the physical laws of the real world.

To implement this module, it should be appreciated that the three-dimensional simulation module 101 is not just a generator of mere images and physical effects, but is also a highly integrated and interactive system. This means that the graphics processing unit and the physics engine need to work in close concert to ensure that the results of the image rendering reflect the results of the physics simulation in real time and vice versa. In addition, the module needs to be tightly integrated with the intelligent teaching module 102 and the user interface 103 to ensure timely updating and personalized adjustment of learning content, and accurate execution and feedback of simulation operations.

Furthermore, the graphic processing unit is used for executing a ray tracing algorithm to simulate the ray behavior under the complex illumination condition in the power grid automation environment and the interaction of the ray and the surface material of the power grid automation equipment so as to realize the highly real simulation of the natural light and the artificial light source effect, wherein the ray behavior comprises the scattering, reflection and refraction of the ray; the ray tracing algorithm comprises the following steps:

；

Wherein, Representing the total illumination intensity of the interaction point; /(I)Representing the brightness of the light source; /(I)Is the surface normal; is a unit vector from the interaction point to the light source; /(I) Is a reflection vector; /(I)Is the vector from the surface point to the observer; /(I)Is the diffuse reflection coefficient of the surface; /(I)Is the specular reflection coefficient; /(I)Is the glossiness of the material; /(I)Representing the total color of the interaction points; Representing the effect of ambient light on the color of the object, as a constant value representing the minimum amount of color that the object will reflect even in the absence of direct illumination; /(I) Is the color of the light source, and represents the color characteristic of the light source; /(I)Is the color of the object itself, i.e. the inherent color of the object when not illuminated by any illumination.

In the three-dimensional simulation training system of the power grid automation equipment, the graphic processing unit plays a vital role, and particularly, complex illumination conditions and interaction of light and equipment surface materials are simulated by executing a light ray tracing algorithm, so that the highly real simulation of natural light and artificial light source effects is achieved.

First, scene setting is a preliminary step of ray tracing algorithms, which involves the definition of the type of light source (e.g., point light source, directional light source), location, light intensity, and geometry and material properties of the grid automation device. This step is critical to the accuracy and authenticity of the overall algorithm. The different light source types determine the propagation mode of the light rays in the scene, and the material properties (including reflectivity and refractive index) affect the interaction effect of the light rays with the surface of the device.

Subsequently, a virtual ray is emitted from the light source and its propagation path in the scene is tracked. In this process, the light may encounter the surface of the power grid automation device, and scattering, reflection, refraction, and the like may occur. These interactions are simulated by physical laws and optical principles, providing complex lighting effects and shadows to objects in the scene.

When the illumination intensity and the color of the interaction point are calculated, two core calculation formulas are introduced.

；

Wherein,The total illumination intensity representing the interaction point is the target value that eventually needs to be calculated.

Representing ambient illumination intensity, can be obtained by observing and measuring natural light or indoor lighting conditions for simulating the basic brightness without direct light source illumination.

Representing the brightness of the light source, which is obtained by physical measurement or device specification, depending on the type of light source and the power of the light source.

Is the normal vector of the surface of the interaction point and is calculated by the geometric data of the three-dimensional model of the power grid automation equipment.

Is a unit vector from the interaction point to the light source, and is calculated according to the position of the light source and the position of the interaction point.

Is a reflection vector calculated from the incident angle of the light and the normal vector. In ray tracing algorithms, when a ray of light is emitted from a light source and encounters an object surface, reflection occurs according to the characteristics of the object surface and the angle of incidence of the ray of light. The term "reflection vector (/ >)) "Means the vector of the direction of the reflected light with respect to the surface of the object.

Is a vector from the interaction point to the observer (e.g., camera or eye) and is calculated from the observation position and the interaction point position.

Is the diffuse reflection coefficient of the surface; /(I)Is the specular reflection coefficient; these two parameters are determined by the physical properties of the material and can be obtained from the material science literature or experimental data.

The glossiness of the material reflects the influence of the smoothness of the surface of the material on specular reflection light, and is also determined by the material characteristics.

Representing the total color of the interaction points;

is the basic influence of the ambient light on the color of the object, and determines the basic color expression of the object without direct illumination.

The color of the light source represents the color characteristic of the light source itself and is determined by the physical characteristic of the light source.

The color of the object is obtained through the observation or design parameters of the real object of the power grid automation equipment.

The ray tracing algorithm can provide a highly real visual effect for the three-dimensional simulation training system of the power grid automation equipment, and the training effect and the user experience are enhanced.

；

In the three-dimensional simulation training system of the power grid automation equipment, the texture mapping algorithm executed by the graphic processing unit plays a crucial role, in particular to optimize the surface detail of the power grid automation equipment. The algorithm adjusts the original texture color by using a formula so as to achieve a more real visual effect. The meaning and acquisition method of each term in this process and formula are described in detail below.

The core of the texture mapping algorithm is the pass formulaTo adjust the texture color to reflect the effect of light interaction with the object surface on the object texture. In this process, the original texture color/>Will be based on the surface brightness/>And light reflectivity of material/>And (5) adjusting.

Representing the adjusted texture color is the final target of the algorithm, namely the color which the surface of the power grid automation equipment should present under the specific illumination condition. The color considers the illumination effect, so that the texture is more vivid.

Is the original texture color, i.e., the original color of the model texture in the absence of illumination. These colors are typically obtained from photographs or design data of the actual device, converted to texture maps by a digitizing process, and applied to a three-dimensional model.

Represents the surface brightness by the total illumination intensity/>, of each interaction point on the model surfaceAnd performing accumulation and summation. /(I)Reflecting the irradiation effect of the light source on a certain point of the model surface, including direct illumination and ambient illumination. The calculation of this value involves the aforementioned ray tracing algorithm, determined by modeling the propagation of rays in the scene and interactions with the model surface.

Is the light reflectance of a material and indicates the ability of the material to reflect light. The light reflectivity of different materials is different, and the parameter is usually set according to the actual physical characteristics of the materials, and can be obtained through experimental measurement or reference to related material science documents. The light reflectivity directly affects the adjustment degree of the texture color, and a higher light reflectivity means that the illumination effect has a greater influence on the texture color when calculating the adjusted texture color.

By comprehensively considering the original texture color, the surface brightness and the light reflectivity of the material, the texture mapping algorithm can accurately adjust the texture color of the power grid automation equipment, so that the power grid automation equipment shows more real and fine visual effects under different illumination conditions. The implementation of the algorithm not only enhances the visual sense reality of the model, but also greatly improves the effect and user experience of three-dimensional simulation training.

The intelligent teaching module 102 is used for determining a teaching plan and a difficulty level matched with the actual operation level of the user according to the interactive behavior and learning progress of the user and the three-dimensional simulation module through the user interface; and sending a teaching adjustment instruction to a user according to the determined teaching plan and difficulty level.

The intelligent teaching module 102 is a key component of the three-dimensional simulation training system of the power grid automation equipment provided by the embodiment, and aims to realize individuation and dynamic adjustment of teaching contents through advanced technical means so as to maximally meet learning requirements of different users. The module utilizes artificial intelligence algorithms, including but not limited to machine learning and data analysis techniques, to automatically identify the learning behavior, progress, and level of understanding of the user, and to formulate or adjust teaching plans and difficulty levels accordingly.

In a specific implementation aspect, the intelligent teaching module first needs to collect interaction data between a user and the three-dimensional simulation module through a user interface. These data include the user's operating conditions, operating frequency, operating results, feedback time, etc. for the simulated grid automation device, thereby providing the intelligent teaching module with sufficient information to assess the user's learning efficiency and operating skill level.

Next, the intelligent teaching module analyzes the data through a built-in algorithm processing unit. The algorithm processing unit may include pattern recognition algorithms, cluster analysis, neural networks, etc. to recognize patterns and trends in the user learning process. Based on these analysis results, the module can determine the current operational capabilities and understanding depth of the user, thereby identifying the areas where the user needs reinforcement learning.

And then, the intelligent teaching module dynamically formulates or adjusts a teaching plan and a difficulty level according to the analysis result. For example, if a user performs poorly in a particular link, the module may reduce the difficulty level of that link or provide more exercises for that link. Conversely, if the user is able to easily accomplish certain tasks, the module may increase difficulty or introduce new learning content to maintain the challenges and effectiveness of learning.

And finally, the intelligent teaching module sends a teaching adjustment instruction to a user through a communication interface with the user interface to guide the user interface to correspondingly adjust the displayed teaching content and difficulty. The process is performed in real time, so that the teaching content can be updated in time, and individuation and adaptability of the teaching content are guaranteed.

To implement this module, it should be appreciated that the intelligent teaching module is not merely a stand-alone algorithm or software program, but is a highly integrated system that needs to work in close cooperation with the three-dimensional simulation module and the user interface. Through the implementation of the intelligent teaching module, the system can dynamically adjust the teaching strategy according to the actual learning condition of the user, thereby greatly improving the teaching effect and learning efficiency.

Furthermore, the intelligent module adopts a hybrid neural network model to dynamically determine teaching plans and difficulty levels matched with the actual operation level of the user; the hybrid neural network model comprises a decision tree network, a behavior pattern recognition network and a comprehensive evaluation network;

the comprehensive evaluation network is realized by adopting a fully-connected neural network and is used for processing the preliminary classification result and the learning disability analysis result to obtain a teaching plan and a difficulty level matched with the actual operation level of the user.

In the three-dimensional simulation training system of the power grid automation equipment, the core of the intelligent module is a hybrid neural network model, and the model is combined with a decision tree network, a behavior pattern recognition network and a comprehensive evaluation network, so that the teaching plan and the difficulty level matched with the actual operation level of each user are dynamically determined by analyzing the interactive behavior and the learning progress of the user.

The decision tree network obtains preliminary classification results of the learning progress and skill level of the user by processing direct operation data of the user, such as operation success rate, average operation time, number of errors, and help request frequency. This part can be implemented using conventional decision tree models such as CART or C4.5, where each node represents a decision test, each branch represents the result of the test, and the leaf node represents the final classification result. The skilled person can choose appropriate splitting criteria, such as information gain or genie purity, to construct the decision tree based on specific operational data characteristics.

Further, the decision tree network comprises a dynamic feature selection mechanism, and the dynamic feature selection mechanism automatically identifies and selects the features most relevant to the learning progress and skill level of the user based on a machine learning algorithm, so that the accuracy and efficiency of classification of the user by the decision tree network are improved; wherein the most relevant features include speed variation of user operation, accuracy of operation, and improved speed of user on a particular operational task.

In a three-dimensional simulation training system of a power grid automation device, a Decision Tree Network (DTN) plays a critical role, which primarily classifies learning progress and skill level of a user by analyzing direct operation data of the user. To further enhance the classification capabilities of DTNs, a dynamic feature selection mechanism is introduced that automatically identifies and selects features that best reflect the user's learning state through a machine learning algorithm. This approach aims to improve the accuracy and efficiency of decision tree networks in processing user data, particularly in identifying user learning progress and skill level.

The core of the dynamic feature selection mechanism is to automatically determine which features are most relevant to the user's learning progress and skill level using machine learning algorithms, such as feature importance scoring, recursive Feature Elimination (RFE), or model-based feature selection methods. This process involves the following steps:

1. Characteristic pretreatment: firstly, normalization and standardization processing are carried out on operation data of a user, so that the data are ensured to be on the same scale, and the model is prevented from biasing to the characteristics with larger magnitude. Such operation data may include the user's operation success rate, average operation time, number of errors, help request frequency, etc.

2. Feature importance assessment: the individual features are scored for importance using a machine learning algorithm. For example, the impact of each feature on the user classification can be evaluated by training an auxiliary decision tree or random forest model and using the feature importance index provided by these models.

3. Feature selection: based on the feature importance from the assessment, the features most critical to the user's learning progress and skill level prediction are dynamically selected. By "dynamic" is meant that the process of feature selection is not done at once, but can be updated as more user data is accumulated and the model retrains.

4. Training a decision tree model: the decision tree model is trained using the selected features. This step ensures that the model is focused on those features that best reflect the user's learning state, thereby improving the accuracy and efficiency of classification.

The dynamic feature selection mechanism is innovative in that it automatically recognizes features that are most relevant to the user's learning progress and skill level, such as user operation speed variation, operation accuracy, and user improvement speed over a specific operation task, rather than by preset or manual selection. The self-adaptive feature selection method not only enables the decision tree model to be more flexible and accurate, but also provides a solid foundation for providing personalized learning paths and difficulty setting.

By adopting the dynamic characteristic selection mechanism, the intelligent module of the three-dimensional simulation training system of the power grid automation equipment can more accurately understand and adapt to the learning needs of each user, and provide the most suitable teaching content and difficulty level for the users, so that the learning effect is optimized and the user satisfaction is improved.

The behavior pattern recognition network adopts a Convolutional Neural Network (CNN) to analyze the operation behavior pattern data of the user, including gesture recognition, time distribution of an operation sequence and the like. This network is capable of automatically extracting useful features from the raw operational behavior data for identifying learning disabilities that a user may encounter. A technician can utilize existing deep learning frameworks, such as TensorFlow or PyTorch, to design appropriate CNN architectures, including convolutional, pooling, and fully-connected layers, to handle time-series data and gesture recognition tasks.

Still further, the behavior pattern recognition network employs a hybrid architecture of convolutional neural networks and recurrent neural networks to efficiently process time-series data and capture time-dependence in user operational behavior.

In a three-dimensional simulation training system for power grid automation equipment, a behavior pattern recognition network is a key component and is specifically designed to process and analyze operation behavior pattern data of a user, wherein the data are essentially time-series data, and include dynamic processes of operation of the user, such as continuous actions of gestures and time distribution of operation. To understand the behavior patterns of the user in depth and identify possible learning disorders therefrom, a hybrid convolutional and cyclic neural network (CNN-RNN) architecture may be employed.

The core idea of the hybrid network architecture is the ability to automatically extract features in local areas and to process time series data by Recurrent Neural Networks (RNNs), particularly long and short term memory networks (LSTM) or variants of gated loop units (GRUs), which are part of RNNs, in combination with the ability of Convolutional Neural Networks (CNNs), superior to capturing long term dependencies.

1. Data preprocessing: first, the operation behavior pattern data of the user needs to be appropriately preprocessed. This includes normalization of the data, and possibly reshaping, to make it suitable as an input to the neural network. For example, the gesture recognition data is converted into a sequence of equal length, or the operation time distribution is converted into feature vectors within a fixed time window.

2. Convolutional neural network layer: the data first passes through a series of convolution layers that automatically extract useful local features from the operational behavior data. The convolution layer is typically followed by a pooling layer to reduce the spatial dimensions of features and increase the abstract capability of the model. The purpose of this step is to identify key patterns in the operational behavior, such as the shape of a particular gesture or the cadence of motion, without explicitly encoding the patterns.

3. Cyclic neural network layer: features extracted by the convolutional layer are then fed into a recurrent neural network layer, such as an LSTM or GRU. The purpose of this layer is to process and analyze time dependencies in the time series data, such as the order of operations or time intervals between gestures. By memorizing past information, the RNN layer can understand dynamic changes in operation behavior, and recognize patterns that may indicate learning disabilities.

4. Output and interpretation: the output of the network is the identification and analysis of learning disabilities that the user may encounter, which can be used directly to generate personalized teaching advice. For example, if the network recognizes that the user is often making mistakes in performing a certain continuous gesture, this may indicate a need to add exercises to this gesture in the teaching plan.

By combining the strong items of CNN and RNN, the hybrid network can extract rich information from the operation behaviors of users, including both static features of actions such as gestures and dynamic features of operation sequences.

The innovation of the hybrid structure of the convolutional neural network and the cyclic neural network is that the hybrid structure can effectively process and analyze complex time series data and capture time dependence in user operation behaviors, so that powerful support is provided for personalized teaching in a three-dimensional simulation training system of power grid automation equipment.

The following is an implementation code based on a Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) hybrid architecture implemented by Python using TensorFlow framework.

import tensorflow as tf

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Conv1D, MaxPooling1D, LSTM, Dense, Flatten

# Assuming that the preprocessing of the data has been completed, the data has been in a format suitable for input into the neural network

The # data preprocessing section converts time-series data into equal-length series and performs normalization processing

# Creation model

model = Sequential()

First part #: convolutional neural network layer

Adding convolution layer for extracting local features in time series data

model.add(Conv1D(filters=64, kernel_size=3, activation='relu', input_shape=(None, input_features)))

Adding pooling layers for reducing the spatial dimension of features

model.add(MaxPooling1D(pool_size=2))

model.add(Conv1D(filters=128, kernel_size=3, activation='relu'))

model.add(MaxPooling1D(pool_size=2))

# Flattening layer, unifying multidimensional input, ready to join RNN layer

model.add(Flatten())

Second part #: circulating neural network layer

Add LSTM layer for processing and analyzing time dependencies

model.add(LSTM(units=50, return_sequences=True))

model.add(LSTM(units=50))

# Output layer: for outputting analysis results of learning disabilities

Model. Add (units=10, activation= 'softmax') # assuming 10 different learning difficulties or disorder types

# Compiling model

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

Training model #

# Assume that train_data is training data and train_labes is a tag

# model.fit(train_data, train_labels, epochs=10, batch_size=64)

# Output model structure

model.summary()

Model. Summary () here will print out each layer details of the model, including layer name, output shape, and number of parameters.

In this code, a section is first constructed that contains both convolutional and pooling layers, which is dedicated to automatically extracting local features from operational behavior data. Next, an LSTM layer is added for analysis of the time-dependent relationship of these features, capturing possible long-term dependencies. The prediction results are finally output through a fully connected layer, and the results can be used for guiding teaching or identifying potential learning disabilities.

This architecture is suitable for processing and analyzing complex patterns containing time series data, especially for scenes involving continuous dynamic operations, such as gesture recognition and complex operation sequence analysis. The model not only can improve understanding of the user behavior mode, but also can provide data support for personalized teaching based on the user behavior.

The comprehensive evaluation network uses a Fully Connected Neural Network (FCNN) to comprehensively process the output from the decision tree network and the behavior pattern recognition network, considers other relevant factors such as difficulty coefficients of operation tasks, user learning efficiency and satisfaction evaluation, and finally generates a teaching plan and difficulty level for the user. This network may include multiple fully connected layers, each of which introduces nonlinearities through an activation function (e.g., reLU) to ensure that complex data relationships can be captured. The depth and width of the network are required to be adjusted according to the specific requirements of the task, and network parameters are optimized through methods such as cross validation.

Still further, the comprehensive assessment network includes an adaptive learning module based on user feedback. The self-adaptive learning module dynamically adjusts the weight and parameters of the comprehensive evaluation network by using direct feedback from the user on the satisfaction degree of the teaching contents so as to ensure that the recommendation of the teaching plan and the difficulty level accords with the personal preference and the learning effect of the user as much as possible, and realizes the truly personalized learning experience.

In the three-dimensional simulation training system of the power grid automation equipment, the comprehensive evaluation network comprises an adaptive learning module based on user feedback. The core function of this module is to dynamically adjust the weights and parameters of the network using direct feedback of the user's satisfaction with the teaching content. By the aid of the design, teaching plans and difficulty levels can be matched with personal preferences of users better, learning effects are improved, and personalized learning experience is achieved. The following is a detailed description of how this module is implemented.

1. Collecting user feedback: first, the system needs to collect the feedback of users after they complete the teaching module. This may be achieved by simple satisfaction surveys, direct scoring, or more detailed feedback forms. To automate this process, a feedback collection mechanism, such as a pop-up window or feedback form after the end of the teaching module, may be embedded in the three-dimensional simulation training system.

2. Processing feedback data: the collected user feedback data needs to be processed and standardized in order to be used by the adaptive learning module of the comprehensive evaluation network. The data processing steps may include converting qualitative feedback into quantifiable scoring, screening and summarizing the feedback of multiple users to form a comprehensive view, and identifying key feedback trends and patterns.

3. Dynamically adjusting network weights and parameters: the core of the adaptive learning module based on user feedback is to dynamically adjust the weight and parameters of the comprehensive evaluation network by using the collected and processed feedback data. This step may be accomplished by a variety of methods, such as gradient descent or variants thereof, for parameter updating, or using more advanced algorithms such as reinforcement learning to optimize the decision making process of the network. The key is to guide these adjustments based on feedback of user satisfaction so that the teaching plan more closely fits the needs and preferences of the user.

4. Testing and iterating: in the adaptive adjustment process, it is important to continuously test the influence of new network configuration on teaching plans and difficulty level recommendations. This includes improvements in contrast to user satisfaction changes, learning effects improvements, and personalized experiences before and after adjustment. Through continuous testing and iteration, the self-adaptive learning module can reflect the preference and learning requirement of the user more and more accurately.

5. Feedback loop: and establishing a positive feedback loop, and continuously improving the relevance and individuation degree of the teaching content by continuously collecting user feedback and optimizing network parameters according to the feedback. This loop ensures that the teaching system is able to adapt to changes in user behavior, continually providing an optimal learning experience.

Through the steps, the self-adaptive learning module based on the user feedback can enable the comprehensive evaluation network to respond to the learning progress and feedback of the user in a real sense, dynamically adjust the teaching plan and the difficulty level, and realize personalized learning experience.

From the viewpoint of data flow, the direct operation data of the user is firstly input into a decision tree network to generate a preliminary classification result; meanwhile, the operation behavior pattern data of the user is input to the behavior pattern recognition network, and learning disability analysis results are generated. The outputs of the two parts are then integrated and input into a comprehensive evaluation network, which can also consider additional input factors such as task difficulty coefficients and user feedback, and finally output personalized teaching plans and difficulty level suggestions for the user through the processing of the fully connected network.

Provided below is a reference implementation describing how teaching plans and difficulty levels matching the user's actual operating level are dynamically determined by means of a hybrid neural network model.

import numpy as np

from sklearn.tree import DecisionTreeClassifier

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Conv1D, MaxPooling1D, Flatten, Dense

from tensorflow.keras import Input, Model

# Suppose that direct operation data and operation behavior pattern data have been preprocessed and are ready

Direct manipulation data #: operation success rate, average operation time, number of errors, help request frequency

direct_operation_data = np.load('direct_operation_data.npy')

Operational behavior pattern data: features including gesture recognition and time distribution of sequence of operations

behavior_pattern_data = np.load('behavior_pattern_data.npy')

# Decision Tree Network (DTN)

def decision_tree_network(data):

# Use decision tree classifier

clf = DecisionTreeClassifier()

clf.fit(data[:, :-1], data[:, -1])

prelim_classification = clf.predict(data[:, :-1])

return prelim_classification

# Behavior Pattern Recognition Network (BPRN)

def behavior_pattern_recognition_network(data):

model = Sequential()

model.add(Conv1D(64, 3, activation='relu', input_shape=(data.shape[1], 1)))

model.add(MaxPooling1D(2))

model.add(Flatten())

model.add(Dense(50, activation='relu'))

model.add(Dense(1, activation='sigmoid'))

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

# Suppose that data is already in the correct shape

model.fit(data, np.random.randint(0, 2, size=(data.shape[0], 1)), epochs=10)

learning_obstacle_analysis = model.predict(data)

return learning_obstacle_analysis

# Comprehensive Evaluation Network (CEN)

def integrated_evaluation_network(dtn_output, bprn_output, additional_inputs):

inputs = np.concatenate((dtn_output, bprn_output, additional_inputs), axis=1)

input_layer = Input(shape=(inputs.shape[1],))

dense1 = Dense(128, activation='relu')(input_layer)

dense2 = Dense(64, activation='relu')(dense1)

Output layer=react (10, activation= 'softmax') (Dense 2) # assume 10 different teaching plans and difficulty levels

model = Model(inputs=input_layer, outputs=output_layer)

# Assume that the additional_inputs are already of the correct shape

model.fit(inputs, np.random.randint(0, 10, size=(inputs.shape[0], 10)), epochs=10)

teaching_plan_and_difficulty_level = model.predict(inputs)

return teaching_plan_and_difficulty_level

Additional inputs for the# hypothesis include task difficulty coefficients, user learning efficiency and satisfaction assessment, etc

additional_inputs = np.load('additional_inputs.npy')

# Data flow

dtn_output = decision_tree_network(direct_operation_data)

bprn_output = behavior_pattern_recognition_network(behavior_pattern_data)

final_output = integrated_evaluation_network(dtn_output, bprn_output, additional_inputs)

# Final_output is personalized teaching plan and difficulty level suggestion for user

Note that the reference codes may be adjusted according to the actual data format and network parameters. In actual implementation, the detailed design of each network part (such as the depth of a decision tree, the number of layers and filters of a convolution network, the number of layers of a full connection network, etc.) is optimized according to the specific task and the characteristics of a data set. In addition, data preprocessing, model training details (such as batch size, learning rate, etc.), and selection of superparameters are all key to achieving an efficient network.

To train the hybrid neural network model, data is first prepared and preprocessed, then a Decision Tree Network (DTN) and a Behavior Pattern Recognition Network (BPRN) are trained step by step, and finally the outputs of these networks are integrated to train a Comprehensive Evaluation Network (CEN). This process involves several key steps, ensuring that each network part can learn efficiently and contribute to the generation of the final personalized teaching plan and difficulty level.

First, direct operation data and operation behavior pattern data of a user are collected. The direct operation data may include an operation success rate, an average operation time, the number of errors, a help request frequency, etc., and the operation behavior pattern data may include gesture recognition, time distribution of an operation sequence, etc. These data need to be cleaned and standardized to ensure that the model can learn from a consistent data format.

For direct manipulation of data, min-max normalization or Z-score normalization may be employed to eliminate the effects of different dimensions so that the data is on the same scale. The operational behavior pattern data may require more complex pre-processing, including normalization of the time series and vectorization of the gesture data.

And training a decision tree model by utilizing the preprocessed direct operation data. A decision tree is constructed by selecting appropriate splitting criteria, such as information gain or genie uncertainty. The purpose of this step is to classify the user's skill level initially based on his operational data. The optimal tree depth and parameters are selected by a cross-validation method to avoid overfitting and ensure generalization capability of the model.

For operational behavior pattern data, convolutional Neural Networks (CNNs) are employed for training. Suitable network architectures, including convolutional, pooling, and fully-connected layers, are designed for automatically extracting useful features from the data. In the training process, super parameters of the network, such as learning rate, batch size and the like, need to be adjusted, and a proper regularization method is adopted to improve the performance of the model and avoid over fitting.

The outputs of the decision tree network and the behavior pattern recognition network are combined with additional input factors (such as task difficulty coefficients, user learning efficiency and satisfaction assessment) as inputs to the comprehensive assessment network. The comprehensive evaluation network is realized by adopting a fully-connected neural network, processes input data through a plurality of fully-connected layers, and introduces nonlinearity through an activation function. When the comprehensive evaluation network is trained, the network structure and parameters need to be carefully adjusted, and proper loss functions and optimization algorithms are used to ensure that the network can effectively learn from the comprehensive data and generate accurate teaching plans and difficulty level suggestions.

The entire training process is iterative and may require multiple adjustments and optimization of network parameters and architecture. For each network part, performance indicators during training, such as accuracy and loss, and performance on an independent validation set should be monitored to evaluate the learning effect and generalization ability of the model. Finally, by integrating the capabilities of different networks and analyzing the multi-source data, the hybrid neural network model can provide each user with a personalized teaching plan that exactly matches his skill level.

A user interface 103 for presenting to a user the grid automation device and environment rendered by the three-dimensional simulation module through the graphics processing unit; providing an interactive interface to enable a user to perform a simulation operation on the three-dimensional model, wherein the simulation operation performs feedback simulation through a physical engine to enhance learning experience; and receiving a teaching adjustment instruction from the intelligent teaching module, and adjusting the displayed teaching content and difficulty according to the received teaching adjustment instruction, so as to ensure timely updating and personalized adjustment of the teaching content.

The user interface 103 forms the front end of the three-dimensional simulation training system of the power grid automation equipment provided by the embodiment, and provides an intuitive and interactive platform for users to perform simulation operation and learning. The user interface aims to present complex three-dimensional simulation and intelligent teaching functions in a user-friendly manner, and ensures that a user can learn and practice from a training system effectively regardless of the technical background.

In designing and implementing a user interface, first of all, a presentation function is considered. The user interface needs to be able to clearly show the three-dimensional model of the grid automation device and environment generated by the three-dimensional simulation module 101. This includes detail rendering of the device, lighting effects of the environment, and dynamic changes produced by simulated operations, such as changes in device state or real-time presentation of physical feedback. To achieve this, the user interface employs a high resolution graphical presentation technique and optimizes the transmission of the data stream from the three-dimensional simulation module to the display screen to reduce latency and ensure immediate visual feedback to the user.

Second, the user interface provides an interactive interface that allows the user to interact directly with the three-dimensional model. This includes performing operations such as opening or closing a device switch, adjusting a device setting, or simulating a fault response. To make these operations intuitive and understandable, the user interface has designed a series of interactive elements, such as buttons, sliders and drag controls, that intuitively correspond to the corresponding device components in the three-dimensional model. In addition, to further enhance the learning experience, the simulated operation performed may be feedback simulated by the physical engine and presented to the user in visual or other sensory form through a user interface, such as by sound or vibration feedback of the physical effects of the simulated operation.

Finally, the user interface is responsible for receiving teaching adjustment instructions from the intelligent teaching module 102 and adjusting the displayed teaching content and difficulty accordingly. This means that the user interface needs to have the ability to dynamically update the content in order to adjust the teaching strategy and learning materials according to the learning progress and performance of the user. To achieve this goal, the user interface has devised a flexible content management system that enables rapid loading and replacement of tutorials, including text descriptions, images, video tutorials, and interactive simulation exercises.

To implement this user interface, it should be appreciated that its design is not just for aesthetic and user-friendly purposes, but rather to support efficient learning and operational exercises. This requires a tight integration and coordination of the user interface with the three-dimensional simulation module and the intelligent teaching module to ensure a seamless, interactive learning environment is provided. Through the design, the user interface 103 not only greatly improves the usability and effectiveness of the three-dimensional simulation training system of the power grid automation equipment, but also provides an attractive and participatory learning platform for users.

Still further, the user interface includes a dashboard that dynamically displays the user's learning progress, completed tutorial units and upcoming tutorials.

In the three-dimensional simulation training system of the power grid automation equipment, one of design keys of a user interface is to provide a dashboard, which not only enhances user experience, but also is crucial for promoting the learning process of a user. The core function of the instrument panel is to dynamically show the learning progress of the user, including the completed teaching units and upcoming teaching contents, so that the user can grasp the learning state of the user at a glance, thereby planning and adjusting the learning plan of the user more effectively.

The key steps for realizing the user dashboard comprise:

1. Design and layout: the design of the instrument panel is simple and visual, and is easy to understand and operate by a user. The dashboard may be divided into sections in layout, such as "my learning progress", "completed teaching unit" and "content to learn", each of which should provide clear, specific information. Learning progress and achievements can be presented more intuitively using charts, progress bars, or color coding.

2. Data integration and real-time update: the dashboard needs to integrate data from the three-dimensional simulation module and the intelligent teaching module and update this information in real time. This requires the back-end system to be able to process the user's interaction data, identify and record the teaching units completed by the user, and update the upcoming teaching content according to the recommendations of the intelligent teaching module.

3. And (3) personalized display: the dashboard should be able to present customized information based on the user's personal learning history and preferences. This may involve an analysis of the user's behavior by an algorithm to predict the next teaching unit that is most likely to be of interest to the user, or to adjust the content and difficulty of the presentation to accommodate the learning speed and level of the user.

4. Interaction function: in addition to presenting information, the dashboard should provide certain interactive functions, such as enabling the user to mark specific teaching units for review, or to adjust the priority of the content to be learned. In addition, the dashboard may also provide a feedback channel that allows the user to suggest or evaluate the teaching.

5. The technology is realized: at a technical level, achieving such an instrument panel requires close cooperation of the front end and the back end. The front end is responsible for design and user interaction of the instrument panel, and a user-friendly interface is constructed by using technologies such as HTML, CSS, javaScript and the like; the backend then needs to handle the storage, querying and updating of the data, possibly involving the operation of the database and server-side programming.

Through the steps, a user dashboard in the three-dimensional simulation training system of the power grid automation equipment can be realized, and an intuitive, practical and personalized learning progress tracking and planning tool is provided for users. The instrument panel not only can help users to better master the learning condition of the users and motivate the users to make progress, but also can further optimize teaching contents through feedback and interaction, thereby realizing more efficient and personalized learning experience.

Furthermore, the user interface is provided with an interactive question-answering function, through which a user can directly put forth questions related to the three-dimensional simulation module on the user interface and receive customized answers generated based on the intelligent teaching module, so that the interactivity and depth of learning are enhanced.

In the three-dimensional simulation training system of the power grid automation equipment, in order to make the learning process more interactive and deep, the user interface is designed to comprise an interactive question-and-answer function. This feature allows the user to directly ask questions through the interface, which may be guidelines for the operation of the three-dimensional simulation module, the interpretation of theoretical knowledge points, or the question of specific teaching. The system may then utilize the intelligent teaching module to generate and provide customized answers to the user questions.

The step of realizing the interactive question-answering function comprises the following steps:

1. And (3) designing a question-answering interface: first, it is necessary to design an easy-to-use question-answering portion in the user interface. This may be a simple text entry box, allowing users to enter their questions, and a submit button to send the questions. To enhance the user experience, voice input options may also be provided to enable the user to ask questions directly via voice.

2. Natural Language Processing (NLP): when a user submits a question, the system needs to understand and analyze the question. This involves natural language processing techniques including steps of language recognition, word segmentation, semantic understanding, etc. The purpose is to convert the natural language questions of the user into a form that the system can understand and handle.

3. Integration of intelligent teaching modules: the intelligent teaching module needs to have the ability to process user questions and generate customized answers. This may involve queries of the knowledge base, rule-based logical reasoning, or generating answers from existing teaching content and past question-answer records using machine learning models. Importantly, this module needs to be able to handle a wide range of question types and provide accurate, relevant information as answers.

4. Answer presentation: the generated answer needs to be presented on the user interface in a user friendly manner. This includes not only presentation of text answers, but also charts, pictures, multimedia elements linked to related teaching units or videos, etc. to enrich the answer content, enhance understanding and learning effects.

5. Feedback and iteration: to continuously optimize the interactive question-and-answer function, the system should provide a way for the user to evaluate the answers received, such as a "useful" or "not useful" flag. The system learns and adjusts based on these feedback to improve the accuracy and relevance of question processing and answer generation.

Through integrating the interactive question-answering function, the three-dimensional simulation training system of the power grid automation equipment can provide a more interactive and responsive learning environment, greatly enhance the interactivity and depth of learning, simultaneously provide instant help for users, solve the doubt in the learning process, and further promote the overall learning experience and effect.

In the three-dimensional simulation training system of the power grid automation equipment, one core function of the user interface design is a mode switching function. This functionality aims to improve user experience and learning efficiency by allowing the user to freely switch between different views and modes of operation according to personal preferences to suit the needs and learning environment of different users.

To achieve this, a flexible and easy to operate user interface needs to be designed and developed first. This interface should contain obvious user controls such as buttons, sliders or drop-down menus that enable the user to easily access and activate the mode switch function.

The implementation steps of the design mode switching function include:

Realization of different views: first, it is necessary to define the different view types that may be involved in the system. This may include, but is not limited to, simplified views, detailed views, night views, and the like. The simplified view can provide basic information and the most necessary operation control for a beginner or a user needing quick review; the detailed view is suitable for deep learning and provides rich teaching contents and advanced operation options; the night view adjusts the color scheme to reduce the influence of the screen light on the eyes of the user, and is suitable for night study.

Differentiation of operation modes: the differentiated design of the operation modes should take into account the different demands of the user in terms of interaction. For example, the touch screen mode optimizes touch operation, and the keyboard mouse mode is suitable for the traditional desktop environment. In addition, it is also conceivable to introduce a voice control mode, allowing the user to operate by voice commands to support unobstructed learning.

Recording and application of user preferences: the system needs to be able to record the user's mode selection preferences and automatically apply these preference settings the next time the user logs in. This requires the backend development to support the storage and retrieval functions of the user profile and to apply these configurations when the user interface is loaded.

Through the steps, the user interface of the three-dimensional simulation training system of the power grid automation equipment has a highly flexible and personalized mode switching function, so that a user can adjust a learning interface according to personal preference and the current environment, and learning experience is optimized. The design considers the requirements and the use scenes of different users, so that the system is more humanized and easy to use, and the users can fully utilize the three-dimensional simulation training system to learn effectively.

While the application has been described in terms of preferred embodiments, it is not intended to be limiting, but rather, it will be apparent to those skilled in the art that various changes and modifications can be made herein without departing from the spirit and scope of the application as defined by the appended claims.

Claims

1. A three-dimensional simulated training system for power grid automation equipment, comprising:

2. The three-dimensional simulation training system of the power grid automation equipment according to claim 1, wherein the graphic processing unit is used for executing a ray tracing algorithm to simulate the ray behavior under the complex illumination condition in the power grid automation environment and the interaction of the ray and the surface material of the power grid automation equipment, wherein the ray behavior comprises the scattering, reflection and refraction of the ray; the ray tracing algorithm comprises the following steps:

；

3. The grid automation device three-dimensional simulation training system of claim 2, wherein the graphics processing unit is configured to perform a texture mapping algorithm; the texture mapping algorithm uses the following equation 3 for texture adjustment:

；

4. The three-dimensional simulation training system of the power grid automation equipment according to claim 1, wherein the intelligent module adopts a trained hybrid neural network model to dynamically determine teaching plans and difficulty levels matched with actual operation levels of users; the hybrid neural network model comprises a decision tree network, a behavior pattern recognition network and a comprehensive evaluation network;

5. The grid automation device three dimensional simulation training system of claim 4, wherein the decision tree network comprises a dynamic feature selection mechanism that automatically identifies and selects features most relevant to user learning progress and skill level based on a machine learning algorithm; wherein the most relevant features include speed variation of user operation, accuracy of operation, and improved speed of user on operation task.

6. The three-dimensional simulation training system of power grid automation equipment according to claim 4, wherein the behavior pattern recognition network adopts a mixed structure of a convolutional neural network and a cyclic neural network.

7. The three-dimensional simulation training system of power grid automation equipment of claim 4, wherein the comprehensive evaluation network comprises an adaptive learning module based on user feedback, the adaptive learning module dynamically adjusting weights and parameters of the comprehensive evaluation network using direct feedback from a user on satisfaction of the teaching content.

8. The grid automation device three dimensional simulation training system of claim 1, wherein the user interface comprises a dashboard for dynamically showing the user's learning progress, completed teaching units and upcoming teaching content.

9. The three-dimensional simulation training system of the power grid automation device of claim 1, wherein the user interface comprises an interactive question-and-answer function that enables a user to submit questions related to the three-dimensional simulation module directly through the user interface and to receive customized answers generated by the intelligent teaching module.

10. The grid automation device three dimensional simulation training system of claim 1, wherein the user interface includes a mode switching function that allows a user to switch between different views and modes of operation according to personal preferences.