Operation method and operation terminal for data center electrical system simulation

Abstract

The invention relates to the technical field of electric appliance simulation, in particular to an operation method and an operation terminal for data center electrical system simulation, and the method comprises the following steps: S1, obtaining a trainee terminal operation video collected by a video collection device and trainee terminal operation step text data collected by a simulation system database; s2, performing feature extraction on the trainee terminal operation video to generate a trainee terminal operation attention key feature vector; s3, performing feature extraction on the trainee terminal operation step text data to generate a terminal operation step text association feature vector; s4, fusing the trainee terminal operation attention key feature vector and the terminal operation step text association feature vector based on a feature topology association method driven by feature value granularity association information migration; and S5, generating a trainee operation process report based on the trainee operation feature vector.

Description

Technical Field

[0001] The invention relates to the technical field of electrical appliance simulation, and in particular to an operation method and an operation terminal for simulating an electrical system of a data center. Background Art

[0002] The power supply system relies on multiple independent power sources as the main power supply source, adopts a distributed redundant architecture, and supplies power in parallel operation. In addition, the power supply system is equipped with an N+1 full-load capacity emergency power diesel generator system, and uses dual bus output to achieve physical isolation. Therefore, the professional skills of the operation and maintenance personnel of the power supply system are extremely high. In the prior art, the training of operation and maintenance personnel usually includes two forms: on-site teaching and simulation teaching. Currently, simulation teaching on the market mostly uses digital twin technology to develop a three-dimensional power supply and distribution simulation software system. Due to the strict operating procedures and complex steps of the power supply and distribution system, the traditional digital twin three-dimensional power supply and distribution simulation system interacts through a mouse, keyboard or VR helmet. This virtual software operation lacks a sense of reality and immersion in the actual situation, resulting in poor operation training results and low efficiency, which seriously hinders the improvement of the operation skills of operation and maintenance personnel.

[0003] Therefore, there is an urgent need to design an operation method and an operation terminal for data center electrical system simulation. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an operation method and an operation terminal for data center electrical system simulation, which are used to collect operation videos of trainees and text data of trainees' operation steps, and analyze and associate the video and text data through deep learning technology to generate a report to provide personalized training plans for different trainees, enhance the learning experience of trainees, improve operation skills, make up for the lack of real scenes in traditional simulation teaching, and improve training efficiency.

[0005] In order to achieve the above-mentioned purpose, an operation method for data center electrical system simulation is designed, comprising the following steps: S1. Acquire a terminal operation video of a trainer collected by a video acquisition device and a text data of the terminal operation steps of the trainer collected by a simulation system database; S2. Perform feature extraction on the terminal operation video of the trainer to generate a key feature vector of the terminal operation attention of the trainer; S3. Perform feature extraction on the text data of the terminal operation steps of the trainer to generate a text-related feature vector of the terminal operation steps; S4. Based on a feature topology association method driven by feature value granularity association information walking, the key feature vector of the terminal operation attention of the trainer and the text-related feature vector of the terminal operation steps are merged, specifically comprising: S41. Construct a training distance topology between the key feature vector of the terminal operation attention of the trainer and the text-related feature vector of the terminal operation steps. matrix; S42. Construct a training eigenvalue granularity association matrix between the trainer's terminal operation attention key feature vector and the terminal operation step text associated feature vector; S43. Perform topological association modulation on the training eigenvalue granularity association matrix based on the training distance topological matrix to obtain a training topological modulation association matrix; S44. Project the trainer's terminal operation attention key feature vector and the terminal operation step text associated feature vector to the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text associated feature vector; S45. Fusion the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text associated feature vector to generate a trainer operation feature vector; S5. Generate a trainer operation process report based on the trainer operation feature vector.

[0006] Preferably, the operation method provided by the present invention also includes other technical features, wherein the feature extraction of the trainer terminal operation video in step S2 includes: extracting multiple key frames from the video; performing feature encoding based on the spatial attention mechanism on the multiple key frames to generate multiple trainer terminal operation attention key feature maps; pooling and cascading the multiple trainer terminal operation attention key feature maps to generate trainer terminal operation attention key feature vectors; the feature encoding based on the spatial attention mechanism includes: using the convolution encoding part to perform deep convolution encoding on the multiple key frames to obtain an initial convolution feature map; inputting the initial convolution feature map into the spatial attention part to obtain a spatial attention map; processing the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; calculating the positional point multiplication of the spatial attention feature map and the initial convolution feature map to obtain multiple trainer terminal operation attention key feature maps.

[0007] Preferably, the operation method provided by the present invention also includes other technical features, wherein the feature extraction of the terminal operation step text data of the trainer in step S3 includes: dividing the operation step text data into multiple terminal operation step text data; using a converter-based terminal operation step text context encoder to encode the multiple terminal operation step text data to generate multiple terminal operation step text feature vectors; cascading the multiple terminal operation step text feature vectors into a terminal operation step text associated feature vector; the converter-based terminal operation step text context encoder includes: an embedding layer, used to map each operation step word in the operation step text data into a word embedding vector; a converter-based Bert model, used to perform semantic encoding on the word embedding vector based on the global context to generate a terminal operation step text feature vector.

[0008] Preferably, the operating method provided by the present invention also includes other technical features, wherein generating a training personnel operation process report in step S5 includes: analyzing the training personnel operation feature vector to generate key findings and improvement suggestions for the operation process; converting the analysis results into an easy-to-understand language description through natural language generation technology; generating a structured report containing a summary, key findings, data analysis results and improvement suggestions.

[0009] The present invention also provides an operation terminal for simulating the electrical system of a data center, which includes the simulation operation method, and comprises: a training staff operation data acquisition module, which is used to acquire the training staff terminal operation video collected by a camera and the training staff terminal operation step text data collected by a simulation system database; a training staff operation data extraction module, which is used to extract the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector from the trainer terminal operation video collected by the camera and the trainer terminal operation step text data collected by the simulation system database; a training staff operation process report generation module, which is used to generate a trainer operation process report based on the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector.

[0010] Preferably, the operation terminal provided by the present invention also includes other technical features, wherein the trainer operation data extraction module includes: a trainer terminal operation video feature extraction unit, which is used to perform feature extraction on the trainer terminal operation video collected by the camera to obtain the trainer terminal operation attention key feature vector; a trainer terminal operation step text feature extraction unit, which is used to perform feature extraction on the trainer terminal operation step text data collected by the simulation system database to obtain the terminal operation step text associated feature vector; wherein the trainer terminal operation video feature extraction unit includes: a trainer terminal operation key frame extraction subunit, which is used to extract key frames from the trainer terminal operation video collected by the camera to obtain multiple trainer terminal operation key frames; a trainer terminal operation key frame attention encoding subunit, which is used to pass the multiple trainer terminal operation key frames through a trainer terminal operation feature encoder based on a spatial attention mechanism to obtain multiple trainer terminal operation attention key feature maps; a trainer terminal operation attention feature aggregation subunit, which is used to pool and cascade the multiple trainer terminal operation attention key feature maps to obtain a trainer terminal operation attention key feature vector.

[0011] Preferably, the operation terminal provided by the present invention also includes other technical features, wherein the trainer terminal operation key frame attention encoding subunit is used to: use the convolution encoding part of the trainer terminal operation feature encoder based on the spatial attention mechanism to perform deep convolution encoding on the multiple trainer terminal operation key frames to obtain an initial convolution feature map; input the initial convolution feature map into the spatial attention part of the trainer terminal operation feature encoder based on the spatial attention mechanism to obtain a spatial attention map; pass the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; calculate the position point multiplication of the spatial attention feature map and the initial convolution feature map to obtain the multiple trainer terminal operation attention key feature maps.

[0012] Preferably, the operation terminal provided by the present invention also includes other technical features, wherein the training personnel terminal operation step text feature extraction unit includes: an operation step text division subunit, used to divide the training personnel terminal operation step text data collected by the simulation system database into multiple terminal operation step text data; an operation step text context encoding subunit, used to pass the multiple terminal operation step text data through a converter-based terminal operation step text context encoder to obtain multiple terminal operation step text feature vectors; a terminal operation step feature cascade subunit, used to cascade the multiple terminal operation step text feature vectors into a terminal operation step text associated feature vector; wherein the terminal operation step feature cascade subunit manages the thread pool by creating an ExecutorService instance, decomposes the cascade task of multiple terminal operation step text feature vectors into multiple subtasks and submits them to the thread pool, uses a Future object to track the execution status of each subtask, merges the results after all subtasks are completed, and finally closes the thread pool to release resources.

[0013] Preferably, the operation terminal provided by the present invention also includes other technical features, wherein the trainer operation process report generation module includes: a trainer operation feature association unit, used to perform feature topology association based on feature value granularity association information walking-driven feature topology association on the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector to obtain the trainer operation feature vector; an operation process report generation unit, used to pass the trainer operation feature vector through a generator to generate a trainer operation process report.

[0014] Preferably, the operation terminal provided by the present invention also includes other technical features, wherein the trainer operation feature association unit includes: constructing a training distance topological matrix between the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector; constructing a training eigenvalue granularity association matrix between the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector; performing topological association modulation on the training eigenvalue granularity association matrix based on the training distance topological matrix to obtain a training topological modulation association matrix; using the training topological modulation association matrix as the perspective co-projection space, projecting the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector to the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text association feature vector; fusing the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text association feature vector to obtain the trainer operation feature vector.

[0015] Compared with the prior art, the present invention has the following advantages:

[0016] Compared with the prior art, the data center electrical system simulation operation terminal proposed in this application first collects the training personnel's operation video recorded by the camera and the operation step text data stored in the simulation system database. Then, deep learning technology is used to extract features and analyze the correlation of these data. Finally, the generator creates a trainer's operation process report, customizes a personalized training plan for each trainer, enhances their learning experience, and improves their operating skills. This process aims to make up for the lack of real scenes in traditional simulation teaching, thereby improving the efficiency of training. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 , is a structural block diagram of the operation terminal;

[0018] Figure 2 , is a structural block diagram of a training personnel operation data extraction module in an operation terminal;

[0019] Figure 3 , is a structural block diagram of a training personnel terminal operation video feature extraction unit in an operation terminal;

[0020] Figure 4 , is a structural block diagram of a text feature extraction unit for terminal operation steps of trainees in an operation terminal;

[0021] Figure 5 , is the structural block diagram of the training personnel operation process report generation module in the operation terminal. DETAILED DESCRIPTION

[0022] In order to make the purpose, principle and structure of the present invention more clear, it is further described below in conjunction with the drawings and specific embodiments.

[0023] Embodiment 1:

[0024] This embodiment provides an operation terminal for data center electrical system simulation

[0025] See also Figure 1The operation terminal 100 for data center electrical system simulation of the embodiment of the present application includes: a training staff operation data acquisition module 110, which is used to obtain the training staff terminal operation video collected by the camera and the training staff terminal operation step text data collected by the simulation system database; a training staff operation data extraction module 120, which is used to extract the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector from the trainer terminal operation video collected by the camera and the trainer terminal operation step text data collected by the simulation system database; a training staff operation process report generation module 130, which is used to generate a trainer operation process report based on the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector.

[0026] In the above-mentioned operation terminal 100 for data center electrical system simulation, the training personnel operation data acquisition module 110 is used to obtain the training personnel terminal operation video collected by the camera and the training personnel terminal operation step text data collected by the simulation system database. It should be understood that the design of the power supply system adopts multiple independent power supplies as common power supplies, and adopts a distributed redundant architecture to supply power in a parallel operation mode. In addition, the system is also equipped with an emergency power supply diesel generator system with N+1 full load capacity, and physical isolation is achieved through dual bus output. This design places extremely high demands on the professional level of operation and maintenance personnel. The training of operation and maintenance personnel usually adopts two methods: on-site teaching and simulation teaching. At present, most of the simulation teaching on the market is based on digital twin technology, and a three-dimensional power supply and distribution simulation software system has been developed. Due to the strict operating procedures and complex steps of the power supply and distribution system, the three-dimensional simulation system of the digital twin interacts through a mouse, keyboard or VR helmet. This virtual software operation lacks the sense of substitution of the actual situation, resulting in poor operation training effect and low efficiency, which seriously affects the improvement of the operation level of the operation and maintenance personnel.

[0027] In response to the above technical problems, the technical solution of this application designs an operation terminal for data center electrical system simulation. The terminals are medium voltage operation terminals and low voltage operation terminals. The terminal structure is consistent with the power supply and distribution equipment of the data center. The operation terminal is connected to the power supply and distribution simulation system through the network, receives instructions from the simulation system, and the trainees operate according to the operating steps of the actual system. All operation data are automatically fed back to the simulation system. Since the trainees operate according to the operating steps of the actual equipment on the real equipment, the problem of the original simulation system being fully virtual, unrealistic, and having poor operation effect is solved.

[0028] Specifically, in the technical solution of the present application, by acquiring the terminal operation video of the trainer collected by the camera and the text data of the terminal operation steps of the trainer collected by the simulation system database, and combining the deep learning technology, the operation process report of the trainer is generated, and then a personalized training program is tailored for each trainer, so as to enhance their learning experience and improve their operation skills. In this way, the lack of real scenes in traditional simulation teaching is made up, thereby improving the overall efficiency of training. Among them, the acquisition of video data usually depends on the camera installed in the training environment. The camera can record the operation of the trainer on the terminal in real time, and the captured picture includes the changes in the operation interface, the operation gestures and expressions of the personnel. Data storage can choose local hard disk or cloud storage for subsequent retrieval and analysis. The acquisition of text data depends on the database of the simulation system. The database records every step of the trainer when operating the terminal, including the input instructions, system response, and the decision-making process of the trainer. Usually, these data are saved in the form of log records to ensure that the detailed information of each operation can be captured in a timely and accurate manner. Through the database interface, these text data can be easily exported, usually in CSV or JSON format, for subsequent data processing and analysis. In this way, training institutions can comprehensively analyze the operating behaviors of trainees, evaluate the training effects, identify problems in operations, and make targeted improvements. This data integration not only helps optimize the training process, but also provides a basis for the subsequent training course design, improving the training effect and efficiency.

[0029] In the above-mentioned operation terminal 100 for data center electrical system simulation, the training staff operation data extraction module 120 is used to extract the training staff terminal operation attention key feature vector and the terminal operation step text association feature vector from the training staff terminal operation video collected by the camera and the training staff terminal operation step text data collected by the simulation system database. It should be understood that in this way, a comprehensive analysis of the training staff's operation behavior can be achieved, providing data support for the formulation of personalized training programs, and effectively guiding operation and maintenance personnel to better deal with complex scenarios in actual operations and improve their operation capabilities and efficiency.

[0030] Embodiment 2:

[0031] See also Figure 2On the basis of Embodiment 1, this embodiment specifically provides a technical solution for the training staff operation data extraction module 120, including: a training staff terminal operation video feature extraction unit 121, used for performing feature extraction on the training staff terminal operation video collected by the camera to obtain the trainer terminal operation attention key feature vector; a training staff terminal operation step text feature extraction unit 122, used for performing feature extraction on the training staff terminal operation step text data collected by the simulation system database to obtain the terminal operation step text associated feature vector.

[0032] Among them, feature extraction of the terminal operation video of the trainer collected by the camera helps to deeply understand the trainer's behavior and decision-making during the operation process. Among them, by analyzing the trainer's line of sight distribution, hand movements and operation steps in the video, their focus and attention distribution when performing tasks can be identified, thereby revealing the complexity and error-prone links of the operation. Specifically, first, the video data needs to be preprocessed, including decomposing into separate frames, adjusting resolution and standardization to ensure the consistency and analyzability of the data. Next, a deep learning model is applied to process each frame. Through model training, key actions such as the trainer's facial expressions, eye movement trajectories and hand operations can be identified. These key actions can be converted into attention key feature vectors, reflecting the trainer's focus area and attention concentration during operation. Finally, through the analysis of feature vectors, institutions can improve the training process and optimize teaching strategies in a targeted manner, thereby effectively improving the trainer's operational capabilities and work efficiency.

[0033] Furthermore, feature extraction of the text data of the terminal operation steps of trainees collected by the simulation system database can reveal the logical relationship and dependency between the operation steps, and provide trainees with more structured and systematic learning resources. This associated feature vector can not only reflect the content of each operation step, but also help identify the order and relationship between different steps, thereby providing guidance for the correctness and efficiency of the operation. By analyzing these feature vectors, training managers can evaluate the learning situation of trainees, optimize training content and strategies, and thus improve the accuracy and efficiency of operations.

[0034] Embodiment three:

[0035] See him Figure 3On the basis of the first or second embodiment, the present embodiment specifically provides a technical solution of the trainer terminal operation video feature extraction unit 121, including: a trainer terminal operation key frame extraction subunit 1211, used to extract key frames from the trainer terminal operation video collected by the camera to obtain multiple trainer terminal operation key frames; a trainer terminal operation key frame attention encoding subunit 1212, used to pass the multiple trainer terminal operation key frames through a trainer terminal operation feature encoder based on a spatial attention mechanism to obtain multiple trainer terminal operation attention key feature maps; a trainer terminal operation attention feature aggregation subunit 1213, used to pool and cascade the multiple trainer terminal operation attention key feature maps to obtain a trainer terminal operation attention key feature vector.

[0036] Among them, extracting key frames from the terminal operation video of the trainer captured by the camera can effectively capture the important moments and behavioral changes of the trainer during the operation process, help analyze the operation process, identify problems and optimize the training content. Specifically, first, the video needs to be decomposed and divided into several continuous frames. Next, algorithms (such as scene change detection or image similarity calculation) are used to identify key frames. These algorithms can compare the differences between consecutive frames and find those frames with significant changes in visual content, such as the start, end or completion of important steps of the operation. By observing the key frames, training institutions can deeply understand the performance of participants during the training process, and then adjust the training content to meet the needs of trainees.

[0037] Furthermore, multiple trainer terminal operation key frames are processed by the trainer terminal operation feature encoder based on the spatial attention mechanism, which can dynamically adjust the network's attention to different areas, emphasize key operations and ignore unimportant information, so as to more effectively capture the trainer's attention allocation and behavior patterns in terminal operations. Among them, the feature encoder based on the spatial attention mechanism is usually composed of a convolutional neural network, whose task is to extract visual features in the frame. The core of the spatial attention mechanism is to determine which parts the model should focus on when extracting features by calculating attention weights. This can be achieved by generating an attention map, which assigns a weight value to each spatial position, and the area with a higher weight value represents more important information. By multiplying with the feature map, the model can enhance the expression of important features while suppressing the influence of irrelevant areas. This mechanism effectively focuses on key areas in the operation process, such as the user's hand movements, buttons operated on the screen, or important interface changes. In this way, the extracted key feature maps of multiple trainer terminal operations can not only reflect the spatial distribution of key actions in the operation process, but also provide an important basis for subsequent data analysis and model training. These feature maps can help analyze the behavioral patterns of trainees when performing tasks, identify problems and difficulties in operations, and provide data support for the optimization of training programs.

[0038] Furthermore, pooling multiple attention feature maps can reduce the dimension of the feature map while retaining important spatial information. Commonly used pooling methods include maximum pooling and average pooling. Maximum pooling retains the most significant features by selecting the maximum value in the feature map, while average pooling obtains a smoother feature representation by calculating the average value of the feature map. In the technical solution of the present application, the maximum pooling method is used because it can highlight the extreme behaviors or significant features of trainers in key operations. After pooling, the obtained feature map is cascaded to form a long feature vector. This process forms a comprehensive feature representation by connecting multiple pooled feature maps in a certain order. The cascade operation can not only fuse information from different key frames, but also reflect the temporal dynamic changes and attention distribution during the operation. This integration method ensures that all important details of the operation process are included in the feature vector, providing rich information for subsequent analysis and model training.

[0039] In a specific embodiment of the present application, the trainer terminal operation key frame attention encoding subunit 1212 is used to: use the convolution encoding part of the trainer terminal operation feature encoder based on the spatial attention mechanism to perform deep convolution encoding on the multiple trainer terminal operation key frames to obtain an initial convolution feature map; input the initial convolution feature map into the spatial attention part of the trainer terminal operation feature encoder based on the spatial attention mechanism to obtain a spatial attention map; pass the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; calculate the position point multiplication of the spatial attention feature map and the initial convolution feature map to obtain the multiple trainer terminal operation attention key feature maps. More specifically, the initial convolutional feature map is input into the spatial attention part of the trainer terminal operation feature encoder based on the spatial attention mechanism to obtain a spatial attention map, including: performing average pooling and maximum pooling along the channel dimension on the initial convolutional feature map to obtain an average feature matrix and a maximum feature matrix; cascading and channel-adjusting the average feature matrix and the maximum feature matrix to obtain a channel feature matrix; and convolutionally encoding the channel feature matrix using the convolution layer of the spatial attention feature map to obtain the spatial attention map.

[0040] It should be understood that the convolutional coding part in the feature encoder based on the spatial attention mechanism extracts low-level features in the key frame through convolution operations, captures basic information such as edges, textures, and shapes, and lays the foundation for subsequent attention calculations. Next, the initial convolutional feature map is input into the spatial attention part of the feature encoder to calculate the spatial attention map. This process is to generate a weight map by analyzing the importance of each region in the initial feature map, reflecting the contribution of different regions to attention during the operation. Subsequently, the spatial attention map is processed by the Softmax activation function to obtain the spatial attention feature map. The role of the Softmax function is to standardize the attention weights so that it can effectively reflect the relative importance of different features in the overall feature map. After Softmax processing, the sum of the weights in the feature map is 1, which is more convenient for subsequent calculations and comparisons. Finally, the calculated spatial attention feature map is point-by-position multiplied with the initial convolutional feature map to generate the final multiple trainer terminal operation attention key feature map. The significance of this point multiplication operation is to highlight the information that is important in space by weighting, while suppressing irrelevant features, so that the final feature map is more focused on the key area. Through this systematic processing flow, key features can be effectively extracted, providing strong data support for subsequent analysis and decision-making.

[0041] Embodiment 4:

[0042] like Figure 4 As shown, on the basis of Embodiment 1, this embodiment specifically provides the trainer terminal operation step text feature extraction unit 122, including: an operation step text division subunit 1221, used to divide the trainer terminal operation step text data collected by the simulation system database into multiple terminal operation step text data; an operation step text context encoding subunit 1222, used to pass the multiple terminal operation step text data through a converter-based terminal operation step text context encoder to obtain multiple terminal operation step text feature vectors; a terminal operation step feature cascade subunit 1223, used to cascade the multiple terminal operation step text feature vectors into a terminal operation step text association feature vector.

[0043] It should be understood that dividing the terminal operation step text data of the trainer collected by the simulation system database into multiple terminal operation step text data can better understand and analyze the operation behavior in the training process. Specifically, the original terminal operation step text data needs to be preprocessed first. This includes removing irrelevant characters, standardizing text formats, unifying terms, etc. to ensure the consistency and readability of the data. Next, the text data is divided into multiple operation steps by defining clear division criteria. The division criteria can be based on the logical order of operations, timestamps, operation types, or specific keywords. For example, the text can be segmented according to symbols such as the start and end marks of the operation, line breaks, or semicolons. Once the division criteria are defined, the text data can be automatically divided into multiple steps using programming tools or text processing libraries (such as Python's NLTK or regular expressions). Each step should contain complete information, such as the name of the operation, the execution time, and the interface elements involved. Each terminal operation step text data after division can be regarded as an independent unit for further analysis and processing. In addition, the division process should also consider the correlation between the operation steps so that context information can be retained during analysis. By dividing the terminal operation step text data of trainees collected by the simulation system database into multiple independent operation step text data, each operation link in the training process can be analyzed more clearly, key behaviors can be identified and training content can be optimized, thereby improving the effectiveness of training and the learning experience of participants.

[0044] Furthermore, the text data of multiple terminal operation steps are processed by a terminal operation step text context encoder based on a converter, which can make full use of context information and extract deep semantic features in the text. Specifically, the text is processed using a converter-based encoder. The structure of the converter allows the model to capture the relationship and context information between each word in the text during the encoding process, and dynamically calculate the weight of each word using the self-attention mechanism, so as to more accurately understand the meaning of the entire sentence. In this process, multiple terminal operation step text data are input into the model, and the converter encodes the input text through a multi-layer stacked self-attention and feedforward neural network structure to generate an initial feature representation. Each feature vector not only contains the basic information of a single operation step, but also incorporates content related to the context, so that the model can identify the relationship and semantic association between words. Specifically, the multiple terminal operation step text data are passed through a converter-based terminal operation step text context encoder to obtain multiple terminal operation step text feature vectors, including: performing word segmentation processing on the multiple terminal operation step text data to obtain an operation step word sequence; using the embedding layer of the converter-based terminal operation step text context encoder to respectively map each operation step word in the operation step word sequence to a word embedding vector to obtain a sequence of operation step word embedding vectors; using the converter-based Bert model of the converter-based terminal operation step text context encoder to perform global context semantic encoding on the sequence of the operation step word embedding vectors to obtain multiple terminal operation step text feature vectors.

[0045] Furthermore, considering that multiple terminal operation step text feature vectors are generated through a transformer-based encoder, each of these vectors represents the unique semantics and features of different operation steps. In order to form an associated feature vector, these feature vectors need to be concatenated. Concatenation refers to connecting different feature vectors in sequence to generate a long feature vector. This operation helps to retain the information of each step and provide a richer feature basis for subsequent analysis.

[0046] Embodiment five:

[0047] Based on the first embodiment, this embodiment specifically provides the terminal operation step feature cascade subunit, which is used to: create an ExecutorService instance for managing the thread pool. Decompose the cascade task of multiple terminal operation step text feature vectors into multiple subtasks, and submit these subtasks to the thread pool; use the Future object to track the execution status of each subtask and wait for all tasks to be completed; after all subtasks are completed, merge their results into the final terminal operation step text associated feature vector; after the task is completed, close the thread pool to release resources.

[0048] It is worth noting that the above embodiment is implemented by deploying code, and the deployment code is a method for processing the cascade tasks of multiple terminal operation step text feature vectors in parallel using the ExecutorService mechanism of Java. The core of the method is to decompose the cascade task into multiple subtasks by creating a fixed-size thread pool, and execute these subtasks in parallel using multithreading, thereby improving processing efficiency. Among them, the code first creates a fixed-size thread pool, and the number of threads is equal to the number of available processors. This ensures that the thread pool can make full use of system resources while avoiding the waste of resources caused by excessive thread creation. Then, the cascade task of each terminal operation step text feature vector is encapsulated as a ConcatenateTask object and submitted to the thread pool for execution. The ConcatenateTask class implements the Callable interface, and its call method contains specific cascade logic. Then, through the Future object, the code can track the execution status of each subtask and collect their results after all tasks are completed. The Future object is stored in a list to facilitate the subsequent result merging. Then, the results of all subtasks are merged into a final terminal operation step text association feature vector. The concatenateResults merge method receives a list of all subtask results and merges them into a complete feature vector. Finally, after the task is completed, the code calls the shutdown method of ExecutorService to shut down the thread pool and release system resources. Through the above deployment code, the system can efficiently process the cascade tasks of multiple terminal operation step text feature vectors in parallel, significantly improving the processing speed and efficiency. Among them, by using multi-threaded parallel processing, the system can process the cascade tasks of multiple feature vectors at the same time, reducing the time overhead of single-threaded serial processing. Especially when processing large-scale data, this parallel processing method can significantly shorten the task completion time. By creating a fixed-size thread pool, the system can make full use of the computing power of multi-core processors, avoid the overhead of thread creation and destruction, and improve resource utilization.

[0049] Here, the ExecutorService mechanism provides a framework for task scheduling and management, ensuring that tasks can be executed in an orderly manner, avoiding resource competition and deadlock problems that may be caused by concurrent execution of multiple threads, and enhancing the stability of the system. Through the Future object, the system can easily track and manage the execution status of each subtask, ensuring that the results are merged only after all tasks are completed, simplifying the task management process. The deployment code has a clear structure and is easy to expand and maintain. The thread pool size and task decomposition logic can be adjusted according to actual needs to adapt to different application scenarios. By using the ExecutorService mechanism to parallelly process the cascade tasks of the terminal operation step text feature vector, the system has achieved significant improvements in processing speed, resource utilization, system stability, and task management, providing strong support for simulation training of data center electrical systems.

[0050] In the above-mentioned operation terminal 100 for data center electrical system simulation, the trainer operation process report generation module 130 is used to generate a trainer operation process report based on the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector. It should be understood that the operation process report generated by integrating the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector can provide in-depth analysis and insights for training evaluation, help optimize training content and methods, and thus improve the overall effect of training and the learning experience of participants.

[0051] Embodiment six:

[0052] See also Figure 5 On the basis of the above-mentioned embodiments, this embodiment specifically discloses the training staff operation process report generation module 130, including: a training staff operation feature association unit 131, used for performing feature topological association based on feature value granularity association information walking driven feature topology association on the training staff terminal operation attention key feature vector and the terminal operation step text association feature vector to obtain the training staff operation feature vector; an operation process report generation unit 132, used for passing the training staff operation feature vector through a generator to generate a training staff operation process report.

[0053] The trainer's terminal operation attention key feature vector usually reflects the key areas that the trainer focuses on and their attention distribution when performing tasks, and can reveal which operation steps or elements are most critical in their operations. The terminal operation step text association feature vector provides the relationship and context information between the operation steps, helping to understand the interaction and dependency of different steps in the entire operation process.

[0054] In particular, in the technical solution of the present application, the key feature vector of the terminal operation attention of the trainer is extracted from the key frames in the camera video, and the important operation links in the video are encoded using the spatial attention mechanism, which emphasizes the dynamic and temporal characteristics of the operation, is highly dependent on image information, and reflects the behavioral characteristics, gestures, facial expressions and other subtle changes of the trainer when performing the operation, showing a strong spatiotemporal correlation. On the other hand, the terminal operation step text association feature vector is derived from the text data in the simulation system database, which represents the logic and semantic information of the operation steps. The text is processed by the converter-based encoding technology, focusing on the semantic relationship and context information between the steps. Therefore, its features are relatively abstract, and the focus is on the order, structure and rules of the operation, rather than the specific operation execution details. This text feature vector has strong language structure and sequence, but compared with image features, its dynamic information is relatively weak, reflecting more static knowledge and instructions. Due to the fundamental differences in the expression content, data form and generation method of the key feature vector of the terminal operation attention of the trainer and the text association feature vector of the terminal operation step, they have significant incompatibility in the feature space. When trying to fuse the key feature vectors of the trainer's terminal operation attention with the feature vectors associated with the terminal operation steps text, this heterogeneity will cause the feature manifold to be non-smooth, making it difficult for the fused trainer operation feature vectors to capture unified semantics and dynamic patterns. The non-smoothness of this manifold not only reduces the stability of the feature representation, but also affects the accuracy and consistency of the subsequently generated trainer operation process reports. Therefore, in the technical solution of the present application, the key feature vectors of the trainer's terminal operation attention and the feature vectors associated with the terminal operation steps text are subjected to feature topological association based on the feature value granularity association information walk-driven to obtain the trainer operation feature vector.

[0055] Among them, the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the terminal operation steps text are subjected to feature topological association based on the feature value granularity association information walking drive to obtain the trainer operation feature vector, including: constructing a training distance topological matrix between the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the terminal operation steps text; constructing a training feature value granularity association matrix between the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the terminal operation steps text; performing topological association modulation on the training feature value granularity association matrix based on the training distance topological matrix to obtain a training topological modulation association matrix; using the training topological modulation association matrix as the perspective co-projection space, projecting the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the terminal operation steps text to the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation steps text association feature vector; fusing the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation steps text association feature vector to obtain the trainer operation feature vector.

[0056] The fusion step is specifically expressed as:

[0057] M t =D(V1,V2);

[0058] D(V1, V2) is the training distance topological matrix between the key feature vector of the terminal operation attention of the trainer and the text-related feature vector of the terminal operation step, that is, d i,j =d(v 1i ,v 2j ), and V1 and V2 are both column vectors.

[0059]

[0060] V c =α·V1′+β·V2′.

[0061] Among them, V1 represents the key feature vector of the terminal operation attention of the trainer, V2 represents the text-related feature vector of the terminal operation steps, and M t represents the training distance topology matrix, T represents the transpose of the vector, Represents matrix multiplication, M v represents the training eigenvalue granularity association matrix, M c represents the training topology modulation association matrix, V1′ represents the key feature vector of the terminal operation attention of the viewpoint modulation trainer, V2′ represents the text association feature vector of the terminal operation step of the viewpoint modulation, α and β represent weighted hyperparameters, V crepresents the training staff operational feature vector.

[0062] In the technical solution of the present application, the key feature vector of the trainer's terminal operation attention and the feature vector associated with the terminal operation step text have relatively significant feature heterogeneity, which results in that when the feature fusion of the key feature vector of the trainer's terminal operation attention and the feature vector associated with the terminal operation step text is performed, the feature manifold and feature space mismatch caused by the heterogeneity will cause the trainer's operation feature vector to fail to have feature manifold smoothness, thereby affecting the certainty of its feature expression.

[0063] Based on this, in the technical solution of the present application, the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the text of the terminal operation steps are subjected to feature topological association based on the feature value granularity association information walk-driven, which first constructs the training distance topological matrix between the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the text of the terminal operation steps. This step utilizes the idea of ​​graph theory, takes the feature values ​​of each position in the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the text of the terminal operation steps as nodes, and uses the distance metric function to determine the low-dimensional embedding expression of the edges between nodes, so as to construct a feature expression that can quantify the topological association relationship between the key feature vector of the terminal operation attention of the trainer and the feature vector associated with the text of the terminal operation steps. This step is crucial to understanding the spatial distribution of data points and their mutual connections, and also provides basic structural information for subsequent steps.

[0064] Next, a training eigenvalue granularity association matrix is ​​constructed between the key feature vector of the trainer's terminal operation attention and the terminal operation step text-associated feature vector. This step focuses on exploring the correlation between the internal attributes of the feature vector. In a specific example, the product between the transposed vectors of the trainer's terminal operation attention key feature vector and the terminal operation step text-associated feature vector is calculated to obtain the training eigenvalue granularity association matrix. Furthermore, the training eigenvalue granularity association matrix is ​​topologically modulated based on the training distance topological matrix to obtain a training topological modulation association matrix. That is, the eigenvalue granularity association information of the trainer's terminal operation attention key feature vector and the terminal operation step text-associated feature vector is driven to wander on the distance topological map to construct a low-dimensional modulated perspective co-projection space domain for mapping the trainer's terminal operation attention key feature vector and the terminal operation step text-associated feature vector to a continuous high-dimensional regression space property.

[0065] Next, the training topology modulation association matrix is ​​used as the perspective co-projection space, and the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector are projected into the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text association feature vector. That is, the key to this technical solution is to find an implicit third space with modulation capability, in which as much information of the original high-dimensional data as possible is retained, so that the converted new feature vector not only reduces the dimension, but also maintains the relative position relationship between the original features.

[0066] Finally, the view modulation trainer terminal operation attention key feature vector and the view modulation terminal operation step text association feature vector are fused to obtain the trainer operation feature vector. Fusion can be achieved in various forms, ranging from simple arithmetic average to complex ensemble learning algorithms, depending on the application scenario and the goals pursued. The final trainer operation feature vector combines the main features of the two sets of input data and provides a more comprehensive data representation, which is not only conducive to improving the performance of machine learning models, but also promotes the ability of cross-modal or multi-source data analysis, thereby significantly enhancing the understanding and utilization efficiency of complex data structures.

[0067] Furthermore, the main reason for generating a trainer operation process report from the trainer operation feature vector through the generator is that it can present the key information of the training process in a systematic and structured manner, thereby providing intuitive and easy-to-understand data analysis for training managers. This kind of report can not only help identify the advantages and disadvantages in the operation, but also provide data support for the optimization of subsequent training and improve the overall training effect. Specifically, the generator receives the operation feature vector as input and generates structured report content after a series of processing steps. In the specific processing process, the generator can use natural language generation (NLG) technology to convert the numerical information in the feature vector into an easy-to-understand language description. For example, by analyzing the feature vector, the generator can point out which steps the trainer performs well and in which aspects there is room for improvement. Next, the report generation process can also include a data visualization step. For example, a chart, a bar chart or a heat map can be used to show the attention distribution, success rate and relationship between the steps of the trainer during the operation. This not only makes the report content richer, but also makes the presentation of data more intuitive. Finally, the generated trainer operation process report should include a summary, key findings, data analysis results and improvement suggestions. This structured report can not only help training managers quickly understand the training effect, but also provide important reference for subsequent decision-making. By converting the trainer's operation feature vector into a trainer's operation process report through the generator, complex data analysis results can be effectively converted into structured information, helping training managers to conduct in-depth analysis and effective decision-making, thereby optimizing the training process and improving the quality of training.

[0068] As described above, the operation terminal 100 for data center electrical system simulation according to the embodiment of the present application can be implemented in various terminal devices. In one example, the operation terminal 100 for data center electrical system simulation can be integrated into the terminal device as a software module and / or a hardware module. For example, the operation terminal 100 for data center electrical system simulation can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the operation terminal 100 for data center electrical system simulation can also be one of the many hardware modules of the terminal device.

[0069] In summary, the embodiment of the present application first obtains the trainer terminal operation video collected by the camera and the trainer terminal operation step text data collected by the simulation system database, and then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally generates a trainer operation process report through a generator, thereby providing personalized training plans for different trainers, enhancing the trainers' learning experience, improving their operational skills, making up for the lack of real scenes in traditional simulation teaching, and improving training efficiency.

[0070] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent substitutions or changes according to the technical solutions and novel concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An operating method for data center electrical system simulation, characterized in that: The following steps are involved: S1. Obtaining terminal operation video of trainers collected by video acquisition equipment and text data of terminal operation steps of trainers collected by simulation system database; S2. Extracting features from the training staff terminal operation video to generate a key feature vector of the training staff terminal operation attention; S3. Extracting features from the training personnel's terminal operation step text data to generate a terminal operation step text-related feature vector; S4. Based on the feature topology association method driven by feature value granularity association information wandering, the terminal operation attention key feature vector of the trainer and the terminal operation step text association feature vector are fused, specifically including: S41. Constructing a training distance topology matrix between the terminal operation attention key feature vector of the trainer and the terminal operation step text associated feature vector; S42. Constructing a training feature value granularity association matrix between the terminal operation attention key feature vector of the trainer and the terminal operation step text association feature vector; S43. Performing topological correlation modulation on the training eigenvalue granularity correlation matrix based on the training distance topological matrix to obtain a training topological modulation correlation matrix; S44. Projecting the trainer terminal operation attention key feature vector and the terminal operation step text associated feature vector to the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text associated feature vector; S45. Fusion of the visual modulation trainer terminal operation attention key feature vector and the visual modulation terminal operation step text associated feature vector to generate a trainer operation feature vector; S5. Generate a training personnel operation process report based on the training personnel operation feature vector.

2. The operating method for data center electrical system simulation according to claim 1, characterized in that: The feature extraction of the training personnel terminal operation video in step S2 includes: extracting a plurality of key frames from the video; Performing feature encoding based on a spatial attention mechanism on the multiple key frames to generate multiple trainer terminal operation attention key feature maps; Pooling and cascading the multiple trainer terminal operation attention key feature maps to generate a trainer terminal operation attention key feature vector; The feature encoding based on the spatial attention mechanism includes: Using the convolutional coding part to perform deep convolutional coding on the multiple key frames to obtain an initial convolutional feature map; Inputting the initial convolutional feature map into the spatial attention part to obtain a spatial attention map; Processing the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; The position-wise point multiplication of the spatial attention feature map and the initial convolution feature map is calculated to obtain multiple trainee terminal operation attention key feature maps.

3. The operating method for data center electrical system simulation according to claim 1, characterized in that: The feature extraction of the training personnel terminal operation step text data in step S3 includes: Dividing the operation step text data into a plurality of terminal operation step text data; Encoding the plurality of terminal operation step text data using a converter-based terminal operation step text context encoder to generate a plurality of terminal operation step text feature vectors; Cascading the plurality of terminal operation step text feature vectors into a terminal operation step text associated feature vector; The converter-based terminal operation step text context encoder includes: An embedding layer, used for mapping each operation step word in the operation step text data into a word embedding vector; The converter-based Bert model is used to perform semantic encoding on the word embedding vector based on the global context to generate a terminal operation step text feature vector.

4. The operating method for data center electrical system simulation according to claim 1, characterized in that: The step S5 of generating the training personnel operation process report includes: Analyze the operational feature vectors of the trainees to generate key findings and improvement suggestions for the operational process; Through natural language generation technology, the analysis results are converted into easy-to-understand language descriptions; Generate a structured report containing a summary, key findings, data analysis results, and recommendations for improvements.

5. An operation terminal for data center electrical system simulation comprising the simulation operation method according to any one of claims 1 to 4, characterized in that: include: The training personnel operation data acquisition module is used to acquire the training personnel terminal operation video collected by the camera and the training personnel terminal operation step text data collected by the simulation system database; A training staff operation data extraction module is used to extract the training staff terminal operation attention key feature vector and the terminal operation step text association feature vector from the training staff terminal operation video collected by the camera and the training staff terminal operation step text data collected by the simulation system database; The training staff operation process report generating module is used to generate the training staff operation process report based on the training staff terminal operation attention key feature vector and the terminal operation step text association feature vector.

6. An operation terminal for data center electrical system simulation according to claim 5, characterized in that: The training personnel operation data extraction module includes: A trainer terminal operation video feature extraction unit, used for extracting features from the trainer terminal operation video collected by the camera to obtain a key feature vector of the trainer terminal operation attention; A trainer terminal operation step text feature extraction unit is used to extract features from the trainer terminal operation step text data collected by the simulation system database to obtain a terminal operation step text associated feature vector; Wherein, the training personnel terminal operation video feature extraction unit includes: A trainer terminal operation key frame extraction subunit, used to extract key frames from the trainer terminal operation video collected by the camera to obtain a plurality of trainer terminal operation key frames; A trainer terminal operation key frame attention encoding subunit, used for passing the plurality of trainer terminal operation key frames through a trainer terminal operation feature encoder based on a spatial attention mechanism to obtain a plurality of trainer terminal operation attention key feature graphs; The trainer terminal operation attention feature aggregation subunit is used to pool and cascade the multiple trainer terminal operation attention key feature maps to obtain the trainer terminal operation attention key feature vector.

7. An operation terminal for data center electrical system simulation according to claim 6, characterized in that: The trainer terminal operates a keyframe attention encoding subunit for: Using the convolutional encoding part of the trainer terminal operation feature encoder based on the spatial attention mechanism, deep convolutional encoding is performed on the plurality of trainer terminal operation key frames to obtain an initial convolutional feature map; Inputting the initial convolutional feature map into the spatial attention part of the trainer terminal operational feature encoder based on the spatial attention mechanism to obtain a spatial attention map; Pass the spatial attention map through a Softmax activation function to obtain a spatial attention feature map; The spatial attention feature map and the initial convolution feature map are multiplied by position points to obtain the multiple trainer terminal operation attention key feature maps.

8. An operation terminal for data center electrical system simulation according to claim 6, characterized in that: The training personnel terminal operation step text feature extraction unit comprises: An operation step text division subunit is used to divide the training personnel terminal operation step text data collected by the simulation system database into a plurality of terminal operation step text data; An operation step text context encoding subunit, used for passing the plurality of terminal operation step text data through a converter-based terminal operation step text context encoder to obtain a plurality of terminal operation step text feature vectors; A terminal operation step feature cascading subunit, configured to cascade the plurality of terminal operation step text feature vectors into a terminal operation step text associated feature vector; Among them, the terminal operation step feature cascade sub-unit manages the thread pool by creating an ExecutorService instance, decomposes the cascade task of multiple terminal operation step text feature vectors into multiple subtasks and submits them to the thread pool, uses the Future object to track the execution status of each subtask, merges the results after all subtasks are completed, and finally closes the thread pool to release resources.

9. An operation terminal for data center electrical system simulation according to claim 5, characterized in that: The training personnel operation process report generation module includes: A trainer operation feature association unit is used to perform feature topology association based on feature value granularity association information walking driven on the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector to obtain the trainer operation feature vector; The operation process report generating unit is used for passing the training personnel operation feature vector through a generator to generate a training personnel operation process report.

10. An operation terminal for data center electrical system simulation according to claim 9, characterized in that: The training personnel operation feature association unit comprises: Constructing a training distance topology matrix between the terminal operation attention key feature vector of the trainer and the terminal operation step text association feature vector; Constructing a training feature value granularity association matrix between the trainer's terminal operation attention key feature vector and the terminal operation step text association feature vector; Performing topological correlation modulation on the training eigenvalue granularity correlation matrix based on the training distance topological matrix to obtain a training topological modulation correlation matrix; Taking the training topology modulation association matrix as the perspective co-projection space, projecting the trainer terminal operation attention key feature vector and the terminal operation step text association feature vector into the perspective co-projection space to obtain the perspective modulation trainer terminal operation attention key feature vector and the perspective modulation terminal operation step text association feature vector; The view angle modulation trainer terminal operation attention key feature vector and the view angle modulation terminal operation step text associated feature vector are fused to obtain the trainer operation feature vector.

Images