Method and device for constructing temperature rise digital twin model of bridge arm reactor and storage medium

Abstract

This application relates to the field of power system technology and provides a method, apparatus, and storage medium for constructing a digital twin model of temperature rise of a bridge arm reactor. The method includes: acquiring input current data on the busbar corresponding to the bridge arm reactor to be modeled in the power system; constructing a long short-term memory network model based on an attention mechanism, training and optimizing the model using a simulation dataset to obtain a thermal performance prediction model for the bridge arm reactor; inputting the input current data into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data; and simulating the temperature field of the bridge arm reactor based on the thermal performance prediction data to construct a temperature rise digital twin model corresponding to the bridge arm reactor. This method can improve the accuracy of temperature prediction for bridge arm reactors and enhance the modeling effect of the digital twin model.

Description

Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method, apparatus and storage medium for constructing a digital twin model of the temperature rise of a bridge arm reactor. Background Technology

[0002] In the field of power system technology, digital twins, as a cutting-edge technology, lie in accurately mapping actual physical systems onto digital models. This can be divided into physical modeling methods and data-driven methods. Physical modeling-based methods are a traditional approach to physical field simulation. They involve establishing mathematical models of physical entities and solving these models using numerical computation methods to obtain the states and behaviors of the physical entities. Data-driven digital twin methods, on the other hand, utilize large amounts of monitoring or simulation data to train the model, and then use the trained model to quickly solve for the target physical field.

[0003] However, regarding physical model-based digital twin methods, due to the complexity of bridge arm reactors, establishing accurate physical models often requires a large amount of computational resources and time, making it difficult to meet the needs of rapid design and optimization. Regarding data-driven digital twin methods, the physical field data and network models selected in traditional approaches are not very reasonable, resulting in poor performance in real-time simulation modeling of bridge arm reactors.

[0004] In summary, traditional digital twin technology for bridge arm reactors suffers from poor modeling performance. Summary of the Invention

[0005] This application provides a method, apparatus, and storage medium for constructing a digital twin model of the temperature rise of a bridge arm reactor, which can improve the accuracy of temperature prediction for the bridge arm reactor and enhance the modeling effect of the digital twin model.

[0006] In a first aspect, embodiments of this application provide a method for constructing a digital twin model of the temperature rise of a bridge arm reactor, comprising:

[0007] Obtain the input current data on the reactor busbar corresponding to the bridge arm reactor to be modeled in the power system;

[0008] A thermal performance prediction model for the bridge arm reactor is constructed by adjusting the long short-term memory network based on the attention mechanism.

[0009] The model is trained and optimized based on the simulation dataset;

[0010] The input current data is input into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data.

[0011] Based on the predicted thermal performance data, the temperature field of the bridge arm reactor is simulated, and a digital twin model of the temperature rise corresponding to the bridge arm reactor is constructed.

[0012] Secondly, embodiments of this application provide a device for constructing a digital twin model of the temperature rise of a bridge arm reactor, which has the function of implementing the method for constructing a digital twin model of the temperature rise of a bridge arm reactor corresponding to the first aspect described above. This function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above function, and the modules can be software and / or hardware.

[0013] In one possible design, the device includes:

[0014] The data acquisition module is used to acquire the input current data on the reactor busbar corresponding to the bridge arm reactor to be modeled in the power system;

[0015] The model building module is used to adjust the long short-term memory network based on the attention mechanism to build a thermal performance prediction model for the bridge arm reactor.

[0016] The training and optimization module is used to train and optimize the thermal performance prediction model based on the simulation dataset.

[0017] A thermal performance prediction module is used to input the input current data into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data.

[0018] The digital twin model generation module is used to simulate the temperature field of the bridge arm reactor based on the thermal performance prediction data, and construct a temperature rise digital twin model corresponding to the bridge arm reactor.

[0019] In another aspect, this application provides an apparatus for constructing a digital twin model of the temperature rise of a bridge arm reactor, which includes at least one connected processor and a memory, wherein the memory is used to store program code, and the processor is used to call the program code in the memory to execute the methods described in the above aspects.

[0020] In another aspect, embodiments of this application provide a computer storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0021] In another aspect, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described in the above aspects.

[0022] Compared to traditional digital twin methods, the technical solution of this application utilizes a long short-term memory network to capture the dynamic dependencies between data, combines an attention mechanism to extract key features, and constructs a rapid simulation model of the temperature field of the bridge arm reactor. This accurately simulates the thermal characteristics of the bridge arm reactor, avoids the waste of computing resources in physical model-based digital twin methods, improves model building efficiency, selects reasonable physical field data, and modifies the network model to improve the accuracy of temperature prediction for the bridge arm reactor and enhance the modeling effect of the digital twin model. Attached Figure Description

[0023] Figure 1 This is an application environment diagram from one embodiment;

[0024] Figure 2 This is a flowchart illustrating the method for constructing a digital twin model of the temperature rise of the bridge arm reactor in one embodiment;

[0025] Figure 3 This is a schematic diagram of the structure of a long short-term memory network in one embodiment;

[0026] Figure 4 This is a schematic diagram of the activation function in one embodiment;

[0027] Figure 5 This is a flowchart illustrating the method for constructing a digital twin model of the temperature rise of the bridge arm reactor in another embodiment;

[0028] Figure 6 This is an overall flowchart of one embodiment;

[0029] Figure 7 A simulation contour plot of the bridge arm reactor in one embodiment;

[0030] Figure 8 This is a structural block diagram of the device for constructing a digital twin model of the temperature rise of the bridge arm reactor in one embodiment;

[0031] Figure 9 This is an internal structural diagram of the device used to construct a digital twin model of the temperature rise of the bridge arm reactor in one embodiment;

[0032] Figure 10 This is an internal structural diagram of the device for constructing a digital twin model of the temperature rise of the bridge arm reactor in another embodiment. Detailed Implementation

[0033] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules in the embodiments of this application is merely a logical division; in actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. Additionally, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms, none of which are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0034] Figure 1 As shown in the application environment diagram of one embodiment, this application provides a method for constructing a digital twin model of the temperature rise of a bridge arm reactor, which can be applied to, for example... Figure 1 In the application scenario shown, terminal 102 communicates with server 104 via a network.

[0035] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers.

[0036] It should be specifically noted that the terminal 102 involved in this application embodiment can be a wired terminal or a wireless terminal. It can be a device that provides voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. The wireless terminal can communicate with one or more core networks via a wireless access network. The wireless terminal can be a mobile terminal, such as a mobile phone (or "cellular" phone) or a computer with a mobile terminal. For example, it can be a portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile device that exchanges voice and / or data with the wireless access network. Examples include personal communication service telephones, cordless phones, session initiation protocol phones, wireless local loop stations, personal digital assistants, and other devices. Wireless terminals can also be referred to as systems, subscriber units, subscriber stations, mobile stations, mobile stations, remote stations, access points, remote terminals, access terminals, user terminals, terminal equipment, user agents, user devices, or user equipment.

[0037] Figure 2 This is a flowchart illustrating the method for constructing a digital twin model of the temperature rise of the bridge arm reactor in one embodiment. The following refers to... Figure 2 This application provides a method for constructing a digital twin model of the temperature rise of a bridge arm reactor, the method comprising:

[0038] S201, Obtain the input current data on the busbar of the bridge arm reactor to be modeled in the power system.

[0039] Among them, the bridge arm reactor is a key inductive component in the converter valve of the power system, and is commonly found in scenarios such as high voltage DC transmission and flexible converter stations; the bridge arm reactor to be modeled refers to the bridge arm reactor to be modeled to simulate its working state or characteristics; the reactor bus is one of the components of the bridge arm reactor; and the input current data is an important data indicator of the reactor bus.

[0040] This embodiment is based on the modeling situation in a real scenario. The actual physical field data is selected as the input current data on the reactor bus, which improves the adaptability of the modeling.

[0041] S202, based on the attention mechanism, adjusts the long short-term memory network to construct a thermal performance prediction model for the bridge arm reactor.

[0042] Among them, the attention mechanism (AM) has the ability to automatically capture the correlation between different components within the data, thereby enhancing the model's efficiency and generalization; the long short-term memory network (LSTM) is a special type of recurrent neural network that can effectively process time series data and solve the problem of long-term dependencies.

[0043] Among them, the thermal performance prediction model of the bridge arm reactor refers to the model associated with the bridge arm reactor that has been constructed, and this model has the function of predicting thermal performance.

[0044] S203, The thermal performance prediction model is trained and optimized based on the simulation dataset.

[0045] Among them, the simulation dataset is a dataset obtained through simulation methods for model training and optimization, and the thermal performance prediction model is the model used in the inference stage after training and parameter optimization adjustment.

[0046] S204: Input the input current data into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data.

[0047] Among them, the input current data is the data to be predicted input during the inference stage, that is, the input current data on the busbar of the corresponding bridge arm reactor; the thermal performance prediction data is the predicted data, such as the temperature data of the bridge arm reactor at a certain current.

[0048] It should be noted that the thermal performance prediction model inputting the input current data in step S204 is the model that has been trained and optimized in step S203, that is, the model after training.

[0049] S205, based on the thermal performance prediction data, simulates the temperature field of the bridge arm reactor and constructs a digital twin model of the temperature rise corresponding to the bridge arm reactor.

[0050] Among them, the temperature rise digital twin model is a model that reflects the changes in the temperature field of the bridge arm reactor, and can be used for health monitoring and fault diagnosis.

[0051] Compared to traditional digital twin methods, this embodiment first acquires the input current data on the busbar corresponding to the arm reactor to be modeled in the power system. Then, it adjusts the long short-term memory network based on an attention mechanism to construct a thermal performance prediction model. Next, it trains and optimizes the thermal performance prediction model based on a simulation dataset. Then, it inputs the input current data into the trained thermal performance prediction model to obtain thermal performance prediction data. Finally, it simulates the temperature field of the arm reactor based on the thermal performance prediction data to construct a temperature rise digital twin model. This embodiment utilizes a long short-term memory network to capture the dynamic dependencies between data, combines an attention mechanism to extract key features, and constructs a rapid simulation model of the arm reactor's temperature field. This accurately simulates the thermal characteristics of the arm reactor, avoiding the waste of computational resources in physical model-based digital twin methods, improving model construction efficiency. By selecting reasonable physical field data and choosing and modifying a suitable network model, the accuracy of temperature prediction for the arm reactor is improved, thus enhancing the modeling effect of the digital twin model.

[0052] Optionally, in some embodiments of this application, the method further includes: obtaining a simulation dataset, specifically including: using the input current on the busbar of the bridge arm reactor as the driving variable for reactor simulation, obtaining the initial input data corresponding to the driving variable; selecting a preset number of current values ​​within the driving current range based on the full factorial method as standard current values ​​for constructing the simulation dataset; calculating the output results corresponding to each standard current value; and constructing the simulation dataset based on the initial input data and the output results.

[0053] Among them, the driving variable refers to the key input quantity that can actively affect the system state or output result; in the reactor temperature simulation, the input current is the driving variable, and its magnitude directly determines the degree of loss and heat generation of the reactor, thus affecting the final temperature output; the preset quantity can be set according to the simulation requirements.

[0054] For example, the process of generating and acquiring the simulation dataset can be as follows: the main driving variable of the reactor simulation is the input current on the reactor busbar. A certain number of current values ​​are selected within the driving current range using the full factorial method. Then, the output results corresponding to different current values ​​are calculated through the temperature field simulation model. The input current and the output results correspond one-to-one to form the simulation dataset of the reactor temperature field.

[0055] In this embodiment, the input current that affects the state of the reactor is taken as the core driving variable. The standard current value is scientifically selected by the full factorial method, which effectively overcomes the problem of insufficient sample representativeness in the selection of traditional datasets, improves the rationality and coverage of the simulation dataset, and ensures that the dataset can accurately reflect the influence of current on reactor output, providing a data foundation for subsequent reactor-related simulation analysis.

[0056] Optionally, in some embodiments of this application, the method further includes: preprocessing the simulated dataset to obtain a preprocessed simulation dataset, specifically including: cleaning the simulation dataset to adjust abnormal data in the simulation dataset to obtain a cleaned dataset; and normalizing the cleaned dataset based on a preset standardization method to obtain a preprocessed simulation dataset.

[0057] Adjusting outlier data in the simulation dataset refers to removing or correcting errors and outliers in the data; the default standardization method can be Z-score.

[0058] For example, the preprocessing process includes data cleaning and normalization. Specifically, data cleaning can remove or correct errors and outliers in the data to ensure the quality of the input data; while normalization uses the Z-score method to dedimensionalize the data and accelerate the convergence speed of the gradient descent algorithm.

[0059] Specifically, the standardization process of the Z-Score method is as follows.

[0060] Data features used for physics simulations often come from a variety of physical parameters, which are usually in different numerical ranges and have different data distributions. Therefore, this application will use the Z-score normalization method to preprocess the input features in order to eliminate the adverse effects of different data dimensions and distributions on the feature extraction of neural networks, while improving the performance and convergence speed of the model.

[0061] The Z-score transforms the original data linearly, resulting in a mean of 0 and a standard deviation of 1. Its formula is: In the formula, X represents the original data. and represents the mean and standard deviation of the data, respectively, and Z represents the standardized data.

[0062] In this embodiment, outliers are removed or corrected by data cleaning, which effectively overcomes the problem of poor quality of the original data and improves the reliability of the data; normalization is used to achieve dimensionless transformation, which speeds up the convergence of subsequent algorithms.

[0063] Optionally, in some embodiments of this application, the thermal performance prediction model of the bridge arm reactor is constructed by adjusting the long short-term memory network based on the attention mechanism, including: using the long short-term memory network as the main architecture, introducing the attention mechanism to adjust the long short-term memory network to obtain a preliminarily adjusted network model; setting the activation function of the preliminarily adjusted network model to a preset activation function to determine the thermal performance prediction model of the bridge arm reactor.

[0064] Figure 3 This is a schematic diagram of the structure of a Long Short-Term Memory network in one embodiment, such as... Figure 3 As shown, Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network capable of effectively processing time-series data and solving long-term dependency problems. LSTM consists of input gates, forget gates, output gates, and cell states. Through gating mechanisms, it controls the flow and storage of information, achieving both long-term and short-term memory of time-series data. In the temperature rise monitoring of the bridge arm reactor, temperature is a typical time-series data point; therefore, LSTM is very suitable for constructing this physical field simulation model.

[0065] The computation process of LSTM involves the following steps: First, the structure of the LSTM network needs to be defined, including parameters such as input dimension, output dimension, number of time steps, and hidden state dimension. For a given input sequence, it can be decomposed into multiple time steps, which are then input into the LSTM network sequentially. At each time step, the LSTM network will adjust the current input... and the hidden state of the previous moment The network calculates the outputs of the three gates, representing the control signals for input, forget, and output, respectively. Simultaneously, it updates the state of memory cell C to store and retrieve information from the previous time step. Finally, the LSTM network calculates the output for the current time step based on the current gate signals and memory cell state, and passes it to the next time step. The output expressions for each gate are as follows:

[0066]

[0067]

[0068]

[0069] in, This represents the hidden state from the previous time step. It is the output of the LSTM network after processing information in the previous time step and contains key information from the previous time step. The input data at the current time step is the feature vector input into the LSTM network. , , These are the outputs of the forget gate, input gate, and output gate, respectively. For the sigmoid function, , , and , , These represent the weights and biases corresponding to different gates. The formula for calculating memory cells is:

[0070]

[0071] Finally, the matrix is ​​multiplied element by element. To obtain entirely new long-term information and short-term information The calculation formula is as follows:

[0072]

[0073]

[0074] The attention mechanism introduced in this application can be a self-attention mechanism, which has the ability to automatically capture the correlation between different components within the data, thereby enhancing the efficiency and generalization of the model. In the scenario of predicting the thermal characteristics of the bridge arm reactor, the contribution of the current excitation of each coil to the temperature change is different. In order to effectively identify and emphasize these key input features, an attention mechanism layer is introduced, which aims to deeply explore the inherent coupling relationship between the input features.

[0075] First, the output information h of the LSTM hidden layer at time t. it Perform a nonlinear transformation to obtain its implicit representation u it Next, the attention mechanism matrix u, which is randomly initialized, is used. w with u it A dot product operation is performed, followed by softmax normalization, to obtain the weight coefficients of the LSTM hidden layer output. This process constructs the attention mechanism matrix, which can be described as:

[0076]

[0077]

[0078]

[0079] In the formula: For attention weights, For the output of the attention mechanism layer, , These are the weighting coefficients. This is the bias vector. The fully connected layer uses this output vector as input data and generates the final prediction result accordingly.

[0080] Figure 4 This is a schematic diagram of an activation function in one embodiment, such as... Figure 4 As shown, the preset activation function is the Mish activation function. The Mish activation function offers better smoothness and exhibits differentiable continuity across the entire input range. When processing data, it can better capture non-linear relationships present in the data, thereby improving the model's accuracy and generalization ability. Figure 4 As can be seen, the Mish activation function has a higher curvature near the input of 0, which can effectively avoid the problems of gradient explosion and gradient saturation during training, thereby improving the performance of the prediction model. Its expression is shown below:

[0081]

[0082] The initially adjusted network model is a neural network model obtained by incorporating an attention mechanism. Naturally, the thermal performance prediction model is a model obtained by setting an activation function based on the initially adjusted network model.

[0083] For example, in the model architecture design and training, a long short-term memory network is used as the main architecture to capture long-term dependencies in the simulation data. At the same time, in order to further improve the model's predictive ability, an attention mechanism is introduced, which enables the model to dynamically focus on important features in the input sequence, thereby improving the accuracy of prediction. In terms of activation function selection, the Mish function is used, which has smoother gradient characteristics than the traditional ReLU function, helping to avoid gradient vanishing or exploding problems during training and improving the stability and generalization ability of the model.

[0084] In this embodiment, the attention mechanism is used to enhance the ability to capture key temporal information, which effectively overcomes the problem that traditional long short-term memory networks do not pay enough attention to important features. This improves the model's sensitivity to core factors related to thermal performance. Combined with the optimization of nonlinear mapping by the preset activation function, the model's fitting effect on the change law of reactor thermal performance is improved, and the accuracy and stability of the prediction results are enhanced.

[0085] Optionally, in some embodiments of this application, training and optimizing the model based on a simulation dataset specifically includes: training the model based on the training set in the simulation dataset, performing backpropagation based on the normalized root mean square error of the training set to update the model parameters; and finely adjusting and controlling the model parameters based on an adaptive learning rate optimizer during each parameter update process.

[0086] In addition, during model training, the normalized root mean square error (NRMSE) of the test set is used to evaluate the model's loss and determine its generalization ability. If the test set loss no longer decreases after several consecutive rounds, training is terminated to prevent overfitting. Furthermore, the optimal model is saved by monitoring the changes in the test set loss during training. The above process can be collectively referred to as the training and optimization process, which aims to improve the model's performance by adjusting its parameters.

[0087] In this embodiment, by performing phased validation on the training and test sets, the overfitting problem that is easily caused by training on a single dataset is effectively overcome, and the generalization ability of the model is improved. By using the normalized root mean square error to accurately quantify the loss, and combining it with the adaptive optimizer to dynamically adjust the parameters, the pertinence and efficiency of parameter optimization are enhanced, and the model's prediction accuracy of the reactor's thermal performance is improved.

[0088] Optionally, in some embodiments of this application, the thermal performance prediction model includes a long short-term memory network layer, an attention layer, and a fully connected layer. The thermal performance prediction model is trained based on the training set in the simulation dataset. The specific data calculation process includes: preprocessing the original data in the training set to obtain new feature representations; inputting the new feature representations into the long short-term memory network layer to capture long-term dependencies and dynamic changes in the data to obtain hidden state outputs; inputting the hidden state outputs into the attention layer to calculate the attention score; performing weighted summation based on the attention score to obtain a context feature vector; inputting the context feature vector into the fully connected layer to output the predicted value; calculating the loss based on the predicted value and the true value; updating the parameters of the thermal performance prediction model through the loss backpropagation mechanism; and iteratively completing the training process of the thermal performance prediction model.

[0089] For example, using the training set X∈R n×d Model training specifically includes:

[0090] Original data X∈R n×d After preprocessing, a new feature representation X'∈R is obtained. n×d’ To filter out redundant information, X' is then input into an LSTM to capture long-term dependencies and dynamic changes in the data, yielding the hidden state output L∈R. n×l Next, L is input to the attention layer, where an attention score is calculated, and a weighted sum is obtained to obtain the context feature vector C∈R. n×c Finally, the vector C is input into the fully connected layer, and the predicted value Y is output.

[0091] In this embodiment, the long short-term memory network is used to accurately capture the long-term dependence of time-series data, overcoming the problem that traditional models are insufficient in characterizing dynamic changes. The attention mechanism is combined to highlight the weight of key features, enhance the pertinence of contextual features, thereby improving the reliability of the predicted values ​​and improving the model's fitting effect on the dynamic changes of reactor thermal performance.

[0092] Optionally, in some embodiments of this application, during the update process of each model parameter, the update process of the model parameters is finely adjusted and controlled based on the adaptive learning rate optimizer, including: during the update process of each model parameter, calculating the current gradient squared and performing an exponentially weighted average with the cumulative gradient squared to obtain a weighted average reflecting the historical magnitude of the gradient; normalizing the current gradient according to the weighted average to obtain the adaptively adjusted learning rate; and adjusting the model parameters according to the adaptively adjusted learning rate.

[0093] For example, in engineering simulation and prediction problems, the data that the model needs to process is often high-dimensional and complex. Bridge arm reactor data not only contains a large amount of feature information, but also may involve complex electromagnetic field interactions between these features. This high dimensionality and complexity pose significant challenges to model training and optimization.

[0094] Therefore, this embodiment employs a highly efficient adaptive learning rate optimization algorithm, RMSProp, whose core lies in adjusting the learning rate of each parameter by calculating the moving average of the squared gradient. This characteristic makes RMSProp perform excellently when handling parameters with different gradient scales, effectively avoiding the problems of gradient vanishing or gradient exploding. When used in conjunction with an LSTM network, RMSProp can adaptively adjust the learning rate of the weights in the LSTM, thereby improving the feature processing effect of the bridge arm reactor data.

[0095] Specifically, RMSProp stores historical gradient information by maintaining a variable of accumulated squared gradients. Each time the parameters are updated, the current squared gradient is calculated and exponentially weighted with the accumulated squared gradients to obtain a weighted average that reflects the magnitude of the historical gradients. The formula is as follows:

[0096]

[0097] in, It is the moving average of the squared gradient of the t-th iteration. is the gradient of the t-th iteration, and β is the decay rate, which takes a value between 0 and 1 and controls the weight of the moving average.

[0098] The weighted average is used to normalize the current gradient, thereby adaptively adjusting the learning rate: decreasing the step size when the gradient is large to avoid oscillations, and increasing the step size when the gradient is small to accelerate convergence. During iteration, the model calculates the gradient based on the input and hidden state, adjusts the learning rate using RMSProp, and updates the parameters until the model converges or meets the stopping condition.

[0099]

[0100] in, Here, α is the parameter for the t-th iteration, and α is the learning rate. This is an offset value used to prevent division by zero errors; it is usually taken as a very small value.

[0101] In summary, for model optimization, the RMSProp optimizer (Root Mean Square Propagation) was chosen. It adaptively adjusts the learning rate of each parameter based on historical gradient information, making the adjustment of the learning rate smoother and avoiding some problems caused by the global learning rate, thereby accelerating the convergence speed of the model.

[0102] In this embodiment, the historical gradient information is captured by exponential weighted averaging, enabling the learning rate to dynamically adapt to gradient changes, thereby improving the accuracy and efficiency of parameter adjustment. The adaptive scaling of the learning rate is achieved by gradient normalization, which enhances the adaptability of parameter optimization to different feature dimensions, improves the model convergence speed and final performance, and further enhances the stability and accuracy of prediction.

[0103] Figure 5 This is a flowchart illustrating the method for constructing a digital twin model of the temperature rise of the bridge arm reactor in another embodiment. In another embodiment, Figure 5 A method for constructing a digital twin model of the temperature rise of a bridge arm reactor is provided, including the following steps:

[0104] S501: Obtain the input current data on the busbar of the bridge arm reactor to be modeled in the power system.

[0105] S502 uses a long short-term memory network as the main architecture and introduces an attention mechanism to adjust the long short-term memory network to obtain a preliminary adjusted network model. The activation function of the preliminary adjusted network model is set to a preset activation function to determine the thermal performance prediction model of the bridge arm reactor.

[0106] S503 uses the input current on the busbar of the bridge arm reactor as the driving variable for reactor simulation and obtains the initial input data corresponding to the driving variable; based on the full factorial method, a preset number of current values ​​are selected within the driving current range as standard current values ​​for constructing the simulation dataset.

[0107] S504 calculates the output results corresponding to each standard current value; based on the initial input data and the output results, a simulation dataset is formed.

[0108] S505 performs data cleaning on the simulation dataset to adjust the abnormal data in the simulation dataset and obtain the cleaned dataset; based on the preset standardization method, the cleaned dataset is normalized to obtain the preprocessed simulation dataset.

[0109] S506 preprocesses the original data in the training set to obtain new feature representations; these new feature representations are then input into the Long Short-Term Memory (LSTM) network layer to capture long-term dependencies and dynamic changes in the data, resulting in hidden state outputs.

[0110] S507: Input the hidden state output into the attention layer to calculate the attention score; perform a weighted summation based on the attention score to obtain the context feature vector; input the context feature vector into the fully connected layer to output the predicted value.

[0111] S508 calculates the loss based on the predicted and actual values, updates the parameters of the thermal performance prediction model through the backpropagation mechanism of the loss, and iterates to complete the training process of the thermal performance prediction model.

[0112] S509: During each update of model parameters, the current squared gradient is calculated and exponentially weighted with the cumulative squared gradient to obtain a weighted average that reflects the historical magnitude of the gradient; the current gradient is normalized according to the weighted average to obtain the adaptively adjusted learning rate; and the model parameters are adjusted according to the adaptively adjusted learning rate.

[0113] S510 inputs the input current data into the thermal performance prediction model to obtain the thermal performance prediction data associated with the input current data; based on the thermal performance prediction data, it simulates the temperature field of the bridge arm reactor and constructs the temperature rise digital twin model corresponding to the bridge arm reactor.

[0114] It should be noted that the specific limitations of the above steps can be found in the above description of the specific limitations of the construction method of a digital twin model of temperature rise of bridge arm reactor, and will not be repeated here.

[0115] The following describes the research process and other technical details of the method for constructing the digital twin model of the bridge arm reactor temperature rise provided in this application, using a specific embodiment.

[0116] In traditional technologies, digital twins, as a cutting-edge technology, focus on accurately mapping actual physical systems onto digital models. This process fully utilizes multi-dimensional information such as physical models, real-time sensor data updates, and system operating history. Through the high integration of interdisciplinary and multi-physical quantities, it aims to achieve efficient and accurate physical simulation. The implementation paths of digital twins can be broadly divided into physical modeling methods and data-driven methods.

[0117] The physical model-based approach is a traditional method for physics simulation. It involves establishing a mathematical model of a physical entity and solving the model using numerical computation methods to obtain the entity's state and behavior. While the physical model-based approach offers high accuracy and reliability and can reflect the essential characteristics of physical entities, its solution efficiency is low, its real-time application is insufficient, and it is difficult to achieve the effect of digital twins.

[0118] Data-driven digital twin methods utilize large amounts of monitoring or simulation data to train models, and then use the trained models to quickly solve for the target physical field. This method can leverage sufficient monitoring data to uncover the underlying physical distribution patterns within an object. Common data-driven methods include machine learning algorithms and deep learning algorithms. Machine learning algorithms, such as support vector machines and random forests, can achieve rapid calculations of the physical state of bridge arm reactors to a certain extent, but their performance is often limited for complex nonlinear problems. Deep learning algorithms have powerful feature extraction and modeling capabilities, and can better handle complex nonlinear problems. Classic models such as feedforward neural networks, recurrent neural networks, and convolutional neural networks have been widely promoted and applied in the construction of real-time simulation models.

[0119] However, the above-mentioned technologies still have the following problems:

[0120] The digital twin approach based on physical models describes the working principle and behavior of bridge arm reactors by establishing precise physical equations. This method typically requires a deep understanding of the reactor's structure, material properties, and operating environment. However, due to the complexity of bridge arm reactors, establishing accurate physical models often requires significant computational resources and time, making it difficult to meet the needs of rapid design and optimization.

[0121] Data-driven digital twin methods also have certain drawbacks: For machine learning, feature selection and extraction capabilities are significantly limited, leading to limitations in the simulation accuracy of physical fields; deep learning requires a large amount of monitoring data, and the accuracy and reliability of the model depend on the quality and quantity of data; furthermore, complex deep learning models are prone to overfitting during training. Therefore, it is necessary to combine actual physical field data and select an appropriate neural network model for real-time simulation modeling of the bridge arm reactor.

[0122] To address the aforementioned issues, this application provides a method for constructing a digital twin model of the temperature rise of a bridge arm reactor, also known as a deep learning construction method for a digital twin model of the temperature rise of a bridge arm reactor, a deep learning rapid simulation method for a digital twin model of the temperature rise of a bridge arm reactor, or a solution for the temperature field simulation problem of a bridge arm reactor, as detailed below.

[0123] This application innovatively introduces an attention mechanism, a Mish activation function, and an adaptive learning rate adjustment method into the basic long short-term memory network. This innovative design effectively captures long-term dependencies and key information in time series data, fully fits the complex temperature distribution characteristics, and significantly improves the model accuracy and generalization ability, thereby enabling rapid solution of the temperature distribution state of the bridge arm reactor.

[0124] This application is applicable to temperature simulation engineering scenarios for bridge arm reactors. By accurately simulating and analyzing temperature distribution, it can promptly identify potential overheating risks and provide strong protection for the safe operation of bridge arm reactors. It comprehensively considers factors such as the physical parameters, mesh division, and external load conditions of bridge arm reactors, deeply mines data characteristics, realizes rapid calculation of temperature field, and thus achieves the effect of digital twin.

[0125] Figure 6 Here is an overall flowchart of one embodiment, such as Figure 6 As shown in the flowchart, an algorithm for predicting the thermal performance of bridge arm reactors by integrating LSTM and attention mechanisms is proposed. The overall framework includes data preprocessing, model building, parameter optimization, and evaluation prediction.

[0126] First, the temperature field sample data was reasonably divided and standardized. Then, a fusion model was used to perform deep training on the preprocessed sample data. At the same time, the model integrated the Mish activation function and the adaptive learning rate optimization method. Finally, the validated model was saved and integrated into the digital twin system to provide users with rapid solution results of temperature field distribution under complex electromagnetic environments and to display the results in a visual cloud map, so as to achieve the digital twin effect of the bridge arm reactor.

[0127] When building a bridge arm reactor thermal performance prediction system based on LSTM and attention mechanism, the entire construction process is divided into four key steps: dataset acquisition, data preprocessing, model architecture design and training, and model performance optimization.

[0128] like Figure 6As shown, the overall process of this application includes: (1) Data acquisition: Within the driving current range of the reactor in operation, a certain number of current values ​​are selected using the full factorial method, and the current values ​​are input into the finite element simulation model to calculate the corresponding reactor temperature result data. (2) Data processing: The source data is cleaned, invalid and abnormal data are removed, and the data is arranged in order and divided into training set, validation set and test set. (3) Model training: The training set X∈R is used. n×d (4) Calculate model loss: Input the test set into the trained model and use the normalized root mean square error (NRMSE) of the test set to evaluate the model loss. (5) Parameter optimization: Adjust the model parameters through the adaptive learning rate optimizer RMSProp until the upper limit of the number of iterations is reached, then save the final parameters to complete the training. (6) Model testing: Evaluate its performance and prediction accuracy.

[0129] Figure 7 Here is a simulation cloud diagram of the bridge arm reactor in one embodiment, combined with the following... Figure 7 Here is a specific example from this application.

[0130] The bridge arm reactor is an important component of the converter valve. It contains 6 aluminum coils, one busbar at the top and one at the bottom, 6 encapsulations, several insulating support bars, 4 aluminum support plates, 4 insulators, and 4 pairs of flanges above and below the insulators.

[0131] The model is simplified by using the 3D simulation module of the finite element software to construct a finite element simulation model consisting of the reactor and the air domain. When the reactor is working, the coil and busbar will generate losses and heat, causing the reactor temperature to rise.

[0132] The overall mesh size of this case model is 15 million. The load on the model is the operating current of the reactor, which has a normal operating range of 0~1000A. Within this range, 100 current values ​​are selected as the input of the sample dataset using the full factorial method. The finite element temperature field simulation results of the reactor corresponding to each current are used as the output of the sample dataset, and the output data file is saved in *.csv format.

[0133] For this case, a hold-out method was used to allocate the sample dataset in an 8:2 ratio (training set: test set). Z-Score normalization was applied to the training set to transform the feature values ​​to the same scale, accelerating the model's convergence. Next, the normalized bridge arm reactor data was input into the Long Short-Term Memory (LSTM) network for training. The attention layer received the LSTM output as input and calculated the attention weights for each time step. The LSTM outputs were then weighted and summed according to the weights to obtain the prediction result.

[0134] After model training was completed, the best model was selected and saved based on rigorous evaluation according to multiple key indicators. In the visualization phase, the model can perform second-level physics field data calculations based on input conditions, accurately simulating the temperature field of the bridge arm reactor; specific simulation results are as follows: Figure 7 As shown.

[0135] In this embodiment, the physical field sample data is first standardized to transform it into a standard normal distribution, thus eliminating the impact of dimensionality issues between different features on model performance. Then, LSTM is used to capture the dynamic dependencies between data, combined with the self-attention mechanism's ability to extract key features, thereby constructing a rapid simulation model of the bridge arm reactor's temperature field. A series of simulation experiments verified the effectiveness and real-time performance of this method. Furthermore, the accurate simulation of the bridge arm reactor's thermal characteristics provides strong support for the health monitoring and fault diagnosis of the bridge arm reactor.

[0136] Figures 1 to 7 Any technical feature in the embodiments corresponding to any of the above items is also applicable to the embodiments of this application. Figures 8 to 10 The corresponding implementation examples will not be repeated hereafter.

[0137] The above describes a method for constructing a digital twin model of the temperature rise of a bridge arm reactor in an embodiment of this application. The following describes the apparatus for performing the above method.

[0138] Figure 8 Here is a structural block diagram of the device for constructing a digital twin model of the temperature rise of the bridge arm reactor in one embodiment. (Refer to the following...) Figure 8 The apparatus for constructing a digital twin model of the temperature rise of the bridge arm reactor is described, and the apparatus includes:

[0139] Data acquisition module 801 is used to acquire input current data on the reactor busbar corresponding to the bridge arm reactor to be modeled in the power system;

[0140] Model building module 802 is used to adjust the long short-term memory network based on the attention mechanism to build a thermal performance prediction model for the bridge arm reactor;

[0141] The training and optimization module 803 is used to train and optimize the thermal performance prediction model based on the simulation dataset;

[0142] The thermal performance prediction module 804 is used to input the input current data into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data.

[0143] The digital twin model generation module 805 is used to simulate the temperature field of the bridge arm reactor based on the thermal performance prediction data, and to construct the temperature rise digital twin model corresponding to the bridge arm reactor.

[0144] In this embodiment of the application, based on, as follows Figure 8 The connections between the modules shown in the diagram demonstrate how the cooperation between these modules can improve the accuracy of temperature prediction for the bridge arm reactor and enhance the modeling effect of the digital twin model.

[0145] In another embodiment, a device for constructing a digital twin model of the temperature rise of a bridge arm reactor is provided. This device can be a computer device, such as a server, and its internal structure diagram can be as follows: Figure 9 As shown, the device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The device's database stores relevant data. The I / O interfaces are used for exchanging information between the processor and external devices. The device's communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements the various methods described in the above embodiments.

[0146] In yet another embodiment, a device for constructing a digital twin model of the temperature rise of a bridge arm reactor is provided. This device can be a computer device, such as a terminal, and its internal structure diagram can be as follows: Figure 10 As shown, the device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the various methods described in the above embodiments. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device.

[0147] Those skilled in the art will understand that Figure 9 and Figure 10The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the construction device for the digital twin model of the bridge arm reactor temperature rise applied thereto. Specifically, the device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, in order to achieve the functions of computer equipment such as terminals or servers.

[0148] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0149] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0150] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0151] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0152] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0153] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0154] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0155] The technical solutions provided in the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A method for constructing a digital twin model of the temperature rise of a bridge arm reactor, characterized in that, The method includes: Obtain the input current data on the reactor busbar corresponding to the bridge arm reactor to be modeled in the power system; A thermal performance prediction model for the bridge arm reactor is constructed by adjusting the long short-term memory network based on the attention mechanism. The thermal performance prediction model is trained and optimized based on the simulation dataset; The input current data is input into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data. Based on the predicted thermal performance data, the temperature field of the bridge arm reactor is simulated, and a digital twin model of the temperature rise corresponding to the bridge arm reactor is constructed.

2. The method according to claim 1, characterized in that, The method further includes: obtaining the simulation dataset, specifically including: The input current on the busbar of the bridge arm reactor is used as the driving variable for reactor simulation, and the initial input data corresponding to the driving variable is obtained. Based on the full factorial method, a preset number of current values ​​are selected within the driving current range as standard current values ​​to constitute the simulation dataset. The output results corresponding to each standard current value are calculated; The simulation dataset is formed based on the initial input data and the output results.

3. The method according to claim 1, characterized in that, The method further includes: preprocessing the simulated dataset to obtain the preprocessed simulation dataset, specifically including: The simulation dataset is cleaned to adjust the abnormal data in the simulation dataset, resulting in a cleaned dataset. Based on a preset standardization method, the cleaned dataset is normalized to obtain the preprocessed simulation dataset.

4. The method according to claim 1, characterized in that, The process of adjusting the long short-term memory network based on an attention mechanism to construct a thermal performance prediction model for the bridge arm reactor includes: Based on the Long Short-Term Memory (LSTM) network as the main architecture, the attention mechanism is introduced to adjust the LSM network, resulting in a preliminarily adjusted network model. The activation function of the initially adjusted network model is set to a preset activation function to determine the thermal performance prediction model of the bridge arm reactor.

5. The method according to claim 1, characterized in that, The training and optimization of the thermal performance prediction model based on the simulation dataset includes: The thermal performance prediction model is trained based on the training set in the simulation dataset, and backpropagation is performed based on the normalized root mean square error of the training set to update the model parameters of the thermal performance prediction model. During each update of the model parameters, the model parameters are finely adjusted and controlled based on the adaptive learning rate optimizer.

6. The method according to claim 5, characterized in that, The thermal performance prediction model includes a long short-term memory network layer, an attention layer, and a fully connected layer. The model is trained using a training set from the simulation dataset. Backpropagation is performed based on the normalized root mean square error of the training set to update the model parameters, including: The original data in the training set is preprocessed to obtain new feature representations; The new feature representation is input into the long short-term memory network layer to capture long-term dependencies and dynamic changes in the data, and to obtain the hidden state output. The hidden state output is input into the attention layer to calculate the attention score; The attention scores are weighted and summed to obtain the context feature vector. The context feature vector is input into the fully connected layer, and the predicted value is output. The loss is calculated based on the predicted and actual values, and the model parameters are updated through the backpropagation mechanism of the loss. The training process of the thermal performance prediction model is completed iteratively.

7. The method according to claim 5, characterized in that, During the update process of each model parameter, the model parameters are finely adjusted and controlled based on the adaptive learning rate optimizer, including: During each update of the model parameters, the current squared gradient is calculated and exponentially weighted with the cumulative squared gradient to obtain a weighted average that reflects the historical magnitude of the gradient. The current gradient is normalized based on the weighted average value to obtain the adaptively adjusted learning rate; The model parameters are adjusted based on the adaptively adjusted learning rate.

8. A device for constructing a digital twin model of the temperature rise of a bridge arm reactor, characterized in that, The device includes: The data acquisition module is used to acquire the input current data on the reactor busbar corresponding to the bridge arm reactor to be modeled in the power system; The model building module is used to adjust the long short-term memory network based on the attention mechanism to build a thermal performance prediction model for the bridge arm reactor. The training and optimization module is used to train and optimize the thermal performance prediction model based on the simulation dataset. A thermal performance prediction module is used to input the input current data into the thermal performance prediction model to obtain thermal performance prediction data associated with the input current data. The digital twin model generation module is used to simulate the temperature field of the bridge arm reactor based on the thermal performance prediction data, and construct a temperature rise digital twin model corresponding to the bridge arm reactor.

9. A device for constructing a digital twin model of the temperature rise of a bridge arm reactor, characterized in that, The device includes: At least one processor and memory; The memory is used to store program code, and the processor is used to call the program code stored in the memory to execute the method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 7.

Images