Construction method and construction device of aluminum electrolysis digital twinning practical training platform

Abstract

The invention provides a construction method and device of an aluminum electrolysis digital twinning training platform, and the method comprises the steps: obtaining data related to aluminum electrolysis through big data, obtaining first target data, training an aluminum electrolysis question and answer model according to the first target data, and obtaining a first target model; obtaining related data of the aluminum electrolysis cell through big data to obtain second target data, and constructing a digital twinborn model of aluminum electrolysis according to the second target data to obtain a second target model; and loading the second target model to a simulation platform, and loading the simulation platform and the first target model to a practical training platform to obtain the aluminum electrolysis digital twinning practical training platform. According to the method, the problems that the training process is not visual enough and the training effect is poor when an aluminum electrolysis training platform carries out training through character and practical operation driving in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method for constructing an aluminum electrolysis digital twin training platform, a construction device, a computer-readable storage medium, and an aluminum electrolysis digital twin training system. Background Art

[0002] Aluminum reduction cells are important equipment in the aluminum production process. Understanding their equipment composition and process flow is of great significance to understanding the aluminum metallurgical process. However, the inside of the electrolytic cell is a high-temperature and highly corrosive environment. In order to fully understand the internal operation of the electrolytic cell, it is often necessary to complete it through a virtual simulation training system.

[0003] Digital twin refers to the collection, modeling, simulation, visualization and analysis of various data of actual physical systems (such as geometry, material properties, motion state, temperature, pressure, etc.), and synchronization with the actual system in real time to form a technology that integrates virtual and real. Digital twin technology can achieve complete simulation and visualization of actual physical systems, allowing users to operate and control in real time through digital twin systems, so as to better understand and predict the behavior of actual systems. Due to the effectiveness of digital twin technology in guiding industrial production, the digital twin platform technology of aluminum electrolysis process is currently developing rapidly.

[0004] In recent years, with the rapid development of artificial intelligence technology, semantic models have become one of the research hotspots in the field of natural semantic processing. Among them, semantic models based on deep learning, such as recurrent neural networks (RNN) and transformer models, have strong natural semantic understanding and generation capabilities, and are widely used in machine translation, text generation, dialogue systems and other fields. In the field of aluminum electrolysis, accurate semantic models can effectively analyze and predict the electrolysis process, thereby helping to understand the principles of each part of the aluminum production process.

[0005] In the prior art, the training of relevant skills is mostly carried out by first teaching with text materials and then familiarizing with the operation through practical operations. However, teaching in text form is not intuitive enough, and the environment of aluminum electrolysis is high temperature and highly corrosive. Misoperation can easily affect the health of the trainees. Summary of the invention

[0006] The main purpose of the present application is to provide a method for constructing a digital twin training platform for aluminum electrolysis, a construction device, a computer-readable storage medium and a digital twin training system for aluminum electrolysis, so as to at least solve the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, the process is not intuitive enough and the training effect is poor.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for constructing a digital twin training platform for aluminum electrolysis is provided, comprising: obtaining first target data through big data related to aluminum electrolysis, training a question-and-answer model for aluminum electrolysis according to the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user; obtaining second target data related to an aluminum reduction cell through big data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; loading the second target model into a simulation platform, and loading the simulation platform and the first target model into a training platform to obtain a digital twin training platform for aluminum electrolysis, wherein the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user.

[0008] Optionally, training a question-answering model for aluminum electrolysis according to the first target data to obtain a first target model includes: performing data processing on the first target data to obtain third target data and dividing the third target data into a first training set, a second training set and a third training set, the data processing including one or more of data cleaning, data completion, word segmentation, data format conversion and smoothing processing; training a GPT-2 model according to the first training set using an autoregressive model training method to obtain a first model; freezing the first model, and training the first model after the freezing operation according to the second training set to obtain a second model, the freezing operation being to set the weights corresponding to the parts of the first model except the last fully connected layer to remain unchanged during the model training process; thawing the second model, and training the second model after the thawing operation according to the third training set to obtain the first target model, the thawing operation being to set the weights in the second model that are subjected to the freezing operation to be changeable during the model training process.

[0009] Optionally, performing data processing on the first target data to obtain third target data includes: analyzing the first target data to determine first data, and performing data cleaning on the first target data according to the first data to obtain fourth target data, wherein the first data is repeated or erroneous data in the first target data; analyzing the fourth target data to determine second data, and performing data completion on the second target data according to the second data to obtain fifth target data, wherein the second data is missing or blank data in the fifth target data; performing smoothing processing on the fifth target data to obtain sixth target data, performing word segmentation on the sixth target data and marking the corresponding part of speech for the word segmentation result to obtain seventh target data; and converting the seventh target data into a corresponding data format according to the data type of the seventh target data to obtain the third target data.

[0010] Optionally, the first model is frozen, and the first model after the freezing operation is trained according to the second training set to obtain a second model, including: analyzing the first model to obtain a first weight set and a second weight set, determining a third weight set according to the first weight set and the second weight set, the second weight set including the weights of the last fully connected layer of the first model, and the third weight set including non-repeated weights in the first weight set and the second weight set; adding target identification information to the weights in the third weight set, and the target identification information is used to characterize that the weights are not allowed to be changed; inputting the second training set into the first model, and calculating the first loss value through the cross entropy loss function, and determining the target gradient according to the first loss value; optimizing the first model according to the target gradient until the first loss value corresponding to the first model is less than the first preset value or the number of iterations reaches the maximum number, to obtain the second model.

[0011] Optionally, the second model is unfrozen, and the unfrozen second model is trained according to the third training set to obtain the first target model, including: deleting the target identification information corresponding to each weight in the third weight set; inputting the third training set into the second model, and obtaining a second loss value by calculating the cross entropy loss function, and updating the target gradient according to the second loss value; optimizing the second model according to the updated target gradient until the second loss value corresponding to the second model is less than a second preset value or the number of iterations reaches a maximum number, so as to obtain the first target model.

[0012] Optionally, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain the second target model includes: performing the data processing on the second target data to obtain eighth target data; and constructing a digital twin model of aluminum electrolysis according to the eighth target data to obtain the second target model.

[0013] Optionally, loading the second target model into the simulation platform includes: configuring an API interface call, a message queue and a file transfer channel between the second target model and the simulation platform to ensure that the simulation platform converts the second target model into the 3D simulation environment.

[0014] According to another aspect of the present application, a device for constructing a digital twin training platform for aluminum electrolysis is provided, the device comprising: a first acquisition unit, used to acquire information related to aluminum electrolysis through big data to obtain first target data, and train a question-and-answer model for aluminum electrolysis according to the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user; a second acquisition unit, used to acquire relevant information of an aluminum reduction cell through big data to obtain second target data, and construct a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; a processing unit, used to load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain a digital twin training platform for aluminum electrolysis, wherein the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user.

[0015] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0016] According to another aspect of the present application, a digital twin training system for aluminum electrolysis is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the described methods.

[0017] By applying the technical solution of the present application, in the method for constructing the above-mentioned digital twin training platform for aluminum electrolysis, first, the information related to aluminum electrolysis is obtained through big data to obtain the first target data, and the question-answering model of aluminum electrolysis is trained according to the above-mentioned first target data to obtain the first target model, and the above-mentioned first target model is used to answer questions in response to the question text input by the user; then, the relevant information of the aluminum electrolysis cell is obtained through big data to obtain the second target data, and the digital twin model of aluminum electrolysis is constructed according to the above-mentioned second target data to obtain the second target model, and the above-mentioned second target model is used to simulate the process of aluminum electrolysis in the aluminum electrolysis cell; finally, the above-mentioned second target model is loaded into the simulation platform, and the above-mentioned simulation platform and the above-mentioned first target model are loaded into the training platform to obtain the digital twin training platform for aluminum electrolysis, the above-mentioned simulation platform is used to convert the above-mentioned second target model into a 3D simulation environment, and the above-mentioned training platform is used to change the operating state of the above-mentioned second target model in response to the control instructions input by the user. The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A hardware structure block diagram of a mobile terminal according to a method for constructing an aluminum electrolysis digital twin training platform provided in an embodiment of the present application is shown;

[0019] Figure 2 A schematic flow chart of a method for constructing a digital twin training platform for aluminum electrolysis provided in accordance with an embodiment of the present application is shown;

[0020] Figure 3 A structural block diagram of a device for constructing an aluminum electrolysis digital twin training platform provided according to an embodiment of the present application is shown.

[0021] The above drawings include the following reference numerals:

[0022] 102, processor; 104, memory; 106, transmission device; 108, input and output devices. DETAILED DESCRIPTION

[0023] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] As introduced in the background technology, the training platform for aluminum electrolysis in the prior art is trained through text and practical driving. In order to solve the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, the process is not intuitive enough and the training effect is poor, the embodiments of the present application provide a method for constructing an aluminum electrolysis digital twin training platform, a construction device, a computer-readable storage medium and an aluminum electrolysis digital twin training system.

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0028] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal of a method for constructing a digital twin training platform for aluminum electrolysis according to an embodiment of the present invention. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0029] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the display method of device information in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The transmission device 106 is used to receive or send data via a network. The above-mentioned specific examples of the network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] In this embodiment, a method for constructing a digital twin training platform for aluminum electrolysis that runs on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0031] Figure 2 : is a flow chart of a method for constructing a digital twin training platform for aluminum electrolysis according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0032] Step S201, obtaining information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0033] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0034] Step S202, obtaining relevant information of the aluminum reduction cell through big data to obtain second target data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell;

[0035] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0036] Step S203: load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

[0037] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0038] Through this embodiment, first, the first target data is obtained by acquiring information related to aluminum electrolysis through big data, and the question-answering model of aluminum electrolysis is trained according to the above first target data to obtain the first target model, and the above first target model is used to answer questions in response to the question text input by the user; then, the second target data is obtained by acquiring relevant information of the aluminum electrolysis cell through big data, and the digital twin model of aluminum electrolysis is constructed according to the above second target data to obtain the second target model, and the above second target model is used to simulate the process of aluminum electrolysis in the aluminum electrolysis cell; finally, the above second target model is loaded into a simulation platform, and the above simulation platform and the above first target model are loaded into a training platform to obtain an aluminum electrolysis digital twin training platform, the above simulation platform is used to convert the above second target model into a 3D simulation environment, and the above training platform is used to change the operating state of the above second target model in response to the control instructions input by the user. The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor.

[0039] In order to train the first target model, in an optional implementation, the step S201 includes:

[0040] Step S2011, performing data processing on the first target data to obtain third target data and dividing the third target data into a first training set, a second training set and a third training set, wherein the data processing includes one or more of data cleaning, data completion, word segmentation, data format conversion and smoothing processing;

[0041] Specifically, firstly, a large amount of corpus data needs to be prepared, that is, the first target data mentioned above is used as a training set. These corpus data can be relevant literature, patents, papers, experimental reports, etc. in the field of aluminum electrolysis. Then the data is processed through data cleaning, data completion, acne, stem extraction and part-of-speech tagging.

[0042] Step S2012, training the GPT-2 model using an autoregressive model training method according to the first training set to obtain a first model;

[0043] Specifically, a suitable model is selected according to the task requirements. In this application, OpenAI's GPT-2 model is selected, and the autoregressive model training method is used to train the model to obtain the first model.

[0044] Step S2013, performing a freezing operation on the first model, and training the first model after the freezing operation according to the second training set to obtain a second model, wherein the freezing operation is to set the weights corresponding to the parts of the first model except the last fully connected layer to remain unchanged during the model training process;

[0045] Specifically, the fine-tuning of the model includes freezing and thawing, where, in the freezing stage, most of the weights in the model are retained and only the weights of the last fully connected layer are modified to make it better suited to the task of the aluminum electrolysis semantic model.

[0046] Step S2014, unfreeze the second model, and train the unfrozen second model according to the third training set to obtain the first target model, wherein the unfreezing operation is to set the weights in the second model that are subjected to the freezing operation so that they can be changed during the model training process.

[0047] Specifically, in the unfreezing phase, most of the weights are unfrozen, allowing the entire model to participate in training. In this phase, the entire model is fine-tuned using the aluminum electrolysis dataset to further improve its performance and obtain the first target model.

[0048] In order to obtain the third target data, in an optional implementation, the step S2011 includes:

[0049] Step S20111, analyzing the first target data to determine first data, and performing data cleaning on the first target data according to the first data to obtain fourth target data, wherein the first data is repeated or erroneous data in the first target data;

[0050] Specifically, the first target data is analyzed to determine duplicate data, erroneous data and data irrelevant to the model in the first data to obtain the first data, and the first data is deleted, that is, the first target data is cleaned to obtain the fourth target data.

[0051] Step S20112, analyzing the fourth target data to determine second data, and completing the second target data according to the second data to obtain fifth target data, wherein the second data is data that is missing or blank in the fifth target data;

[0052] Specifically, the fourth target data is analyzed to determine blank data and missing data in the fourth target data to obtain the second data, and then the second data is completed to obtain the fifth target data.

[0053] Step S20113, performing smoothing processing on the fifth target data to obtain sixth target data, performing word segmentation on the sixth target data and marking the corresponding part of speech on the word segmentation result to obtain seventh target data;

[0054] Specifically, the data is smoothed, such as normalized, or the numerical data is limited to a preset threshold to obtain the sixth target data, the sixth target data is segmented and the segmentation result is tagged with parts of speech to obtain the seventh target data.

[0055] Step S20114: converting the seventh target data into a corresponding data format according to the data type of the seventh target data to obtain the third target data.

[0056] Specifically, the data is converted into a data format that can be recognized by the model to obtain the third target data.

[0057] In order to obtain the second model, in an optional implementation, the step S2013 includes:

[0058] Step S20131, analyzing the first model to obtain a first weight set and a second weight set, and determining a third weight set based on the first weight set and the second weight set, wherein the second weight set includes the weights of the last fully connected layer of the first model, and the third weight set includes the weights in the first weight set that are not repeated in the second weight set;

[0059] Specifically, the weights in the first model are extracted to obtain the first weight set, and then the weights of the last fully connected layer in the first model are extracted to obtain the second weight set, and then the third weight set is determined based on the first weight set and the second weight set.

[0060] Step S20132, adding target identification information to the weights in the third weight set, wherein the target identification information is used to indicate that the weights are not allowed to be changed;

[0061] Specifically, flag information is added to the weights in the third weight set to ensure that the weights in the third weight set do not change during the training process, that is, a freezing operation is performed on the first model.

[0062] Step S20133, inputting the second training set into the first model, calculating the first loss value through the cross entropy loss function, and determining the target gradient according to the first loss value;

[0063] Specifically, the loss function of the first model is calculated according to the addition function L = -∑[y log(p)+(1-y)log(1-p)] to obtain the first loss value, and the gradient for updating the model is determined according to the loss function, where y is the true label and p is the predicted probability of the model.

[0064] Step S20134, optimizing the first model according to the target gradient until the first loss value corresponding to the first model is less than a first preset value or the number of iterations reaches a maximum number, thereby obtaining the second model.

[0065] Specifically, according to the formula Fine-tune the first model until the loss function is less than the first preset value to determine that the accuracy of the model meets the requirements. Or the number of iterations reaches the maximum number to determine that continuing training will have little effect on improving the model performance, and obtain the second model, where θ is the model parameter, α is the learning rate, is the gradient of the loss function with respect to the model parameters.

[0066] In order to obtain the first target model, in an optional implementation, the above step S2014 includes:

[0067] Step S20141, deleting the target identification information corresponding to each of the weights in the third weight set;

[0068] Specifically, the identification information is deleted, and the restriction on the weight of the second model is released, that is, the unfreezing operation is performed.

[0069] Step S20142, inputting the third training set into the second model, calculating a second loss value through the cross entropy loss function, and updating the target gradient according to the second loss value;

[0070] Specifically, the loss function of the second model is calculated according to the addition function L=-∑[y log(p)+(1-y)log(1-p)] to obtain the second loss value, and the gradient for updating the model is determined according to the loss function.

[0071] Step S20143, optimizing the second model according to the updated target gradient until the second loss value corresponding to the second model is less than a second preset value or the number of iterations reaches a maximum number, thereby obtaining the first target model.

[0072] Specifically, according to the formula W[l]=W[l]-α*dW[l], the second model is fine-tuned until the loss function is less than the second preset value to determine that the accuracy of the model meets the requirement. Or the number of iterations reaches the maximum number to determine that continuing training will have little effect on improving the model performance, and the first target model is obtained, α is the learning rate, dW[l] is the gradient of the loss function with respect to W[l], which is calculated by the back propagation algorithm.

[0073] In order to obtain the second target model, in an optional implementation, the step S202 includes:

[0074] Step S2021, performing the above data processing on the above second target data to obtain eighth target data;

[0075] First, collect data related to the aluminum reduction cell. These data can come from sensors, controllers and other equipment of the experimental control module, or the actual operation records of the aluminum reduction cell by the operator. These data include indicators such as temperature, current, voltage, and the operation records of the operator, i.e., the second target data, and then repeat the above data processing on the second target data to obtain the eighth target data.

[0076] Step S2022: construct a digital twin model of aluminum electrolysis based on the eighth target data to obtain the second target model.

[0077] Specifically, a detailed numerical model is constructed using these data to obtain a second target model that can accurately reflect the dynamic behavior of the real aluminum electrolysis process.

[0078] In order to obtain the above-mentioned training platform, in an optional implementation manner, the above-mentioned step S203 includes:

[0079] Step S2031, configuring the API interface call, message queue and file transfer channel between the second target model and the simulation platform to ensure that the simulation platform converts the second target model into the 3D simulation environment.

[0080] Use simulation platforms such as ANSYS to model the real aluminum electrolysis plant environment, including electrolytic cells, cables, joysticks and other equipment, as well as various physical phenomena that may occur during aluminum electrolysis, such as high temperature and strong corrosiveness. Define the user's interaction method in the virtual environment. For example, users can use VR devices to operate virtual joysticks, or control the working state of the electrolytic cell or view the internal state of the electrolytic cell by moving the mouse. Use existing advanced simulation platforms such as ANSYS for dynamic simulation to achieve real physical reactions and effects. For example, when the user operates the virtual joystick, the state inside the electrolytic cell will change according to the physical laws of aluminum electrolysis.

[0081] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0082] The embodiment of the present application also provides a device for constructing a digital twin training platform for aluminum electrolysis. It should be noted that the device for constructing a digital twin training platform for aluminum electrolysis in the embodiment of the present application can be used to execute the method for constructing a digital twin training platform for aluminum electrolysis provided in the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and those that have been explained will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0083] The following introduces the construction device of the aluminum electrolysis digital twin training platform provided in the embodiment of the present application.

[0084] Figure 3 1 is a structural block diagram of a construction device for an aluminum electrolysis digital twin training platform according to an embodiment of the present application. Figure 3 As shown, the device comprises:

[0085] A first acquisition unit 10 is used to acquire information related to aluminum electrolysis through big data to obtain first target data, and to train a question-answering model for aluminum electrolysis according to the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0086] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0087] A second acquisition unit 20 is used to acquire relevant information of the aluminum electrolysis cell through big data to obtain second target data, and to construct a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum electrolysis cell;

[0088] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0089] The processing unit 30 is used to load the above-mentioned second target model into the simulation platform, and load the above-mentioned simulation platform and the above-mentioned first target model into the training platform to obtain an aluminum electrolysis digital twin training platform. The above-mentioned simulation platform is used to convert the above-mentioned second target model into a 3D simulation environment, and the above-mentioned training platform is used to change the operating state of the above-mentioned second target model in response to the control instructions input by the user.

[0090] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0091] Through this embodiment, the first acquisition unit obtains the first target data by acquiring the information related to aluminum electrolysis through big data, and obtains the first target model by training the question-answering model of aluminum electrolysis according to the first target data, and the first target model is used to answer questions in response to the question text input by the user; the second acquisition unit obtains the second target data related to the aluminum electrolysis cell through big data, and obtains the second target model by constructing the digital twin model of aluminum electrolysis according to the second target data, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum electrolysis cell; the processing unit loads the second target model to the simulation platform, and loads the simulation platform and the first target model to the training platform to obtain the digital twin training platform for aluminum electrolysis, and the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to the control instructions input by the user. The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor.

[0092] In order to train the first target model, in an optional implementation, the first acquisition unit includes:

[0093] A first processing module is used to process the first target data to obtain third target data and divide the third target data into a first training set, a second training set and a third training set, wherein the data processing includes one or more of data cleaning, data completion, word segmentation, data format conversion and smoothing processing;

[0094] Specifically, firstly, a large amount of corpus data needs to be prepared, that is, the first target data mentioned above is used as a training set. These corpus data can be relevant literature, patents, papers, experimental reports, etc. in the field of aluminum electrolysis. Then the data is processed through data cleaning, data completion, acne, stem extraction and part-of-speech tagging.

[0095] A first training module is used to train the GPT-2 model using an autoregressive model training method according to the first training set to obtain a first model;

[0096] Specifically, a suitable model is selected according to the task requirements. In this application, OpenAI's GPT-2 model is selected, and the autoregressive model training method is used to train the model to obtain the first model.

[0097] A second training module is used to freeze the first model and train the first model after the freezing operation according to the second training set to obtain a second model, wherein the freezing operation is to set the weights corresponding to the parts of the first model except the last fully connected layer to remain unchanged during the model training process;

[0098] Specifically, the fine-tuning of the model includes freezing and thawing, where, in the freezing stage, most of the weights in the model are retained and only the weights of the last fully connected layer are modified to make it better suited to the task of the aluminum electrolysis semantic model.

[0099] The third training module is used to unfreeze the second model and train the unfrozen second model according to the third training set to obtain the first target model. The unfreezing operation is to set the weights in the second model that are subjected to the freezing operation so that they can be changed during the model training process.

[0100] Specifically, in the unfreezing phase, most of the weights are unfrozen, allowing the entire model to participate in training. In this phase, the entire model is fine-tuned using the aluminum electrolysis dataset to further improve its performance and obtain the first target model.

[0101] In order to obtain the third target data, in an optional implementation, the first processing module includes:

[0102] A first processing submodule is used to analyze the first target data to determine first data, and to perform the data cleaning on the first target data according to the first data to obtain fourth target data, wherein the first data is repeated or erroneous data in the first target data;

[0103] Specifically, the first target data is analyzed to determine duplicate data, erroneous data and data irrelevant to the model in the first data to obtain the first data, and the first data is deleted, that is, the first target data is cleaned to obtain the fourth target data.

[0104] A second processing submodule is used to analyze the fourth target data to determine second data, and to complete the second target data according to the second data to obtain fifth target data, wherein the second data is data that is missing or blank in the fifth target data;

[0105] Specifically, the fourth target data is analyzed to determine blank data and missing data in the fourth target data to obtain the second data, and then the second data is completed to obtain the fifth target data.

[0106] A third processing submodule is used to perform smoothing processing on the fifth target data to obtain sixth target data, perform word segmentation on the sixth target data and label the word segmentation result with corresponding parts of speech to obtain seventh target data;

[0107] Specifically, the data is smoothed, such as normalized, or the numerical data is limited to a preset threshold to obtain the sixth target data, the sixth target data is segmented and the segmentation result is tagged with parts of speech to obtain the seventh target data.

[0108] The fourth processing submodule is used to convert the seventh target data into a corresponding data format according to the data type of the seventh target data to obtain the third target data.

[0109] Specifically, the data is converted into a data format that can be recognized by the model to obtain the third target data.

[0110] In order to obtain the second model, in an optional implementation, the second training module includes:

[0111] A determination submodule is used to analyze the first model to obtain a first weight set and a second weight set, and determine a third weight set based on the first weight set and the second weight set, wherein the second weight set includes the weights of the last fully connected layer of the first model, and the third weight set includes the weights in the first weight set that are not repeated in the second weight set;

[0112] Specifically, the weights in the first model are extracted to obtain the first weight set, and then the weights of the last fully connected layer in the first model are extracted to obtain the second weight set, and then the third weight set is determined based on the first weight set and the second weight set.

[0113] A first training submodule is used to add target identification information to the weights in the third weight set, where the target identification information is used to indicate that the weights are not allowed to be changed;

[0114] Specifically, flag information is added to the weights in the third weight set to ensure that the weights in the third weight set do not change during the training process, that is, a freezing operation is performed on the first model.

[0115] A second training submodule is used to input the second training set into the first model, calculate the first loss value through the cross entropy loss function, and determine the target gradient according to the first loss value;

[0116] Specifically, the loss function of the first model is calculated according to the addition function L = -∑[y log(p)+(1-y)log(1-p)] to obtain the first loss value, and the gradient for updating the model is determined according to the loss function, where y is the true label and p is the predicted probability of the model.

[0117] The third training submodule is used to optimize the above-mentioned first model according to the above-mentioned target gradient until the above-mentioned first loss value corresponding to the above-mentioned first model is less than the first preset value or the number of iterations reaches the maximum number, so as to obtain the above-mentioned second model.

[0118] Specifically, according to the formula Fine-tune the first model until the loss function is less than the first preset value to determine that the accuracy of the model meets the requirements. Or the number of iterations reaches the maximum number to determine that continuing training will have little effect on improving the model performance, and obtain the second model, where θ is the model parameter, α is the learning rate, is the gradient of the loss function with respect to the model parameters.

[0119] In order to obtain the first target model, in an optional implementation, the third training module includes:

[0120] A deletion submodule, used to delete the target identification information corresponding to each of the weights in the third weight set;

[0121] Specifically, the identification information is deleted, and the restriction on the weight of the second model is released, that is, the unfreezing operation is performed.

[0122] A fourth training submodule is used to input the third training set into the second model, calculate a second loss value through the cross entropy loss function, and update the target gradient according to the second loss value;

[0123] Specifically, the loss function of the second model is calculated according to the addition function L=-∑[y log(p)+(1-y)log(1-p)] to obtain the second loss value, and the gradient for updating the model is determined according to the loss function.

[0124] The fifth training submodule is used to optimize the above-mentioned second model according to the updated above-mentioned target gradient until the above-mentioned second loss value corresponding to the above-mentioned second model is less than the second preset value or the number of iterations reaches the maximum number, so as to obtain the above-mentioned first target model.

[0125] Specifically, according to the formula W[l]=W[l]-α*dW[l], the second model is fine-tuned until the loss function is less than the second preset value to determine that the accuracy of the model meets the requirement. Or the number of iterations reaches the maximum number to determine that continuing training will have little effect on improving the model performance, and the first target model is obtained, α is the learning rate, dW[l] is the gradient of the loss function with respect to W[l], which is calculated by the back propagation algorithm.

[0126] In order to obtain the second target model, in an optional implementation manner, the second acquisition unit includes:

[0127] A second processing module, used for performing the above data processing on the above second target data to obtain eighth target data;

[0128] First, collect data related to the aluminum reduction cell. These data can come from sensors, controllers and other equipment of the experimental control module, or the actual operation records of the aluminum reduction cell by the operator. These data include indicators such as temperature, current, voltage, and the operation records of the operator, i.e., the second target data, and then repeat the above data processing on the second target data to obtain the eighth target data.

[0129] A construction module is used to construct a digital twin model of aluminum electrolysis according to the eighth target data to obtain the second target model.

[0130] Specifically, a detailed numerical model is constructed using these data to obtain a second target model that can accurately reflect the dynamic behavior of the real aluminum electrolysis process.

[0131] In order to obtain the above-mentioned training platform, in an optional implementation manner, the above-mentioned processing unit includes:

[0132] The configuration module is provided with the API interface call, message queue and file transfer channel configured between the second target model and the simulation platform to ensure that the simulation platform converts the second target model into the 3D simulation environment.

[0133] Use simulation platforms such as ANSYS to model the real aluminum electrolysis plant environment, including electrolytic cells, cables, joysticks and other equipment, as well as various physical phenomena that may occur during aluminum electrolysis, such as high temperature and strong corrosiveness. Define the user's interaction method in the virtual environment. For example, users can use VR devices to operate virtual joysticks, or control the working state of the electrolytic cell or view the internal state of the electrolytic cell by moving the mouse. Use existing advanced simulation platforms such as ANSYS for dynamic simulation to achieve real physical reactions and effects. For example, when the user operates the virtual joystick, the state inside the electrolytic cell will change according to the physical laws of aluminum electrolysis.

[0134] The construction device of the aluminum electrolysis digital twin training platform includes a processor and a memory. The first acquisition unit, the second acquisition unit and the processing unit are stored in the memory as program units, and the processor executes the program units stored in the memory to realize the corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0135] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the effect of aluminum electrolysis training can be improved by adjusting the kernel parameters.

[0136] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0137] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method for constructing the aluminum electrolysis digital twin training platform.

[0138] Specifically, the construction method of the aluminum electrolysis digital twin training platform includes:

[0139] Step S201, obtaining information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0140] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0141] Step S202, obtaining relevant information of the aluminum reduction cell through big data to obtain second target data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell;

[0142] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0143] Step S203: load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

[0144] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0145] An embodiment of the present invention provides a processor, which is used to run a program, wherein the method for constructing the aluminum electrolysis digital twin training platform is executed when the program is running.

[0146] Specifically, the construction method of the aluminum electrolysis digital twin training platform includes:

[0147] Step S201, obtaining information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0148] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0149] Step S202, obtaining relevant information of the aluminum reduction cell through big data to obtain second target data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell;

[0150] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0151] Step S203: load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

[0152] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0153] An embodiment of the present invention provides an aluminum electrolysis digital twin training system, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, at least the following steps are implemented:

[0154] Step S201, obtaining information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0155] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0156] Step S202, obtaining relevant information of the aluminum reduction cell through big data to obtain second target data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell;

[0157] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0158] Step S203: load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

[0159] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0160] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program for initializing at least the following method steps:

[0161] Step S201, obtaining information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user;

[0162] Specifically, prepare some corpora containing texts related to the field of aluminum electrolysis. The corpus can come from academic literature, technical manuals or content on some related websites in some related fields. Then, train the model on a large-scale corpus to learn some general semantic knowledge. The pre-training stage usually adopts the Autoregressive Language Model training method. Specifically, the input of the model is a text sequence, and the output is the next word in the text sequence. During the training process, the model optimizes the model parameters by maximizing the probability of predicting the next word. The advantage of this method is that it can capture the long-term dependencies of the text sequence and generate texts with coherence and rationality. Unsupervised learning is used in pre-training, and no labeled data is involved in the training. In the fine-tuning stage, the model is fine-tuned based on the loss function to ensure the accuracy of the model in question and answering.

[0163] Step S202, obtaining relevant information of the aluminum reduction cell through big data to obtain second target data, constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell;

[0164] Specifically, a digital model of the aluminum electrolysis process is created. First, the module collects various data about the aluminum electrolysis process, including operating parameters such as temperature, current, and voltage. Then, the module uses this data to build a detailed digital model that accurately reflects the dynamic behavior of the real aluminum electrolysis process. The model updates the digital model in real time to reflect changes in the actual aluminum electrolysis process. When operating parameters change, the digital twin modeling module immediately adjusts the digital model to match these changes. In this way, the digital model always remains synchronized with the actual aluminum electrolysis process.

[0165] Step S203: load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

[0166] Specifically, an immersive virtual reality environment is created for users to simulate the aluminum electrolysis process using a high-precision model provided by the digital twin modeling module. Based on simulation platforms such as ANSYS, the virtual reality simulation module first converts the digital twin model into a comprehensive, interactive three-dimensional environment. In this environment, users can directly observe and control the simulated aluminum electrolysis process, such as adjusting operating parameters, observing the impact of parameter changes on the process, etc. In addition, the present application allows users to define and modify operating parameters in the aluminum electrolysis process, such as temperature, current, type and proportion of additives in the electrolytic cell, etc., in the training platform setting. These parameters can be set through a user-friendly graphical user interface, which may include tools such as sliders and input boxes for inputting parameters, and can store and reproduce experiments. Users can save the operating parameters and experimental results they set for later analysis or reproduction of experiments. This allows users to conduct multiple experiments and compare the effects of different operating parameters on the aluminum electrolysis process to find the optimal operating conditions.

[0167] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0168] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0169] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0170] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0171] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0172] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0173] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0174] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0175] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0176] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0177] 1) The method for constructing a digital twin training platform for aluminum electrolysis of the present application comprises the following steps: first, obtaining information related to aluminum electrolysis through big data to obtain first target data, and training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user; then, obtaining relevant information of an aluminum reduction cell through big data to obtain second target data, and constructing a digital twin model of aluminum electrolysis based on the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; finally, loading the second target model onto a simulation platform, and loading the simulation platform and the first target model onto a training platform to obtain a digital twin training platform for aluminum electrolysis, wherein the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user. The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor.

[0178] 2) The device for constructing the digital twin training platform for aluminum electrolysis of the present application, the first acquisition unit acquires information related to aluminum electrolysis through big data to obtain first target data, and trains a question-and-answer model for aluminum electrolysis based on the first target data to obtain a first target model, and the first target model is used to answer questions in response to a question text input by a user; the second acquisition unit acquires relevant information of an aluminum reduction cell through big data to obtain second target data, and constructs a digital twin model of aluminum electrolysis based on the second target data to obtain a second target model, and the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; the processing unit loads the second target model into a simulation platform, and loads the simulation platform and the first target model into a training platform to obtain a digital twin training platform for aluminum electrolysis, the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user. The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor. The first acquisition unit obtains first target data by acquiring information related to aluminum electrolysis through big data, and obtains the first target model by training a question-answering model for aluminum electrolysis based on the first target data. The first target model is used to answer questions in response to a question text input by a user; the second acquisition unit obtains second target data by acquiring relevant information of an aluminum reduction cell through big data, and obtains a second target model by constructing a digital twin model of aluminum electrolysis based on the second target data. The second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; the processing unit loads the second target model to a simulation platform, and loads the simulation platform and the first target model to a training platform to obtain a digital twin training platform for aluminum electrolysis. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user.The present application is trained based on a semantic model to obtain a question-and-answer model of knowledge related to aluminum electrolysis, so as to realize real-time automatic answers to users' questions during the training process, simulate the process of aluminum electrolysis based on a digital twin model, and convert it into virtual reality through a simulation platform, so that users can intuitively observe the process of aluminum electrolysis. In addition, the training platform is equipped with a control input function for realizing human-computer interaction between the second target model and the user, so that users can change the second target model. This method solves the problem that the training platform for aluminum electrolysis in the prior art is trained through text and practical driving, and the process is not intuitive enough and the training effect is poor.

[0179] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a digital twin training platform for aluminum electrolysis, characterized in that: include: Acquiring information related to aluminum electrolysis through big data to obtain first target data, training a question-answering model for aluminum electrolysis based on the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user; Relevant information of the aluminum reduction cell is obtained through big data to obtain second target data, and a digital twin model of aluminum electrolysis is constructed according to the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum reduction cell; The second target model is loaded into a simulation platform, and the simulation platform and the first target model are loaded into a training platform to obtain an aluminum electrolysis digital twin training platform. The simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to control instructions input by a user.

2. The method according to claim 1, characterized in that: Training the aluminum electrolysis question-answering model according to the first target data to obtain a first target model includes: Performing data processing on the first target data to obtain third target data and dividing the third target data into a first training set, a second training set and a third training set, wherein the data processing includes one or more of data cleaning, data completion, word segmentation, data format conversion and smoothing processing; According to the first training set, the GPT-2 model is trained using an autoregressive model training method to obtain a first model; Freezing the first model, and training the first model after the freezing operation according to the second training set to obtain a second model, wherein the freezing operation is to set the weights corresponding to the parts of the first model except the last fully connected layer to remain unchanged during the model training process; The second model is unfrozen, and the unfrozen second model is trained according to the third training set to obtain the first target model, wherein the unfreezing operation is to set the weights of the second model that are subjected to the freezing operation so that they can be changed during the model training process.

3. The method according to claim 2, characterized in that Processing the first target data to obtain third target data includes: Analyze the first target data to determine first data, and perform data cleaning on the first target data according to the first data to obtain fourth target data, wherein the first data is repeated or erroneous data in the first target data; Analyze the fourth target data to determine second data, and complete the second target data according to the second data to obtain fifth target data, where the second data is data that is missing or blank in the fifth target data; Smoothing the fifth target data to obtain sixth target data, segmenting the sixth target data and marking corresponding parts of speech on the segmentation results to obtain seventh target data; The seventh target data is converted into a corresponding data format according to the data type of the seventh target data to obtain the third target data.

4. The method according to claim 2, characterized in that: Freezing the first model, and training the first model after the freezing operation according to the second training set to obtain a second model, comprising: Analyze the first model to obtain a first weight set and a second weight set, and determine a third weight set based on the first weight set and the second weight set, wherein the second weight set includes weights of the last fully connected layer of the first model, and the third weight set includes non-repeated weights in the first weight set and the second weight set; Adding target identification information to the weights in the third weight set, wherein the target identification information is used to indicate that the weights are not allowed to be changed; Input the second training set into the first model, calculate the first loss value through the cross entropy loss function, and determine the target gradient according to the first loss value; The first model is optimized according to the target gradient until the first loss value corresponding to the first model is less than a first preset value or the number of iterations reaches a maximum number, so as to obtain the second model.

5. The method according to claim 4, characterized in that Unfreezing the second model, and training the unfrozen second model according to the third training set to obtain the first target model, including: Deleting the target identification information corresponding to each of the weights in the third weight set; Input the third training set into the second model, calculate a second loss value through the cross entropy loss function, and update the target gradient according to the second loss value; The second model is optimized according to the updated target gradient until the second loss value corresponding to the second model is less than a second preset value or the number of iterations reaches a maximum number, so as to obtain the first target model.

6. The method according to claim 2, characterized in that Constructing a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model includes: performing the data processing on the second target data to obtain eighth target data; The second target model is obtained by constructing a digital twin model of aluminum electrolysis based on the eighth target data.

7. The method according to claim 1, characterized in that Loading the second target model into the simulation platform includes: The API interface call, message queue and file transfer channel between the second target model and the simulation platform are configured to ensure that the simulation platform converts the second target model into the 3D simulation environment.

8. A device for constructing a digital twin training platform for aluminum electrolysis, characterized in that: The device comprises: A first acquisition unit is used to acquire information related to aluminum electrolysis through big data to obtain first target data, and train a question-answering model for aluminum electrolysis according to the first target data to obtain a first target model, wherein the first target model is used to answer questions in response to a question text input by a user; A second acquisition unit acquires relevant information of the aluminum electrolysis cell through big data to obtain second target data, and constructs a digital twin model of aluminum electrolysis according to the second target data to obtain a second target model, wherein the second target model is used to simulate the process of aluminum electrolysis in the aluminum electrolysis cell; A processing unit is used to load the second target model into a simulation platform, and load the simulation platform and the first target model into a training platform to obtain an aluminum electrolysis digital twin training platform, wherein the simulation platform is used to convert the second target model into a 3D simulation environment, and the training platform is used to change the operating state of the second target model in response to a control instruction input by a user.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. An aluminum electrolysis digital twin training system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of claims 1 to 7.

Images