A power transformer temperature monitoring method, device and equipment

Abstract

The application discloses a power transformer temperature monitoring method, device and equipment, and relates to the technical field of power equipment state monitoring and intelligent modeling. First, according to the load factor, the average winding temperature and the position of the tap switch of the power transformer at the current moment, the physical model of the power transformer is used to calculate the heat generation power at the current moment; the heat generation power at the current moment is brought into a heat balance equation for solving to obtain the top-layer oil temperature simulation value at the current moment; then, the physical characteristics at the current moment and a preset number of moments before the current moment are combined to form an enhanced feature sequence; the enhanced feature sequence is input into a deviation prediction model to obtain the dynamic deviation of the top-layer oil temperature at the current moment, the top-layer oil temperature simulation value at the current moment is calibrated, and the top-layer oil temperature monitoring value is obtained. The application provides prior knowledge through the physical model, combines data-driven learning of the deviation dynamic characteristics, and applies stability constraints, so that high-precision and stable oil temperature prediction is realized.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring and intelligent modeling technology, and in particular to a method, device and equipment for monitoring the temperature of a power transformer. Background Technology

[0002] With the continuous advancement of urbanization, the demand for electricity is increasing year by year, while the number of transformers is relatively limited, leading to a continuous rise in their load rate. Especially during the high temperatures of summer and the centralized heating season in winter, the electricity load increases dramatically, and transformer overload and heavy load situations occur frequently. In the long run, this will inevitably pose a serious threat to the safe operation of the power grid. Currently, most transformers in the power grid are oil-immersed transformers. These devices mainly rely on natural oil circulation combined with plate radiators for heat dissipation, but this heat dissipation method is relatively inefficient. When the temperature rise of the insulating oil exceeds a certain limit due to sudden increases in load or other reasons, it can cause problems such as oil becoming electrified and a decrease in insulation capacity. In extreme cases, it can even cause the transformer to explode and catch fire, leading to serious safety accidents. As one of the most important and expensive pieces of equipment in the power system's generation, transformation, transmission, and consumption links, the safety and reliability of the power transformer's operation directly affects the integrity of the entire power system, and has a profound impact on the efficiency of substation companies, the interests of users, and the daily lives of the people. Statistical data shows that overheating faults account for more than 73% of all internal faults in transformers. These faults can directly or indirectly shorten the life of the equipment. Therefore, timely detection and monitoring of overheating faults, and accurate analysis of the causes and specific locations of the faults are of vital importance to ensuring the safe and stable operation of transformers.

[0003] The transformer winding hot spot temperature, as the name suggests, is the highest temperature reached by the transformer windings during operation. Typically, the location of the transformer winding hot spot is difficult to determine accurately and requires sensor measurement, which is demanding, costly, and has low accuracy, thus limiting its application in actual power grids. Given that the top oil temperature can be considered a substitute for characterizing the winding hot spot temperature, accurately calculating the top oil temperature becomes crucial for assessing the thermal condition of the transformer.

[0004] Currently, methods for calculating top-layer oil temperature mainly focus on physical model methods and data-driven methods. Semi-physical model methods, based on simplified thermal balance relationships, encompass thermal circuit model methods and load guideline methods. The most widely used methods are those recommended by IEEE Std C57.91 and GB / T1094.7-2008. GB / T1094.7-2008 assumes that the temperatures of both insulating oil and windings exhibit a linear growth trend, which can be considered as two parallel straight lines, with higher temperatures closer to the top of the transformer. However, this model rarely considers the nonlinear characteristics of the heat transfer process, leading to significant errors in the calculation results. With the continuous emergence of artificial intelligence technology, data-driven algorithms for predicting top-layer oil temperature provide a new approach to obtaining the thermal state of transformers. This method requires the establishment of evaluation indicators for the prediction results to reflect the accuracy of the model. Among these, back propagation (BP) neural networks, support vector machines, and Kalman filter models have all been used to predict top-layer oil temperature and have achieved certain results. However, while traditional neural networks can establish mapping relationships between different data to some extent, they neglect the sequential correlation of time-series data, resulting in relatively limited predictive performance. Then, recurrent neural networks (RNNs) emerged, and their unique network structure has enabled them to perform well in handling time series problems. However, for sequences with long-term dependencies, RNNs are prone to gradient vanishing or gradient exploding. Therefore, data-driven models have achieved satisfactory results in oil temperature prediction, but because most models use deep neural networks, they still suffer from problems such as subjective parameter assignment, long model training time, and the need to improve the quality of raw input data. These issues hinder the improvement of model accuracy and speed.

[0005] In summary, some progress has been made in transformer temperature monitoring. However, traditional physical model methods have large prediction biases, while data-driven methods rely on massive amounts of labeled data and have poor generalization ability when data is sparse or operating conditions change abruptly. Summary of the Invention

[0006] The purpose of this application is to provide a method, apparatus, and equipment for monitoring the temperature of power transformers, so as to improve the accuracy and generalization capability of power transformer temperature monitoring.

[0007] To achieve the above objectives, this application provides the following solution.

[0008] In a first aspect, this application provides a method for monitoring the temperature of a power transformer, including:

[0009] Based on the current load factor, average winding temperature, and tap changer position of the power transformer, the heat generation power at the current moment is calculated using the physical model of the power transformer.

[0010] The heat generation power at the current moment is substituted into the heat balance equation of the power transformer to solve for the simulated value of the top oil temperature at the current moment.

[0011] The physical characteristics of the current moment and a preset number of moments prior to the current moment are combined to form an enhanced feature sequence; the physical characteristics include ambient temperature, load factor, and simulated top oil temperature.

[0012] The enhanced feature sequence is input into the deviation prediction model to obtain the dynamic deviation of the top oil temperature at the current moment; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network.

[0013] The current top oil temperature dynamic deviation is used to calibrate the current top oil temperature simulation value, and the calibrated top oil temperature simulation value is used as the top oil temperature monitoring value.

[0014] Secondly, this application provides a power transformer temperature monitoring device, which applies the above-mentioned power transformer temperature monitoring method, and the power transformer temperature monitoring device includes:

[0015] The heat generation power calculation module is used to calculate the heat generation power of the power transformer at the current moment based on the load factor, average winding temperature and tap changer position of the power transformer at the current moment, using the physical model of the power transformer.

[0016] The top oil temperature simulation module is used to input the current heat generation power into the heat balance equation of the power transformer to solve for the current simulated value of the top oil temperature.

[0017] An enhanced feature sequence construction module is used to compose an enhanced feature sequence from the physical features of the current time and a preset number of time steps prior to the current time; the physical features include ambient temperature, load factor, and simulated top oil temperature.

[0018] The top-layer oil temperature dynamic deviation prediction module is used to input the enhanced feature sequence into the deviation prediction model to obtain the current top-layer oil temperature dynamic deviation; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network.

[0019] The calibration module is used to calibrate the simulated value of the top oil temperature at the current moment using the dynamic deviation of the top oil temperature at the current moment, and obtain the calibrated simulated value of the top oil temperature as the monitoring value of the top oil temperature.

[0020] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described power transformer temperature monitoring method.

[0021] According to the specific embodiments provided in this application, this application has the following technical effects.

[0022] This application provides a method, apparatus, and equipment for monitoring the temperature of a power transformer. First, based on the load factor, average winding temperature, and tap changer position of the power transformer at the current moment, the application uses a physical model of the power transformer to calculate the heat generation power at the current moment. The heat generation power at the current moment is then substituted into the heat balance equation of the power transformer to obtain the simulated value of the top oil temperature at the current moment. Next, physical characteristics from the current moment and a preset number of moments prior to the current moment are combined to form an enhanced feature sequence. This enhanced feature sequence is input into a deviation prediction model to obtain the dynamic deviation of the top oil temperature at the current moment. The simulated value of the top oil temperature at the current moment is calibrated using the dynamic deviation of the top oil temperature at the current moment to obtain the monitored value of the top oil temperature. This application provides prior knowledge through the heat balance equation, combines data-driven learning of the dynamic characteristics of the deviation, and applies stability constraints to achieve high-precision and stable oil temperature prediction. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a method for monitoring the temperature of a power transformer, as provided in an embodiment of this application.

[0025] Figure 2 Provided for an embodiment of this application Network structure diagram.

[0026] Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] This application provides a method, apparatus, and equipment for monitoring the temperature of power transformers, and a data-physical fusion error calibration scheme for transformer temperature monitoring. It provides prior knowledge through a physical model, combines data-driven learning of the dynamic characteristics of deviations, and applies stability constraints to achieve high-precision and stable oil temperature prediction. It is suitable for online temperature monitoring, condition assessment, and fault early warning scenarios of power transformers.

[0030] In one exemplary embodiment, a method for monitoring the temperature of a power transformer is provided, such as... Figure 1 As shown, it includes the following steps 101-105.

[0031] Step 101: Based on the load factor, average winding temperature, and tap changer position of the power transformer at the current moment, calculate the heat generation power at the current moment using the physical model of the power transformer.

[0032] Step 102: Substitute the current heat generation power into the heat balance equation of the power transformer to solve for the current simulated value of the top oil temperature.

[0033] Step 103: Combine the physical features of the current time and a preset number of time points prior to the current time into an enhanced feature sequence; the physical features include ambient temperature, load factor, and simulated top oil temperature.

[0034] Step 104: Input the enhanced feature sequence into the deviation prediction model to obtain the dynamic deviation of the top oil temperature at the current moment; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network.

[0035] Step 105: Use the dynamic deviation of the top oil temperature at the current moment to calibrate the simulated value of the top oil temperature at the current moment, and obtain the calibrated simulated value of the top oil temperature as the monitoring value of the top oil temperature.

[0036] In another exemplary embodiment, in steps 101 and 102 above, the ambient temperature is first collected by a temperature sensor, the load current is obtained by a load monitoring device and the load factor is calculated in combination with the rated current, which is used as input to the physical model; then, a model is constructed based on the principle of thermodynamics, and the heat generation power is obtained by calculating iron loss, copper loss and additional loss through the load factor. The heat dissipation characteristics are described by thermal resistance and heat capacity, and the dynamic relationship between the top oil temperature and the ambient temperature and load factor is established; finally, the heat balance equation is solved and the simulated value of the top oil temperature is output.

[0037] In another exemplary embodiment, the upper heat balance equation is constructed based on a physical model of a power transformer, the input parameters of which include Ambient temperature, Transformer load factor Based on the physical model, including the tap changer position, iron loss, copper loss, and additional losses, the heating power can be calculated. The constructed heat balance equation is shown in equation (1).

[0038] (1)

[0039] in, This refers to the heating power of the power transformer. It is the load factor; It is the average winding temperature. This is the position of the tap changer. and These are the heat capacity and thermal resistance of the equivalent thermal circuit of the power transformer, respectively. It is the temperature difference between the simulated top oil temperature and the ambient temperature; This is a simulated value for the top oil temperature; It is the ambient temperature; It is the difference between the average winding temperature and the simulated value of the top oil temperature, k is the ambient temperature transformation coefficient, and t is the time variable.

[0040] Furthermore, in the above formula Representing iron losses, copper losses, and additional losses, the power of the heating oil consists of two parts: losses unrelated to the transformer load (iron losses, assumed to be constant) and losses related to the transformer load (copper losses and additional losses). The latter depends on the average winding temperature and load factor, as follows: ,in ,and ( This is a correction factor related to the material resistivity. This refers to the total Joule loss of the transformer under rated load. The rated copper loss of the transformer at the reference temperature. (This refers to the additional losses of the transformer under rated load). The above formula is derived from... By integrating the coupled effects of load factor, winding temperature, and tap changer position on losses, rather than a single load function, it more closely approximates the actual heating mechanism, and also leverages... and It dynamically depicts the relationship between thermal resistance, heat capacity, and temperature, overcoming the shortcomings of traditional constant values ​​and improving prediction accuracy. Furthermore, it introduces… By capturing the coupling between the rate of change of ambient temperature and the temperature difference between the winding and the oil, the linkage between the environment and internal thermal characteristics is realized, making it more adaptable to complex scenarios.

[0041] Solving the above heat balance equation requires calibration using existing experimental data to ensure that the model outputs the predicted top oil temperature. To get as close as possible to the measured value, the key is to determine... and By using two physical models, one is a linear model, where... and The assumption is that it is a constant; another is a nonlinear physical model, in which... and The temperature is related, meaning its value changes dynamically with the top oil temperature. The two models are applicable to different scenarios, which has different impacts on the accuracy of the top oil temperature prediction value output by the physical model.

[0042] In another exemplary embodiment, the heat capacity and thermal resistance of the equivalent thermal circuit of the power transformer are the core parameters of the thermal balance equation; these core parameters are obtained in the following manner:

[0043] Based on the experimentally obtained measured value of the top oil temperature, the core parameters in the heat balance equation are adjusted so that the absolute value of the difference between the simulated value of the top oil temperature obtained by solving the heat balance equation and the measured value of the top oil temperature is less than a preset threshold.

[0044] In the process of obtaining the core parameters, the core parameters are assumed to be constants or dynamic changes related to the top oil temperature.

[0045] In another exemplary embodiment, in steps 103 and 104 above, the machine learning box model includes: (1) an LSTM layer: used to process the temporal dependencies of ambient temperature, load factor and physical model prediction values, and output a feature vector containing historical information, wherein the historical information contains at least the data of the previous 4 time steps; (2) a ResNet layer: based on the output features of LSTM, learns the incremental changes of the bias, and realizes dynamic bias prediction through residual connections.

[0046] The process of training a machine learning model that integrates a residual network and a long short-term memory network to obtain a deviation prediction model involves collecting measured values ​​of the top oil temperature and calculating the deviation between the measured and simulated values ​​as the learning objective of the dynamic deviation prediction model. A machine learning architecture integrating a residual network (ResNet) and a long short-term memory network (LSTM) is used as input. The simulated top oil temperature, ambient temperature, load factor, and historical time-series data (containing records from at least the first four time steps) of these three core features are used to learn the dynamic characteristics of deviation evolution over time, resulting in the deviation prediction model. Specifically, this includes:

[0047] The measured value of the top oil temperature was obtained by a temperature sensor installed on the top of the transformer. Simulated top oil temperature obtained by solving the heat balance equation The deviation is calculated as shown in equation (2).

[0048] (2)

[0049] Furthermore, an architecture integrating Long Short-Term Memory (LSTM) and ResNet is employed to learn the dynamic deviation of the transformer top oil temperature. The dynamic change pattern over time can be obtained through two different methods: linear residual networks, which are suitable for scenarios with gradual deviation changes and stable operating conditions, and whose dynamic evolution equation is as follows:

[0050] (3)

[0051] in, It represents the rate of change of the dynamic deviation of the top oil temperature over time, reflecting the evolution trend of the deviation over time; For relying solely on enhanced feature sequences The linear constraint term function is used to apply a linear stability constraint of fixed strength to the deviation to prevent the deviation from diverging. For the dynamic deviation of the top oil temperature; For relying solely on enhanced feature sequences The bias dynamic driving term function is used to fit the active driving effect of the enhanced feature set on bias changes.

[0052] In addition, there are nonlinear residual networks, which are suitable for scenarios with drastic deviation fluctuations and complex operating conditions. Their dynamic evolution equations are as follows:

[0053] (4)

[0054] in, For simultaneous dependence on enhanced feature sequences and The nonlinear adaptive constraint term function can dynamically adjust the constraint strength according to the magnitude of the deviation (the larger the deviation, the stronger the constraint strength), adapting to the nonlinear deviation evolution characteristics under sudden changes in operating conditions; other parameters , The definition is consistent with the linear residual network equation.

[0055] In the above two formulas, This represents an enhanced feature sequence, which uses physical feature X (including ambient temperature) as the basis for the enhancement. Transformer load factor Based on this, the simulated top oil temperature obtained by solving the heat balance equation was superimposed. It also includes time-series data from the current moment and the previous four time steps, which is used by the LSTM layer to capture the temporal dependencies of features, providing comprehensive support for the dynamic laws of deviation.

[0056] For linear residual networks, the core function is used to describe them. and Two LSTM architectures are presented for the design. Both architectures consider the extended features involved in the extended feature sets of the current time step and the previous four time steps. The specific LSTM architectures are described in Tables 1 and 2, respectively. Both architectures enhance the feature sequence. Based on this, it not only considers the extended features at the current time step, but also covers the extended features of the previous four time steps, including the enhanced feature sequence. It includes physical characteristics such as ambient temperature and load factor, as well as the simulated top oil temperature obtained by solving the heat balance equation. This ensures that the LSTM structure can fully capture the correlation logic between feature changes and the dynamic evolution of deviations in the time dimension, providing support for the linearized residual network to accurately learn the dynamic characteristics of deviations over time.

[0057] Therefore, the linearized correction dynamic model can be expressed as:

[0058] (5)

[0059] in, Let n be the dynamic deviation of the top oil temperature at time n. The dynamic deviation of the top oil temperature at time n-1. For time intervals, For relying solely on enhanced feature sequences The linear constraint term function, For relying solely on enhanced feature sequences The deviation dynamic driving term function, and All were obtained based on the construction of Long Short-Term Memory (LSTM) networks. , , , , The ambient temperatures at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The load factors at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The simulated values ​​of the top oil temperature at times n-4, n-3, n-2, n-1, and n are respectively.

[0060] Formula (5) above is used to describe the function and The neural network architectures are all based on a combination of Long Short-Term Memory (LSTM) layers and deep dense neural network layers, as described in Tables 1 and 2 for Linearized Residual Networks (ResNet). These architectures were built using the Tensorflow Keras library. The inputs involved in Equation (5) are as follows: Figure 2 As shown.

[0061] like Figure 2 As shown, the key coefficient function of the bias differential equation modeled based on the Long Short-Term Memory (LSTM) network architecture in this application is... This enables time-series learning of the dynamic evolution characteristics of top-layer oil temperature deviation; each LSTM unit is input with ambient temperature, loading factor, and simulated top-layer oil temperature values ​​for 5 consecutive time steps (n-4 to n), combined with the initial cell state. (Storage long-term hot property dependence) and hidden state (Transmitting short-term deviation fluctuations), the intermediate variable is output by nonlinear encoding of features through a gating mechanism. Encoding, i.e., the dynamic correlation between "environment-load-oil temperature" and deviation at this time step; multi-time step Learning is achieved through collaborative integration Dependency in characterization Time-series characteristics The changing patterns of these patterns support subsequent stability constraints, breaking the steady-state assumption, learning multi-scale correlations, and possessing scalability (reusable modeling). This provides key support for the data-physical fusion calibration of oil temperature monitoring, taking into account both mechanism constraints and error compensation.

[0062] Table 1. LSTM-based Building blocks of alternative models

[0063]

[0064] Table 2. LSTM-based Building blocks of alternative models

[0065]

[0066] The training of the above model used the first 80% of the available measurement data, while the testing used the remaining 20%.

[0067] After completing the construction of the bias correction model for the linear residual network as described above, it is important to note that the model has already learned the bias correction parameters through training. and They will be used together to perform integral operations on the deviation time evolution dynamics problem of the linear residual network (i.e., substituting into the core formula (5) of the deviation evolution of the linear residual network); specifically, this integral operation is essentially a time-series recursive process: the dynamic deviation of the top oil temperature at time n It is based on the deviation at time n-1. The result is calculated, and then it is used to calculate the next time step. The deviation, and so on.

[0068] To distinguish between known biases and dynamic characteristics based on learning and The deviation calculated by the integrator (i.e., the deviation predicted by the deviation prediction model), which is denoted by a cap symbol, is:

[0069] (6)

[0070] in, , All of these are deviations predicted by the deviation prediction model.

[0071] The deviation correction amount calculated by the linear residual network can effectively optimize the model, demonstrating excellent accuracy and stability.

[0072] The linearized correction dynamic model exhibits good performance, proving that most of the nonlinear characteristics of the problem have been captured by the first-order simplified model. For the nonlinear part of the correction work, the dynamic form of the model is similar to that of equation (5), and the deviation prediction model is:

[0073] (7)

[0074] in, Let n be the dynamic deviation of the top oil temperature at time n. The dynamic deviation of the top oil temperature at time n-1. For time intervals, For simultaneous dependence on enhanced feature sequences Dynamic deviation of top oil temperature at time n-1 The nonlinear adaptive constraint term function, For relying solely on enhanced feature sequences The deviation dynamic driving term function, and All were obtained based on the construction of Long Short-Term Memory (LSTM) networks. , , , , The ambient temperatures at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The load factors at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The simulated values ​​of the top oil temperature at times n-4, n-3, n-2, n-1, and n are respectively.

[0075] The integral form of the nonlinear deviation prediction model corresponding to the above formula (6) is:

[0076] (8)

[0077] in, , All of these are deviations predicted by the deviation prediction model.

[0078] In another exemplary embodiment, the process of solving the heat balance equation to obtain the simulated top-layer oil temperature, using an oil temperature prediction model to predict the dynamic deviation of the top-layer oil temperature, and then calibrating the simulated top-layer oil temperature is referred to as a hybrid model. To illustrate the performance of this hybrid model, a reference model (pure data-driven model) is constructed using the exact same learning principle as the hybrid model (such as time-series recursive integration, and the core architecture of "constraint function + driving function"). The only difference between the two is the learning object: the hybrid model learns the "deviation between the simulated and measured top-layer oil temperature", while the reference model directly learns the "evolution law of the measured top-layer oil temperature (experimental data)". The specific model formula and input feature set definition are as follows:

[0079] (9)

[0080] in, and It is a parameterized function related to the measured oil temperature, and both depend only on the input feature X.

[0081] Similarly, the same model as shown in Tables 1 and 2 was used for training, and the same proportions (80% and 20%) of the training and test datasets were used.

[0082] By introducing an enhanced feature set containing simulated values ​​of ambient temperature, load factor, and top oil temperature through linear and nonlinear residual networks, a deep fusion of data and physical information is achieved, overcoming the limitation of pure data-driven models lacking mechanistic support. The extended features of the first four time steps are used to characterize temporal dependencies, and a deviation prediction model accurately distinguishes between known deviations and integrally calculated deviations, achieving refined modeling of deviation evolution over time. Simultaneously, both linear and nonlinear forms are provided to flexibly adapt to different deviation characteristic scenarios, making it more universal than traditional single models. Comparison with a reference model demonstrates the significant advantages of this hybrid model in balancing accuracy and stability.

[0083] In another exemplary embodiment, stability constraints are applied during the training of the deviation prediction model. The stability constraints are that the linear constraint term function or the nonlinear adaptive constraint term function in the deviation prediction model is less than or equal to 0. That is, stability constraints are applied during the training of the deviation prediction model. For the linear deviation model, the eigenvalues ​​of its parameter matrix are constrained to be ≤0; for the nonlinear deviation model, the critical coefficients of its nonlinear terms are constrained to be ≤0, ensuring that the deviation model converges during the time integration process.

[0084] In this embodiment, to prevent the bias model from diverging in long-term predictions, all bias models ultimately applied to real-world scenarios must be subject to stability constraints in this step. To demonstrate the impact of stability, a residual network (ResNet) is trained. The formula used during training without stability constraints is as follows:

[0085] (10)

[0086] No conditions were applied during the calculation of G and F.

[0087] Its discrete form is:

[0088] (11)

[0089] While the above formula can provide excellent prediction results, it is important to note that here... Solution of time It is based on Precise deviation of time It was calculated.

[0090] However, in complete integrals, It is based on the previously calculated To calculate, that is:

[0091] (12)

[0092] This will produce extremely poor prediction results, which is a direct consequence of the lack of stability constraints.

[0093] The following section introduces the application of stability constraints to linear dynamic systems, starting with a parameterized dynamic system:

[0094] (13)

[0095] in, state of time Depends on the state at the previous moment and loads that drive state changes Assume that the loading depends not only on the state but also on a set of parameters (input features), which are combined here into a vector X. Therefore, the learning process aims to compute the regression function. .

[0096] However, to obtain a stable integrator, certain constraints must be satisfied. As mentioned earlier, these constraints are relatively simple in the linear case, so we linearize the forcing terms as follows:

[0097] (14)

[0098] Now, as long as the conditions are met Stability can then be guaranteed. This constraint can be achieved by relating it to... This can be easily achieved by introducing a penalty term into the loss function of the relevant neural network.

[0099] Linearized Residual Networks (ResNet) are similar to Dynamic Mode Decomposition (DMD), but they may be simpler in terms of introducing parameter dimensions.

[0100] Inspired by the above principles, a feasible method that guarantees stability without compromising nonlinear behavior is to express it as:

[0101] (15)

[0102] Simultaneously constructing the regression function Applying non-positive constraints at the time, i.e. .

[0103] In another exemplary embodiment, step 105 above feeds back the learned dynamic deviation to the simulated top oil temperature value to achieve the final prediction. The corrected top oil temperature is:

[0104] (16)

[0105] in, The dynamic deviation of the top oil temperature at the current time t is the output of the deviation prediction model.

[0106] The technical solutions provided by the above-described method embodiments have the following advantages:

[0107] 1) By adopting a data-physical fusion strategy, we not only rely on the principle of transformer thermodynamics to build a physical model to provide mechanistic support, but also combine measured data with deviation model dynamic correction. This solves the problems of large prediction deviation of pure physical model and weak generalization ability of pure data-driven method, and significantly improves the accuracy of top oil temperature monitoring.

[0108] 2) Apply stability constraints during the training of the deviation model. For linear models, restrict the eigenvalues ​​of the parameter matrix to ≤0, and for nonlinear models, restrict the key coefficients to ≤0. This ensures that the model converges during long-term time integration and avoids the prediction divergence caused by sudden changes in operating conditions or data sparsity in traditional models. This makes the model suitable for complex and ever-changing operating scenarios.

[0109] 3) This invention is applicable to power transformers of different types and capacities. Through multi-source data acquisition (temperature, load, tap changer position, etc.) and closed-loop optimization mechanism (deviation feedback to update physical model parameters), it can adapt to various operating conditions and reduce temperature prediction error by more than 40% compared with traditional methods, providing a reliable basis for transformer condition assessment and fault early warning.

[0110] Based on the same inventive concept, this application also provides a power transformer temperature monitoring device for implementing the power transformer temperature monitoring method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more power transformer temperature monitoring device embodiments provided below can be found in the limitations of the power transformer temperature monitoring method described above, and will not be repeated here.

[0111] In one exemplary embodiment, a power transformer temperature monitoring device is provided, comprising:

[0112] The heat generation power calculation module is used to calculate the heat generation power of the power transformer at the current moment based on the load factor, average winding temperature and tap changer position of the power transformer at the current moment, using the physical model of the power transformer.

[0113] The top oil temperature simulation module is used to input the current heat generation power into the heat balance equation of the power transformer to solve for the current simulated value of the top oil temperature.

[0114] An enhanced feature sequence construction module is used to compose an enhanced feature sequence from the physical features of the current time and a preset number of time steps prior to the current time; the physical features include ambient temperature, load factor, and simulated top oil temperature.

[0115] The top-layer oil temperature dynamic deviation prediction module is used to input the enhanced feature sequence into the deviation prediction model to obtain the current top-layer oil temperature dynamic deviation; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network.

[0116] The calibration module is used to calibrate the simulated value of the top oil temperature at the current moment using the dynamic deviation of the top oil temperature at the current moment, and obtain the calibrated simulated value of the top oil temperature as the monitoring value of the top oil temperature.

[0117] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 3 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), 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 computational 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 databases. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for monitoring the temperature of a power transformer.

[0118] Those skilled in the art will understand that Figure 3 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0119] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0120] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0121] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0122] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for monitoring the temperature of a power transformer, characterized in that, include: Based on the current load factor, average winding temperature, and tap changer position of the power transformer, the heat generation power at the current moment is calculated using the physical model of the power transformer. The heat generation power at the current moment is substituted into the heat balance equation of the power transformer to solve for the simulated value of the top oil temperature at the current moment. The physical characteristics of the current moment and a preset number of moments prior to the current moment are combined to form an enhanced feature sequence; the physical characteristics include ambient temperature, load factor, and simulated top oil temperature. The enhanced feature sequence is input into the deviation prediction model to obtain the dynamic deviation of the top oil temperature at the current moment; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network. The current dynamic deviation of the top oil temperature is used to calibrate the current simulated value of the top oil temperature, and the calibrated simulated value of the top oil temperature is used as the monitoring value of the top oil temperature. The heat balance equation is: ； in, This refers to the heating power of the power transformer. It is the load factor; It is the average winding temperature. This is the position of the tap changer. and These are the heat capacity and thermal resistance of the equivalent thermal circuit of the power transformer, respectively. It is the temperature difference between the simulated top oil temperature and the ambient temperature; This is a simulated value for the top oil temperature; It is the ambient temperature; It is the difference between the average winding temperature and the simulated value of the top oil temperature, k is the ambient temperature transformation coefficient, and t is the time variable.

2. The method for monitoring the temperature of a power transformer according to claim 1, characterized in that, The heat capacity and thermal resistance of the equivalent thermal circuit of a power transformer are the core parameters of the heat balance equation; these core parameters are obtained in the following manner: Based on the experimentally obtained measured value of top oil temperature, the core parameters in the heat balance equation are adjusted so that the absolute value of the difference between the simulated value of top oil temperature and the measured value of top oil temperature obtained by solving the heat balance equation is less than a preset threshold.

3. The method for monitoring the temperature of a power transformer according to claim 2, characterized in that, In the process of obtaining the core parameters, the core parameters are assumed to be constants or dynamic changes related to the top oil temperature.

4. The method for monitoring the temperature of a power transformer according to claim 1, characterized in that, The deviation prediction model is as follows: ； in, This represents the dynamic deviation of the top oil temperature at time n, as output by the deviation prediction model. The dynamic deviation of the top oil temperature at time n-1 is the output of the deviation prediction model. For time intervals, For relying solely on enhanced feature sequences The linear constraint term function, For relying solely on enhanced feature sequences The deviation dynamic driving term function, and All were obtained based on the construction of Long Short-Term Memory (LSTM) networks. , , , , The ambient temperatures at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The load factors at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The simulated values ​​of the top oil temperature at times n-4, n-3, n-2, n-1, and n are respectively.

5. The method for monitoring the temperature of a power transformer according to claim 1, characterized in that, The deviation prediction model is as follows: ； in, This represents the dynamic deviation of the top oil temperature at time n, as output by the deviation prediction model. The dynamic deviation of the top oil temperature at time n-1 is the output of the deviation prediction model. For time intervals, For simultaneous dependence on enhanced feature sequences The dynamic deviation of the top oil temperature at time n-1 from the deviation prediction model output The nonlinear adaptive constraint term function, For relying solely on enhanced feature sequences The deviation dynamic driving term function, and All were obtained based on the construction of Long Short-Term Memory (LSTM) networks. , , , , The ambient temperatures at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The load factors at times n-4, n-3, n-2, n-1, and n are respectively. , , , , The simulated values ​​of the top oil temperature at times n-4, n-3, n-2, n-1, and n are respectively.

6. The method for monitoring the temperature of a power transformer according to claim 1, characterized in that, During the training of the deviation prediction model, a stability constraint is applied, wherein the linear constraint term function or the nonlinear adaptive constraint term function in the deviation prediction model is less than or equal to 0.

7. The method for monitoring the temperature of a power transformer according to claim 1, characterized in that, The simulated value of the top oil temperature at the current moment is calibrated using the dynamic deviation of the top oil temperature at the current moment, and the calibrated simulated value of the top oil temperature is obtained. The formula for the top oil temperature monitoring value is as follows: ； in, The top oil temperature is the monitored value at the current time t. This is the simulated top-layer oil temperature at the current time t. The dynamic deviation of the top oil temperature at the current time t is the output of the deviation prediction model.

8. A power transformer temperature monitoring device, characterized in that, The power transformer temperature monitoring device uses the power transformer temperature monitoring method according to any one of claims 1-7, and the power transformer temperature monitoring device includes: The heat generation power calculation module is used to calculate the heat generation power of the power transformer at the current moment based on the load factor, average winding temperature and tap changer position of the power transformer at the current moment, using the physical model of the power transformer. The top oil temperature simulation module is used to input the current heat generation power into the heat balance equation of the power transformer to solve for the current simulated value of the top oil temperature. An enhanced feature sequence construction module is used to compose an enhanced feature sequence from the physical features of the current time and a preset number of time steps prior to the current time; the physical features include ambient temperature, load factor, and simulated top oil temperature. The top-layer oil temperature dynamic deviation prediction module is used to input the enhanced feature sequence into the deviation prediction model to obtain the current top-layer oil temperature dynamic deviation; the deviation prediction model is obtained by training a machine learning model that integrates residual network and long short-term memory network. The calibration module is used to calibrate the simulated value of the top oil temperature at the current moment using the dynamic deviation of the top oil temperature at the current moment, and obtain the calibrated simulated value of the top oil temperature as the monitoring value of the top oil temperature.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the power transformer temperature monitoring method according to any one of claims 1-7.

Images