Transformer internal temperature time delay compensation method, system, medium and device
Patent Information
- Application Number
- CN202610660975.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-14
- Publication Date
- 2026-09-11
AI Technical Summary
[0005]针对不足，本发明提供一种基于外部测温信号的变压器内部温度时滞补偿方法、系统、介质及设备，通过结合多物理场耦合仿真和深度学习模型，捕捉外部测温信号与变压器内部温度之间的非线性时滞关系，从而实现准确的温度估算，避免了现有方法中的时滞和不准确问题
[0051] Compared with existing technologies, this invention has the following advantages: By combining deep learning with physical models, it can effectively compensate for the time lag between external temperature measurement signals and internal temperatures, solving the blind spots in existing technologies; this method does not rely on internal sensors and can accurately estimate the internal temperature of the transformer through external temperature measurement signals, reducing hardware costs and maintenance difficulties; by combining multiphysics coupling simulation with deep learning models, it can better adapt to the operating conditions of different transformers, improving the real-time performance and accuracy of temperature estimation; this method is applicable to various oil-immersed transformers and can be widely used for transformer temperature rise monitoring, over-temperature early warning, and operating status assessment.
Smart Images
Figure CN122735319A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of transformer technology, and in particular to a method, system, medium, and device for compensating for internal temperature delay in transformers based on external temperature measurement signals. Background Technology
[0002] Transformers are crucial equipment in power systems, and their operational status directly impacts the stability and security of the power grid. Transformer temperature is a key indicator of its operating status. Since internal temperature is difficult to measure directly, it is typically estimated by measuring external temperature. However, due to the heat transfer process between the internal heating and the external temperature measurement point, there is a significant time lag between the external temperature signal and the actual internal temperature. This is especially problematic under dynamic conditions such as load changes, where the external temperature cannot promptly reflect changes in the internal temperature, leading to delayed warnings or distorted status assessments.
[0003] While some temperature estimation methods based on thermal circuit models exist in the prior art, they typically ignore the time delay relationship between external temperature measurement signals and internal temperatures, or simplify nonlinear time delay effects in the calculations, resulting in inaccurate temperature estimations. Therefore, there is an urgent need for a transformer internal temperature estimation method that can effectively compensate for the time delay of external temperature measurement signals.
[0004] The information disclosed in the background section is only for enhancing the understanding of the background of this invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0005] To address the shortcomings, this invention provides a method, system, medium, and device for compensating for the time delay of transformer internal temperature based on external temperature measurement signals. By combining multiphysics field coupling simulation and deep learning models, the nonlinear time delay relationship between the external temperature measurement signal and the transformer internal temperature is captured, thereby achieving accurate temperature estimation and avoiding the time delay and inaccuracy problems in existing methods.
[0006] A method for compensating for the internal temperature delay of a transformer based on an external temperature measurement signal includes:
[0007] Step 1: Obtain the structural and material parameters of the transformer. The structural parameters include the geometric dimensions of the transformer tank, windings, and core. The material parameters include the BH curve, thermal conductivity, and specific heat capacity.
[0008] Step 2: Establish a steady-state thermal circuit model of the transformer, and determine the equivalent parameters of the thermal circuit model based on the structural and material parameters;
[0009] Step 3: Establish a simulation model to simulate the transfer of heat from the internal heat source of the transformer to the external temperature measuring point, and obtain the transfer time between the external temperature measuring signal and the internal temperature of the transformer under some operating conditions.
[0010] Step 4: Capture the nonlinear correspondence of temperature transfer time under different operating conditions through a deep learning model, and perform time delay compensation on the transformer steady-state thermal circuit model to obtain a deep learning-enhanced time delay compensation model.
[0011] Step 5: Collect the external temperature measurement signal of the transformer and input the external temperature measurement signal into the deep learning-enhanced time delay compensation model;
[0012] Step 6: Obtain the time-delay-free internal temperature distribution inside the transformer based on the deep learning-enhanced time-delay compensation model.
[0013] In the method described, step 2 includes:
[0014] Step 2.1: The heat transfer process of the transformer is simplified by using a thermal resistance-thermal capacity network model, and the temperature change process is a steady-state process.
[0015] Step 2.2: Establish a steady-state thermal circuit model based on the structural and material parameters, which includes the thermal resistance and thermal capacity between each component;
[0016] Step 2.3: The external ambient temperature is considered as a constant voltage source;
[0017] Step 2.4: The dynamic heat generation and dissipation process of the transformer forms a stable heat flow loop;
[0018] Step 2.5: Based on the thermoelectric analogy theory, the thermal circuit model is described by differential equations.
[0019] In the method described, step 3 includes:
[0020] Step 3.1: Establish a simulation model of the transformer to represent the geometry of each component of the transformer;
[0021] Step 3.2, the physics module includes an electromagnetic field module, a heat conduction module, and a fluid dynamics module to simulate the transfer of heat from the inside of the transformer to the external temperature measurement point;
[0022] Step 3.3: Define the material properties of each component in the simulation model;
[0023] Step 3.4: Define the heat source for the simulation model and set the ambient temperature as the boundary condition;
[0024] Step 3.5: Mesh the model, discretizing the continuous geometric domain into a finite number of elements;
[0025] Step 3.6: Set the initial temperature distribution and initial conditions, and run the simulation to solve the model and obtain the simulation results;
[0026] Step 3.7: Analyze the temperature change curve based on the simulation results to determine the time required from the start-up of the heat source to the stabilization of the temperature at the external temperature measuring point.
[0027] In the method described, the deep learning model is any one of LSTM, CNN, Transformer, and their variant deep learning network models.
[0028] In the method described, step 4 includes:
[0029] Step 4.1: Generate internal and external temperature transfer time data under temperature levels through simulation and generate a dataset, which is divided into a training set and a validation set.
[0030] Step 4.2, determine the structure of the neural network;
[0031] Step 4.3: Select an activation function to control the flow and transmission of dataset information in the network;
[0032] Step 4.4: Select a loss function to quantify the difference between the model's predicted values and the actual values;
[0033] Step 4.5: Set up the training loop. The outer loop controls the total number of training rounds, and the inner loop is responsible for traversing each batch in the dataset.
[0034] Step 4.6: Input the data into the model for the first round of training to obtain the prediction results. Calculate the loss value using the predicted value and the true value, update the parameters through backpropagation, and proceed to the next round of training.
[0035] Step 4.7: After each round of training, use the validation set to monitor model performance in real time to prevent overfitting;
[0036] Step 4.8: Optimize model performance by adjusting model hyperparameters to ensure it can predict temperature lag;
[0037] Step 4.9: Apply the time delay compensation output by the deep learning model to the physical model to obtain the final deep learning-enhanced time delay compensation model.
[0038] In the method described, hyperparameters include batch size, learning rate, and number of training epochs.
[0039] In the method described, the external temperature measurement signals of the transformer include the top oil temperature, the bottom oil temperature, and the ambient temperature.
[0040] A system for performing the method includes:
[0041] The parameter acquisition module is used to acquire the structural and material parameters of the transformer.
[0042] The physical modeling module is used to establish a steady-state thermal circuit model of the transformer based on the structural and material parameters, and to determine the equivalent parameters of the thermal circuit model.
[0043] The simulation module is used to build a multi-physics coupled simulation model to simulate the process of heat transfer from the internal heat source of the transformer to the external temperature measuring point, and to obtain the transfer time between the external temperature measuring signal and the internal temperature of the transformer under some operating conditions.
[0044] The model building module is used to capture the nonlinear correspondence of temperature transfer time under different operating conditions through a deep learning model, and to perform time delay compensation on the physical temperature transfer model based on the steady-state thermal circuit model to obtain a deep learning-enhanced time delay compensation model.
[0045] The data acquisition module is used to collect external temperature measurement signals from the transformer.
[0046] The temperature estimation module is used to input the external temperature measurement signal into the deep learning-enhanced time-delay compensation model to obtain the time-delay-free internal temperature distribution inside the transformer.
[0047] A computer storage medium including computer instructions that, when run on a computer, cause the computer to perform the method.
[0048] An electronic device, the electronic device comprising:
[0049] Memory, processor, and computer program stored in memory and executable on the processor, wherein;
[0050] The processor implements the method when executing the program.
[0051] Compared with existing technologies, this invention has the following advantages: By combining deep learning with physical models, it can effectively compensate for the time lag between external temperature measurement signals and internal temperatures, solving the blind spots in existing technologies; this method does not rely on internal sensors and can accurately estimate the internal temperature of the transformer through external temperature measurement signals, reducing hardware costs and maintenance difficulties; by combining multiphysics coupling simulation with deep learning models, it can better adapt to the operating conditions of different transformers, improving the real-time performance and accuracy of temperature estimation; this method is applicable to various oil-immersed transformers and can be widely used for transformer temperature rise monitoring, over-temperature early warning, and operating status assessment. Attached Figure Description
[0052] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.
[0053] In the attached diagram:
[0054] Figure 1 This is the overall flowchart of the transformer internal temperature time delay compensation method based on external temperature measurement signals in this invention;
[0055] Figure 2 This is a diagram of the transformer's steady-state thermal circuit model in this invention;
[0056] Figure 3 This is a schematic diagram of a multi-physics coupling simulation model of a transformer;
[0057] Figure 4 This is a flowchart of the time delay compensation model for deep learning enhancement in this invention;
[0058] Figure 5 This is a sample image of the simulation data in this invention;
[0059] Figure 6 This is a structural diagram of the time delay compensation model in this invention.
[0060] The present invention will be further explained below with reference to the accompanying drawings and embodiments. Detailed Implementation
[0061] Specific embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While specific embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
[0062] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.
[0063] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. The accompanying drawings do not constitute a limitation on the embodiments of the present invention.
[0064] like Figures 1 to 6 As shown, a method for compensating for the internal temperature delay of a transformer based on an external temperature measurement signal includes the following steps:
[0065] Step 1: Obtain the structural and material parameters of the transformer. The structural parameters include the geometric dimensions of the transformer tank, windings, and core. The material parameters include the BH curve, thermal conductivity, and specific heat capacity.
[0066] Step 2, establish a steady-state thermal circuit model of the transformer as follows: Figure 2 And based on the structural parameters and material parameters, the equivalent parameters of the thermal circuit model are determined;
[0067] Step 3: Establish a transformer simulation model to simulate heat transfer from the internal heat source of the transformer to the external temperature measuring point, and obtain the transfer time between the external temperature measurement signal and the internal temperature of the transformer under certain operating conditions. The COMSOL software was used to establish the transformer simulation model, which includes the transformer casing, transformer oil, core, and windings. A load was set on the transformer, and the time it took for the internal and external temperatures to reach dynamic equilibrium was observed; this is the transfer time. "Some operating conditions" refers to the time it takes for several internal temperature values (e.g., 40℃, 60℃, 80℃, etc.) to be transferred to the external environment, obtained through simulation. Only a small amount of sample data corresponding to these temperature values is needed to capture the nonlinear correspondence between temperature transfer times under different operating conditions through deep learning.
[0068] Step 4 involves capturing the nonlinear relationship between temperature transfer time under different operating conditions using a deep learning model, and then performing time-delay compensation on the transformer steady-state thermal circuit model to obtain a deep learning-enhanced time-delay compensation model. This is achieved using time-delay samples of some operating conditions (i.e., some temperature values), and a deep learning network model capable of representing the time delays of all temperature values. The time delays of each temperature value are mapped to the thermal circuit model. For example, the thermal circuit model calculates that when the load is 100%, the external measured temperature is 56℃, and the internal temperature should be 70℃. However, the time delay calculated by deep learning is 10 minutes. Therefore, when the external temperature is measured to be 56℃, we know that the transformer's internal temperature was 70℃ 10 minutes ago.
[0069] Step 5: Collect the external temperature measurement signal of the transformer and input the external temperature measurement signal into the deep learning-enhanced time delay compensation model;
[0070] Step 6: Obtain the time-delay-free internal temperature distribution inside the transformer based on the deep learning-enhanced time-delay compensation model.
[0071] In a preferred embodiment of the method, step 2 includes:
[0072] Step 2.1: The heat transfer process of the transformer is simplified by using a thermal resistance-thermal capacity network model, and the temperature change process is a steady-state process.
[0073] Step 2.2: Establish a steady-state thermal circuit model based on the structural and material parameters, which includes the thermal resistance and thermal capacity between each component;
[0074] Step 2.3: The external ambient temperature is considered as a constant voltage source;
[0075] Step 2.4: The dynamic heat generation and dissipation process of the transformer forms a stable heat flow loop;
[0076] Step 2.5: Based on the thermoelectric analogy theory, the thermal circuit model is described by differential equations.
[0077]
[0078] Where Φ is the heat flow generated by the losses in the windings and core during transformer operation, represented by a current source; Cth is the sum of the heat capacities of the windings and core; Rth is the convective heat transfer resistance from the transformer oil to the external environment, all of which are constant values in the initial model; the constant pressure source is the external environment temperature θe; and the temperature node θ is the temperature at the location to be determined.
[0079] In a preferred embodiment of the method, step 3 includes:
[0080] Step 3.1: Establish a simulation model of the transformer to represent the geometry of each component of the transformer;
[0081] Step 3.2, the physics module includes an electromagnetic field module, a heat conduction module, and a fluid dynamics module to simulate the transfer of heat from the inside of the transformer to the external temperature measurement point;
[0082] Step 3.3: Define the material properties of each component in the simulation model;
[0083] Step 3.4: Define the heat source for the simulation model and set the ambient temperature as the boundary condition;
[0084] Step 3.5: Mesh the model, discretizing the continuous geometric domain into a finite number of elements;
[0085] Step 3.6: Set the initial temperature distribution and initial conditions, and run the simulation to solve the model and obtain the simulation results;
[0086] Step 3.7: Analyze the temperature change curve based on the simulation results to determine the time required from the start-up of the heat source to the stabilization of the temperature at the external temperature measuring point.
[0087] In a preferred embodiment of the method, the deep learning model is any one of LSTM, CNN, Transformer, and variant deep learning network models based thereon.
[0088] In a preferred embodiment of the method, step 4 includes:
[0089] Step 4.1: Generate internal and external temperature transfer time data under temperature levels through simulation and generate a dataset, which is divided into a training set and a validation set.
[0090] Step 4.2, define the structure of the neural network to be used;
[0091] First, the time-delay sequence data containing temperature is divided into multiple non-overlapping subsequence blocks, each representing a local temperature window, preserving its internal structural information.
[0092] Secondly, the model employs an intra-block attention mechanism to focus on the relationships between data points within each block. Within each block, the model calculates the correlation between each temperature point and other temperature points through a cross-spatiotemporal attention mechanism. Simultaneously, the model uses an inter-block attention mechanism to handle the relationships between different temperature window blocks, utilizing self-attention to calculate the correlation weights between blocks in order to capture the correspondence between different temperature points and time delays.
[0093] Subsequently, the model integrates the local detail features extracted by intra-block attention and the global correlation features captured by inter-block attention through a feature fusion layer to form a unified and comprehensive feature representation.
[0094] Finally, the fused feature vector is input into the prediction output layer, mapped to the target prediction space, and outputs the time-delay inversion result at a specific temperature.
[0095] Step 4.3: Select an activation function to control the flow and transmission of dataset information in the network;
[0096] Step 4.4: Select a loss function to quantify the difference between the model's predicted values and the actual values;
[0097] Step 4.5: Set up the training loop. The outer loop controls the total number of training rounds, and the inner loop is responsible for traversing each batch in the dataset.
[0098] Step 4.6: Input the data into the model for the first round of training to obtain the prediction results. Calculate the loss value using the predicted value and the true value, update the parameters through backpropagation, and proceed to the next round of training.
[0099] Step 4.7: After each round of training, use the validation set to monitor model performance in real time to prevent overfitting;
[0100] Step 4.8: Optimize model performance by adjusting model hyperparameters to ensure it can predict temperature lag;
[0101] Step 4.9: Apply the time delay compensation output by the deep learning model to the physical model to obtain the final deep learning-enhanced time delay compensation model.
[0102] In a preferred embodiment of the method, the hyperparameters include batch size, learning rate, and number of training epochs.
[0103] In a preferred embodiment of the method, the external temperature measurement signals of the transformer include the top oil temperature, the bottom oil temperature, and the ambient temperature.
[0104] A system for performing the method includes:
[0105] The parameter acquisition module is used to acquire the structural and material parameters of the transformer.
[0106] The physical modeling module is used to establish a steady-state thermal circuit model of the transformer based on the structural and material parameters, and to determine the equivalent parameters of the thermal circuit model.
[0107] The simulation module is used to build a multi-physics coupled simulation model to simulate the process of heat transfer from the internal heat source of the transformer to the external temperature measuring point, and to obtain the transfer time between the external temperature measuring signal and the internal temperature of the transformer under some operating conditions.
[0108] The model building module is used to capture the nonlinear correspondence of temperature transfer time under different operating conditions through a deep learning model, and to perform time delay compensation on the physical temperature transfer model based on the steady-state thermal circuit model to obtain a deep learning-enhanced time delay compensation model.
[0109] The data acquisition module is used to collect external temperature measurement signals from the transformer.
[0110] The temperature estimation module is used to input the external temperature measurement signal into the deep learning-enhanced time-delay compensation model to obtain the time-delay-free internal temperature distribution inside the transformer.
[0111] A computer storage medium including computer instructions that, when run on a computer, cause the computer to perform the method.
[0112] An electronic device, the electronic device comprising:
[0113] Memory, processor, and computer program stored in memory and executable on the processor, wherein:
[0114] The processor implements the method when executing the program.
[0115] The transformer internal and external temperature time delay compensation method is applicable to various oil-immersed transformers.
[0116] In one embodiment, a method for compensating for the internal temperature delay of a transformer based on an external temperature measurement signal includes:
[0117] Step 1: Obtain the structural and material parameters of the transformer;
[0118] Step 2: Based on the structural and material parameters, establish a steady-state thermal circuit model of the transformer and determine the equivalent parameters of the thermal circuit model;
[0119] Step 2.1: By analogy with circuits, a thermal resistance-thermal capacity network model is used to simplify the heat transfer process of the transformer, assuming that the temperature change process is a steady-state process;
[0120] Step 2.2: Based on the obtained structural and material parameters, establish a steady-state thermal circuit model, including the thermal resistance and heat capacity between each component;
[0121] Step 2.3 In the simplified model, it is assumed that the ambient temperature has little effect on the heat exchange process outside the transformer, so the external ambient temperature is regarded as a constant voltage source.
[0122] Step 2.4: In this simplified model, the dynamic heat generation and dissipation process of the transformer forms a stable heat flow loop.
[0123] Step 2.5: According to the thermoelectric analogy theory, the thermal circuit model can be described by differential equations;
[0124] Step 3: Using a multiphysics coupling simulation model, simulate the transfer of heat from the internal heat source of the transformer to the external temperature measurement point, and obtain the transfer time between some external temperature measurement signals and the internal temperature of the transformer.
[0125] Step 3.1: Establish a simulation model of the transformer to accurately represent the geometry of each component of the transformer;
[0126] Step 3.2: In order to simulate the heat source being transferred from inside the transformer to the external temperature measuring point, select appropriate physical field modules, including electromagnetic field module, heat conduction module and fluid dynamics module;
[0127] Step 3.3: Define material properties for each component in the model;
[0128] Step 3.4: Define the heat source for the simulation model and set the ambient temperature as the boundary condition;
[0129] Step 3.5: Mesh the model, discretizing the continuous geometric domain into a finite number of elements;
[0130] Step 3.6: Set the initial temperature distribution and initial conditions, and run the simulation to solve the model;
[0131] Step 3.7: Analyze the temperature change curve based on the simulation results to determine the time required from the start-up of the heat source to the stabilization of the temperature at the external temperature measuring point;
[0132] Step 4: Capture the nonlinear correspondence of temperature transfer time under different operating conditions using a deep learning model, and perform time delay compensation on the physical temperature transfer model.
[0133] Step 4.1: Generate internal and external temperature transfer time data under different operating conditions and temperature levels through multiphysics coupling simulation;
[0134] Step 4.2: Based on the generated dataset, select appropriate activation and loss functions, and train the model.
[0135] Step 4.3: Train the deep learning model using the training set and use the validation set to monitor the model performance in real time to prevent overfitting;
[0136] Step 4.4: Optimize model performance by adjusting model hyperparameters (such as batch size, learning rate, number of training epochs, etc.) to ensure that it can accurately predict temperature lag.
[0137] Step 4.5: Apply the time delay compensation output by the deep learning model to the physical model to obtain the final deep learning-enhanced time delay compensation model.
[0138] Step 5: Collect the external temperature measurement signal of the transformer and input the external temperature measurement signal into the deep learning-enhanced time delay compensation model;
[0139] Step 6: Based on the deep learning-enhanced time delay compensation model, obtain the time-delay-free internal temperature distribution inside the transformer.
[0140] In one embodiment, step 1, the transformer steady-state thermal circuit model is as follows: Figure 1 Where Φ is the heat flow generated by the losses in the windings and core during transformer operation, represented by a current source; Cth is the sum of the heat capacities of the windings and core; Rth is the convective heat transfer resistance from the transformer oil to the external environment, all of which are constant values in the initial model; the constant pressure source is the external environment temperature θe; and the temperature node θ is the temperature at the location to be determined.
[0141] Step 2: Substitute the transformer structure and material parameters from the example to determine its equivalent parameters;
[0142] Step 3, establish the transformer simulation model of the embodiment, as follows: Figure 3 The simulation yielded time delay values corresponding to some internal temperature values, such as... Figure 5 As shown;
[0143] Step 4: Define the neural network model structure;
[0144] Step 5: Set the parameters as follows: batch size 128, number of multi-head attention heads 8, dropout rate 0.1, learning rate 0.05, block size 4, and step size 2;
[0145] Step 6: Input the sample data into the neural network to train the deep learning model so that it can capture the nonlinear relationship between temperature transfer time under different working conditions.
[0146] Step 7: Apply the time delay compensation output by the deep learning model to the steady-state thermal circuit model to obtain the final deep learning-enhanced time delay compensation model.
[0147] Step 8: Set the load of the deep learning augmented time delay compensation model to 100% and the ambient temperature to 20℃. The time delay is 0.53h, which corresponds completely with the simulation verification results, indicating that the model has excellent performance.
[0148] like Figure 2 As shown, Φ represents the heat flux generated by losses in the windings and core of the transformer during operation, expressed as a current source, C. th R is the sum of the heat capacities of the windings and the core. thThe convective heat transfer thermal resistance from transformer oil to the external environment is a constant value in the initial model, and the constant pressure source is the external ambient temperature θ. e The temperature node θ represents the temperature at the location to be determined.
[0149] See Figure 6 The time delay compensation model structure and the time delay compensation model training process include:
[0150] First, the time-delay data containing temperature is divided into multiple non-overlapping subsequence blocks, each representing a local temperature window, preserving its internal structural information.
[0151] The sample starts at a different temperature and ends at a different temperature. The i-th univariate sequence of length L is represented as follows:
[0152] ,
[0153] Each input temperature range data is divided into sub-module data, with the length of each sub-sequence block denoted as P, and the number of blocks being N.
[0154]
[0155] Secondly, the model employs an intra-block attention mechanism to focus on the relationships between data points within each block. Within each block, the model calculates the correlation between each temperature point and other temperature points through a cross-spatiotemporal attention mechanism. Simultaneously, the model uses an inter-block attention mechanism to handle the relationships between different temperature window blocks, utilizing self-attention to calculate the correlation weights between blocks in order to capture the correspondence between different temperature points and time delays.
[0156] The cross-spatial attention mechanism can be represented as:
[0157]
[0158]
[0159] In the formula, LayerNorm(•) represents layer normalization, and MLP(•) represents feedforward network. (•) indicates a multi-head attention mechanism. This is not required in detail in regular networks, as all dimensions are shared. , and express And the output of MLP.
[0160] The self-attention mechanism can be represented as:
[0161]
[0162] For the input temperature sequence T, That is, the data corresponding to the weight matrix. Let Q and K be the feature dimensions. Divide by This can make gradient descent more stable.
[0163] Subsequently, the model integrates the local detail features extracted by intra-block attention and the global correlation features captured by inter-block attention through a feature fusion layer to form a unified and comprehensive feature representation.
[0164]
[0165] in, Indicates intra-block attention features. This represents the inter-block attention feature.
[0166] Finally, the fused feature vector is input into the prediction output layer, mapped to the target prediction space, and outputs the time-delay inversion result at a specific temperature.
[0167] Furthermore, this invention first utilizes the structural and material parameters of the transformer to establish a steady-state thermal circuit model (thermal resistance-thermal capacity network). This model analogizes the complex heat transfer process to a circuit, providing a physical basis for subsequent temperature estimation. The model clarifies the physical relationships between heat sources (winding and core losses), heat transfer paths (thermal resistance, thermal capacity), and boundary conditions (ambient temperature), avoiding the uninterpretability caused by the "black box" nature of purely data-driven models. As prior knowledge, the steady-state thermal circuit model provides physical constraints for the output of the deep learning model, ensuring that the time-delay compensation results conform to the fundamental laws of energy conservation and heat transfer, thus improving the reliability and extrapolation capability of the estimation results.
[0168] Traditional methods struggle to obtain time-delay data under full operating conditions. This invention introduces a multi-physics coupled simulation model (electromagnetic-thermal-fluid field coupling) to accurately simulate the dynamic process of heat transfer from an internal heat source to an external temperature measurement point. The internal temperature of a transformer is difficult to measure directly, making it challenging to obtain the "input (external temperature) - label (internal temperature)" samples required for supervised learning. The simulation model can generate high-fidelity time-delay data in batches under different loads and ambient temperatures, providing ample training samples for deep learning models.
[0169] Revealing the nonlinear time delay mechanism: The oil flow and the thermal conductivity of the material inside the transformer change with temperature, causing the time delay to exhibit strong nonlinearity. Multiphysics simulation can accurately capture this nonlinear process, and the generated dataset contains complex time delay characteristics under the coupling of multiple physical mechanisms such as heat conduction and convection heat transfer, laying a data foundation for subsequent deep learning to capture nonlinear mappings.
[0170] This invention employs deep learning models (such as LSTM, CNN, or Transformer) to learn the nonlinear correspondence of temperature transfer time under different operating conditions, and introduces this compensation into the physical temperature transfer model. Traditional thermal circuit models typically use lumped parameters or simplified assumptions, making it difficult to accurately describe the dynamic characteristics of time delay as operating conditions change. Deep learning models, with their powerful nonlinear fitting capabilities, can automatically extract the dynamic relationships such as delay time and attenuation characteristics between external temperature measurement signals and internal real temperatures from simulation data, achieving precise quantification and compensation of time delay. While multiphysics simulations are accurate, they are computationally time-consuming and cannot be used for real-time monitoring. In contrast, the trained deep learning model has extremely fast forward inference speed, transforming the physical knowledge accumulated from simulations into a lightweight real-time compensator. Furthermore, through training data covering a wide range of operating conditions, the model can adapt to operating scenarios under different loads and ambient temperatures, exhibiting excellent generalization ability.
[0171] This invention ultimately applies the time-delay compensation output from deep learning to the physical model, forming a deep learning-enhanced time-delay compensation model. The steady-state thermal circuit model provides the main framework and physical constraints for temperature estimation, while the deep learning model is responsible for correcting the dynamic time-delay portion that it cannot accurately describe. This method relies solely on external temperature measurement signals (top oil temperature, bottom oil temperature, and ambient temperature) and, by compensating for the time-delay effect, inversely derives the internal temperature distribution. This significantly reduces the hardware cost and maintenance difficulty of pre-embedding sensors inside the transformer, while ensuring the real-time performance and accuracy of temperature monitoring under dynamic operating conditions, providing a reliable basis for over-temperature early warning and operational status assessment.
[0172] Although embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of the present invention, and all of these are within the scope of protection of the present invention.
Claims
1. A method for compensating for the internal temperature delay of a transformer based on an external temperature measurement signal, characterized in that, Includes the following steps: Step 1: Obtain the structural and material parameters of the transformer. The structural parameters include the geometric dimensions of the transformer tank, windings, and core. The material parameters include the BH curve, thermal conductivity, and specific heat capacity. Step 2: Establish a steady-state thermal circuit model of the transformer, and determine the equivalent parameters of the thermal circuit model based on the structural and material parameters; Step 3: Establish a simulation model to simulate the transfer of heat from the internal heat source of the transformer to the external temperature measuring point, and obtain the transfer time between the external temperature measuring signal and the internal temperature of the transformer under some operating conditions. Step 4: Capture the nonlinear relationship between temperature transfer time under different operating conditions using a deep learning model, and perform time delay compensation on the transformer steady-state thermal circuit model to obtain a deep learning-enhanced time delay compensation model. Step 5: Collect the external temperature measurement signal of the transformer and input the external temperature measurement signal into the deep learning-enhanced time delay compensation model; Step 6: Obtain the time-delay-free internal temperature distribution inside the transformer based on the deep learning-enhanced time-delay compensation model.
2. The method according to claim 1, characterized in that, Preferably, step 2 includes: Step 2.1: The heat transfer process of the transformer is simplified by using a thermal resistance-thermal capacity network model, and the temperature change process is a steady-state process. Step 2.2: Establish a steady-state thermal circuit model based on the structural and material parameters, which includes the thermal resistance and thermal capacity between each component; Step 2.3: The external ambient temperature is considered as a constant voltage source; Step 2.4: The dynamic heat generation and dissipation process of the transformer forms a stable heat flow loop; Step 2.5: Based on the thermoelectric analogy theory, the thermal circuit model is described by differential equations.
3. The method according to claim 1, characterized in that, Step 3 includes, Step 3.1: Establish a simulation model of the transformer to represent the geometry of each component of the transformer; Step 3.2, the physics module includes an electromagnetic field module, a heat conduction module, and a fluid dynamics module to simulate the transfer of heat from the inside of the transformer to the external temperature measurement point; Step 3.3: Define the material properties of each component in the simulation model; Step 3.4: Define the heat source for the simulation model and set the ambient temperature as the boundary condition; Step 3.5: Mesh the model, discretizing the continuous geometric domain into a finite number of elements; Step 3.6: Set the initial temperature distribution and initial conditions, and run the simulation to solve the model and obtain the simulation results; Step 3.7: Analyze the temperature change curve based on the simulation results to determine the time required from the start-up of the heat source to the stabilization of the temperature at the external temperature measuring point.
4. The method according to claim 1, characterized in that, The deep learning model is any one of LSTM, CNN, Transformer, and their variant deep learning network models.
5. The method according to claim 1, characterized in that, Step 4 includes, Step 4.1: Generate internal and external temperature transfer time data under temperature levels through simulation and generate a dataset, which is divided into a training set and a validation set. Step 4.2, determine the structure of the neural network; Step 4.3: Select an activation function to control the flow and transmission of dataset information in the network; Step 4.4: Select a loss function to quantify the difference between the model's predicted values and the actual values; Step 4.5: Set up the training loop. The outer loop controls the total number of training rounds, and the inner loop is responsible for traversing each batch in the dataset. Step 4.6: Input the data into the model for the first round of training to obtain the prediction results. Calculate the loss value using the predicted value and the true value, update the parameters through backpropagation, and proceed to the next round of training. Step 4.7: After each round of training, use the validation set to monitor model performance in real time to prevent overfitting; Step 4.8: Optimize model performance by adjusting model hyperparameters to ensure it can predict temperature lags; Step 4.9: Apply the time delay compensation output by the deep learning model to the physical model to obtain the final deep learning-enhanced time delay compensation model.
6. The method according to claim 5, characterized in that, Hyperparameters include batch size, learning rate, and number of training epochs.
7. The method according to claim 1, characterized in that, External temperature measurement signals for transformers include top oil temperature, bottom oil temperature, and ambient temperature.
8. A system for performing the method as described in any one of claims 1 to 7, characterized in that, It includes: The parameter acquisition module is used to acquire the structural and material parameters of the transformer. The physical modeling module is used to establish a steady-state thermal circuit model of the transformer based on the structural and material parameters, and to determine the equivalent parameters of the thermal circuit model. The simulation module is used to build a multi-physics coupled simulation model to simulate the process of heat transfer from the internal heat source of the transformer to the external temperature measuring point, and to obtain the transfer time between the external temperature measuring signal and the internal temperature of the transformer under some operating conditions. The model building module is used to capture the nonlinear correspondence of temperature transfer time under different operating conditions through a deep learning model, and to perform time delay compensation on the physical temperature transfer model based on the steady-state thermal circuit model to obtain a deep learning-enhanced time delay compensation model. The data acquisition module is used to collect external temperature measurement signals from the transformer. The temperature estimation module is used to input the external temperature measurement signal into the deep learning-enhanced time-delay compensation model to obtain the time-delay-free internal temperature distribution inside the transformer.
9. A computer storage medium, characterized in that, The storage medium includes computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-7.
10. An electronic device, characterized in that, The electronic device includes: Memory, processor, and computer programs stored in memory and executable on the processor, wherein, When the processor executes the program, it implements the method as described in any one of claims 1-7.