A method and system for online prediction of temperature rise trend of high-voltage electrical equipment

Abstract

This invention discloses an online prediction method and system for the temperature rise trend of high-voltage electrical equipment, belonging to the field of electrical measurement and equipment condition monitoring technology. It includes acquiring the operating electrical variables and environmental variables of the high-voltage electrical equipment, preprocessing these variables to generate preprocessed data; constructing a physical temperature rise model based on the preprocessed data to generate physical prediction results; constructing a data-driven prediction model based on the preprocessed data to generate data-driven prediction results; acquiring the actual measured temperature of the high-voltage electrical equipment, and calculating the prediction error by combining the physical prediction results and the data-driven prediction results; and dynamically assigning fusion weights to the physical prediction results and the data-driven prediction results based on the prediction error for weighted fusion to generate the final temperature rise trend prediction. The strategy of parallel prediction using the physical model and the data-driven model, and dynamic weighted fusion based on real-time error, can improve the accuracy, robustness, and adaptability of the temperature rise prediction.

Description

Technical Field

[0001] This invention relates to the field of electrical measurement and equipment condition monitoring technology, and in particular to an online prediction method and system for the temperature rise trend of high-voltage electrical equipment. Background Technology

[0002] High-voltage electrical equipment is a core component of the power system, and its safe and stable operation is crucial. During operation, the equipment generates heat due to the thermal effect of current, causing its internal temperature to rise. Abnormal temperature rises are important indicators of potential risks such as insulation aging, poor connections, or internal faults. Therefore, accurate online prediction of the temperature rise trend of high-voltage electrical equipment is a key technical means to achieve equipment condition early warning, preventive maintenance, and ensure power grid safety.

[0003] Currently, there are two main technical approaches to predicting the temperature rise of high-voltage electrical equipment. One approach is based on thermodynamic principles to construct a physical temperature rise model. This involves establishing the equipment's thermal balance equation and calculating future temperature changes based on measured parameters such as current and ambient temperature. The other approach uses a data-driven method, employing machine learning or deep learning algorithms to directly learn and mine the complex mapping relationship between operating parameters and equipment temperature from massive amounts of historical operating data, thereby constructing a predictive model.

[0004] However, existing technical solutions all have certain limitations. Prediction methods based on physical temperature rise models often require simplification of the actual equipment during model building, and the thermodynamic parameters used are usually fixed values. This makes it difficult to accurately reflect the true thermal characteristics of the equipment under complex and variable operating conditions, especially when the equipment ages or is affected by unmodeled factors, leading to a decrease in prediction accuracy. Purely data-driven prediction methods, on the other hand, rely on the completeness and coverage of historical training data. The models lack physical interpretability, and the reliability of prediction results cannot be guaranteed when the equipment encounters operating conditions not present in the training set. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an online prediction method and system for the temperature rise trend of high-voltage electrical equipment. It employs a strategy of parallel prediction using physical models and data-driven models, and dynamically weighted fusion based on real-time errors, which can improve the accuracy, robustness, and adaptability of temperature rise prediction.

[0006] The above objectives can be achieved through the following approach:

[0007] An online prediction method for the temperature rise trend of high-voltage electrical equipment includes: acquiring the operating electrical variables and environmental variables of the high-voltage electrical equipment, preprocessing the operating electrical variables and environmental variables to generate preprocessed data; constructing a physical temperature rise model based on the preprocessed data, and generating physical prediction results based on the physical temperature rise model; constructing a data-driven prediction model based on the preprocessed data, and generating data-driven prediction results based on the data-driven prediction model; acquiring the actual measured temperature of the high-voltage electrical equipment, and calculating the prediction error by combining the physical prediction results and the data-driven prediction results; dynamically assigning fusion weights to the physical prediction results and the data-driven prediction results based on the prediction error, and performing weighted fusion of the physical prediction results and the data-driven prediction results based on the fusion weights to generate a final temperature rise trend prediction.

[0008] Optionally, the generation of preprocessed data includes: acquiring current and voltage parameters as operating electrical variables, and ambient temperature and humidity parameters as environmental variables; filtering the current parameters, voltage parameters, ambient temperature parameters, and ambient humidity parameters to generate filtered data; and normalizing the filtered data to generate preprocessed data.

[0009] Optionally, generating the physical prediction result includes: extracting thermodynamic parameters based on the preprocessed data; constructing a heat balance equation based on the thermodynamic parameters to generate physical model parameters; and performing calculations based on the physical model parameters and the preprocessed data to generate the physical prediction result.

[0010] Optionally, generating data-driven prediction results includes: acquiring historical temperature data and historical operating data, integrating the historical temperature data and historical operating data to generate a training dataset; training an adaptive prediction algorithm based on the training dataset to generate data-driven model parameters; and performing calculations based on the data-driven model parameters and the preprocessed data to generate data-driven prediction results.

[0011] Optionally, the method further includes: acquiring broadband ultra-high frequency electromagnetic wave signals generated by partial discharge inside the high-voltage electrical equipment using a preset ultra-high frequency partial discharge detection technology; converting the broadband ultra-high frequency electromagnetic wave signals into logarithmic energy characteristic signals, filtering out non-power frequency noise through harmonic analysis, and obtaining partial discharge characteristic values; and integrating the partial discharge characteristic values ​​as additional features characterizing insulation degradation into the training dataset.

[0012] Optionally, generating the final temperature rise trend prediction includes: calculating the absolute difference between the physical prediction result and the actual measured temperature as a first error, and calculating the absolute difference between the data-driven prediction result and the actual measured temperature as a second error; combining the first error and the second error to form a prediction error; dynamically assigning fusion weights to the physical prediction result and the data-driven prediction result according to the proportional relationship between the first error and the second error; and using the fusion weights to perform a weighted average of the physical prediction result and the data-driven prediction result to generate the final temperature rise trend prediction.

[0013] Optionally, the method further includes: determining whether the prediction error exceeds a threshold; if it does, updating the data-driven model parameters by re-acquiring data to generate updated data-driven model parameters; regenerating the data-driven prediction result based on the updated data-driven model parameters, and performing weighted fusion with the physical prediction result.

[0014] Optionally, the method further includes: determining whether the prediction error exceeds a threshold; if it does, correcting the thermodynamic parameters in the physical temperature rise model to generate updated physical model parameters; regenerating the physical prediction result based on the updated physical model parameters, and performing weighted fusion with the data-driven prediction result.

[0015] Optionally, generating updated physical model parameters includes: calculating the thermal resistance coefficient correction value in reverse based on the prediction error to generate corrected thermodynamic parameters; updating the heat balance equation in the physical temperature rise model based on the corrected thermodynamic parameters to generate an adjusted physical model; and generating updated physical model parameters based on the adjusted physical model.

[0016] Based on the same inventive concept, the present invention also provides an online prediction system for the temperature rise trend of high-voltage electrical equipment, the system comprising:

[0017] The multi-dimensional sensing and data preparation module is used to acquire the operating electrical variables and environmental variables of high-voltage electrical equipment, and to preprocess the operating electrical variables and environmental variables to generate preprocessed data.

[0018] The physical temperature rise model prediction module is used to construct a physical temperature rise model based on the preprocessed data and generate physical prediction results based on the physical temperature rise model.

[0019] The data-driven prediction model module is used to construct a data-driven prediction model based on the preprocessed data and generate data-driven prediction results based on the data-driven prediction model.

[0020] The prediction error calculation module is used to obtain the actual measured temperature of the high-voltage electrical equipment and calculate the prediction error by combining the physical prediction result and the data-driven prediction result.

[0021] The intelligent fusion temperature rise trend prediction module is used to dynamically assign fusion weights to the physical prediction result and the data-driven prediction result based on the prediction error, and to perform weighted fusion of the physical prediction result and the data-driven prediction result based on the fusion weights to generate the final temperature rise trend prediction.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] 1. By constructing a dual-track prediction architecture that combines a physical temperature rise model and a data-driven prediction model, and dynamically allocating fusion weights based on actual measurement errors, the advantages of the two models are complemented in real time, improving the overall accuracy and robustness of temperature rise trend prediction. This ensures that the prediction results closely match the actual temperature rise status of the equipment under various complex working conditions.

[0024] 2. By setting a prediction error threshold, an online update and correction mechanism for the data-driven model or physical model is triggered, enabling the prediction system to proactively adapt to characteristic drift caused by factors such as equipment aging, environmental changes, or changes in operating modes, thus ensuring the effectiveness and high accuracy of the prediction method throughout the entire equipment lifecycle.

[0025] 3. By combining physical models with data-driven models, the limitations of single physical models in accurately describing complex nonlinear processes are overcome, and the problems of single data-driven models lacking physical interpretability and having weak extrapolation capabilities on training data are solved. This results in predictions that have both the high accuracy brought by data intelligence and the reliability guaranteed by physical laws.

[0026] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating an online prediction method for the temperature rise trend of high-voltage electrical equipment according to an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of the dual-model prediction and dynamic weighted fusion effect curves in an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the online correction effect curve of the physical model in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the structure of an online prediction system for the temperature rise trend of high-voltage electrical equipment according to an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Reference Figure 1 One embodiment of the present invention proposes an online prediction method for the temperature rise trend of high-voltage electrical equipment. It adopts a strategy of parallel prediction using physical models and data-driven models, and dynamically weighted fusion based on real-time errors, which can improve the accuracy, robustness and adaptability of temperature rise prediction.

[0034] The method described in this embodiment specifically includes:

[0035] The operating electrical variables and environmental variables of high-voltage electrical equipment are acquired, and the operating electrical variables and environmental variables are preprocessed to generate preprocessed data;

[0036] A physical temperature rise model is constructed based on the preprocessed data, and physical prediction results are generated based on the physical temperature rise model.

[0037] A data-driven prediction model is constructed based on the preprocessed data, and data-driven prediction results are generated based on the data-driven prediction model.

[0038] The actual measured temperature of the high-voltage electrical equipment is obtained, and the prediction error is calculated by combining the physical prediction result with the data-driven prediction result.

[0039] Based on the prediction error, a fusion weight is dynamically assigned to the physical prediction result and the data-driven prediction result. The physical prediction result and the data-driven prediction result are then weighted and fused based on the fusion weight to generate the final temperature rise trend prediction.

[0040] Specifically, two complementary temperature rise prediction models are constructed: a physical temperature rise model established using thermodynamic laws to describe the basic laws and intrinsic mechanisms of equipment temperature rise; and a data-driven prediction model trained based on historical operating data to uncover the complex nonlinear mapping relationship between operating variables and temperature rise. By calculating the deviation between the prediction results of each model and the actual temperature in real time, a dynamic prediction error is obtained. Based on this error, the contribution of the two prediction results to the final decision is adjusted, assigning a higher fusion weight to the model with the better current performance, thus generating a dynamically weighted fusion final prediction value. By combining the advantages of the physical model and the data-driven model, the overall performance of high-voltage electrical equipment temperature rise trend prediction is improved. The introduction of the physical temperature rise model ensures the stability and robustness of the prediction results under normal operating conditions, avoiding predictions that violate physical laws that may occur with purely data-driven methods. The incorporation of the data-driven prediction model effectively compensates for errors caused by simplification or parameter inaccuracies in the physical model, accurately capturing the impact of complex factors such as equipment aging and sudden environmental changes on temperature rise.

[0041] Optionally, the generation of preprocessed data includes:

[0042] Obtain the current and voltage parameters as operating electrical variables, and the ambient temperature and humidity parameters as environmental variables;

[0043] The current parameter, the voltage parameter, the ambient temperature parameter, and the ambient humidity parameter are filtered to generate filtered data;

[0044] The filtered data is then normalized to generate preprocessed data.

[0045] Specifically, a sensor system deployed near key temperature measurement points on the equipment and at the monitoring center acquires in real-time the operating electrical variables and voltage parameters affecting the equipment's temperature rise. The operating electrical variables mainly include current parameters measured by current transformers and voltage parameters measured by voltage transformers; these two parameters directly determine the rate of heat generation within the equipment. Environmental variables include ambient temperature and humidity parameters acquired by temperature and humidity sensors; these two parameters are critical boundary conditions for heat exchange between the equipment and the external environment. After acquiring the raw time-series data, to eliminate the impact of sensor noise, electromagnetic interference, or instantaneous fluctuations during data transmission on the subsequent model prediction accuracy, the collected current, voltage, ambient temperature, and ambient humidity parameters need to be filtered. The filtering process employs a moving average filtering algorithm, which smooths the data sequence by calculating the arithmetic mean of data within a specific time window, effectively filtering out high-frequency noise and preserving the long-term trend characteristics of the data, thus generating filtered data. To address the impact of differences in physical dimensions and numerical ranges on model training and calculation, the filtered data needs to be normalized. A maximum-minimum normalization method is used, and its calculation method is as follows:

[0046] ;

[0047] In the formula, The normalized data represents the final preprocessed data; X represents the filtered data value at a certain moment. and These are the maximum and minimum values ​​of the data within a historical operating cycle or determined based on the rated operating range of the equipment.

[0048] Optionally, the generated physical prediction results include:

[0049] Thermodynamic parameters are extracted based on the preprocessed data;

[0050] Based on the aforementioned thermodynamic parameters, a thermal equilibrium equation is constructed, and physical model parameters are generated.

[0051] The physical prediction results are generated by calculating based on the physical model parameters and the preprocessed data.

[0052] Based on the acquired preprocessed data, particularly the current, voltage, and ambient temperature parameters, key thermodynamic parameters characterizing the thermal properties of high-voltage electrical equipment are extracted or calibrated. These thermodynamic parameters mainly include the equivalent heat-generating resistance inside the equipment and the overall equivalent thermal resistance and heat capacity. They are initially set according to the equipment's design data and material properties, and then calibrated and identified using historical operating data. A heat balance equation describing the dynamic process of temperature rise in the equipment is constructed. A simplified lumped-parameter thermal circuit model can be expressed as:

[0053] ;

[0054] In the formula, C represents the equivalent heat capacity of the critical part of the high-voltage electrical equipment, which is used to characterize its ability to store heat; ΔT is the difference between the temperature of the critical part of the equipment and the ambient temperature, also known as temperature rise. It is the rate of change of temperature rise over time; This refers to the total heat generation power inside the equipment, mainly composed of Joule heat generated by the current flowing through the conductor. It can be calculated using the current parameter I and the equivalent heat generation resistance R from the preprocessed data, and is equal to... ; It is the heat dissipation power of the equipment to the environment, which is proportional to the temperature rise ΔT, and can be expressed as... ,in The equivalent thermal resistance of the equipment represents its heat dissipation capacity. By combining and organizing the above thermodynamic parameters, physical model parameters are generated. Based on the determined physical model parameters and real-time preprocessed data, numerical calculations are performed to generate physical prediction results. Specifically, the differential form of the heat balance equation is discretized to obtain a recursive calculation formula. Using this recursive formula, the equipment temperature at the next moment is predicted using the current equipment temperature, ambient temperature, and current parameters; this is the physical prediction result.

[0055] Optionally, the generated data-driven prediction results include:

[0056] Acquire historical temperature data and historical operation data, and integrate the historical temperature data and historical operation data to generate a training dataset;

[0057] An adaptive prediction algorithm is trained based on the training dataset to generate data-driven model parameters.

[0058] The data-driven prediction results are generated by calculating the data-driven model parameters and the preprocessed data.

[0059] Specifically, a training dataset is constructed by acquiring long-term historical temperature and operational data from historical databases. Historical temperature data consists of actual measured temperature sequences at key temperature measurement points of high-voltage electrical equipment, while historical operational data includes time-synchronized operational electrical variables such as current and voltage parameters, and environmental variables such as ambient temperature and humidity parameters. These multi-dimensional historical data are time-aligned and integrated, and a series of input feature and output label pairs are constructed using a sliding time window approach. Input features are typically sequences of operational and temperature data from the past period, while output labels are equipment temperature values ​​for one or more future time steps. These data pairs together constitute the training dataset. An adaptive prediction algorithm is then trained based on this constructed training dataset. The adaptive prediction algorithm employs a deep learning model capable of capturing long-term dependencies in time-series data, such as a Long Short-Term Memory network or a gated recurrent unit. The training process is an iterative optimization process. The algorithm feeds the input features from the training dataset into the model, which calculates and outputs a predicted value based on its internal parameters. Then, it calculates the difference between this predicted value and the true output label using a loss function such as mean squared error. Finally, it uses the backpropagation algorithm and optimizer to adjust the model's internal network weights and biases based on this difference. This process is repeated until the model's prediction error on the validation set converges to a preset range. The network weights and biases obtained at this point are the data-driven model parameters. Real-time acquired and preprocessed data is used as input and fed into the pre-trained prediction model with its data-driven model parameters fixed. The model performs a forward propagation calculation, using its fixed internal parameters to perform nonlinear mapping and computation on the real-time input data, directly outputting a predicted value for the future temperature rise trend—the data-driven prediction result.

[0060] Optionally, the method further includes:

[0061] The broadband ultra-high frequency electromagnetic wave signal generated by the internal partial discharge of the high-voltage electrical equipment is collected by a preset ultra-high frequency partial discharge detection technology.

[0062] The broadband ultra-high frequency electromagnetic wave signal is converted into a logarithmic energy characteristic signal, and non-power frequency noise is filtered out by harmonic analysis to obtain the partial discharge characteristic value;

[0063] The partial discharge feature values ​​are integrated into the training dataset as additional features characterizing insulation degradation.

[0064] Specifically, in constructing the training dataset, to compensate for the inadequacy of conventional operating electrical variables such as current and voltage in reflecting latent defects such as insulation aging within equipment, a partial discharge monitoring dimension is introduced. Partial discharge is a phenomenon where only certain areas of the insulation material in high-voltage electrical equipment discharges, and it is one of the main factors causing insulation damage, insulation breakdown, explosion, and fire in high-voltage equipment. In the early stages and during the development of partial discharge, abundant electromagnetic radiation pulses in the UHF band are radiated outward. Therefore, broadband ultra-high frequency electromagnetic wave signals generated by partial discharge within power equipment are collected through pre-deployed ultra-high frequency partial discharge sensors. Due to the large dynamic range of this raw signal and the presence of complex background interference, a multi-stage demodulation logarithmic detector is used to convert the extracted ultra-high frequency energy into a logarithmic energy characteristic signal characterizing the discharge intensity.

[0065] ;

[0066] in, The output logarithmic energy characteristic signal is typically represented in circuits as a low-frequency voltage signal characterizing the energy envelope of the discharge pulse. The extracted ultra-high frequency energy refers to the input power of the transient partial discharge radio frequency signal sensed by the ultra-high frequency sensor. The preset reference energy standard, usually determined based on the detector's hardware intercept point, is used as the base value for calculating the logarithmic energy. The slope coefficient of the detector reflects the gain ratio of the output signal voltage to the input power in decibels, and is determined by the characteristics of the multi-stage amplifier circuit in the hardware. Subsequently, harmonic analysis is performed using FFT (Fast Fourier Transform) to filter out non-power frequency noise, thereby obtaining partial discharge characteristic values. The obtained partial discharge characteristic values, as key additional features characterizing the internal insulation degradation state of the equipment, are integrated into the training dataset and time-series aligned and stitched with historical temperature data and historical operating data.

[0067] For example, consider a high-voltage switchgear in a wind farm substation. During long-term operation, the staggered contacts in the busbar compartment of this switchgear experienced early contact loosening and slight insulation degradation due to mechanical vibration or long-term heavy load impact. In the early stages of this degradation, the increase in contact resistance was insufficient to trigger a drastic change in macroscopic heat, and conventional operating electrical variables such as current parameters did not undergo significant abrupt changes. The physical temperature rise model, based on the inherent heat balance equation, still showed a stable fluctuation trend in its physical prediction results. However, at this point, the loosened and deteriorated parts began to generate weak partial discharges and radiated broadband ultra-high frequency electromagnetic wave signals. After the ultra-high frequency sensor deployed in the switchgear captured this radio frequency signal, due to the extremely large energy range of the partial discharge, the acquired transient ultra-high frequency energy was compressed and converted into a low-frequency envelope logarithmic energy characteristic signal according to the logarithmic transformation relationship. This effectively avoided the weak precursor signal being submerged by background noise or the strong signal causing saturation of the acquisition channel. Subsequently, FFT harmonic analysis was performed on the logarithmic energy characteristic signal to filter out non-power frequency noise such as communication interference in the substation environment, and to accurately extract partial discharge characteristic values ​​that are strongly correlated with the power frequency.

[0068] Optionally, generating the final temperature rise trend prediction includes:

[0069] The absolute difference between the physical prediction result and the actual measured temperature is calculated as the first error, and the absolute difference between the data-driven prediction result and the actual measured temperature is calculated as the second error. The first error and the second error are combined to form the prediction error.

[0070] Based on the proportional relationship between the first error and the second error, dynamically assign fusion weights to the physical prediction result and the data-driven prediction result;

[0071] The fusion weights are used to perform a weighted average of the physical prediction results and the data-driven prediction results to generate the final temperature rise trend prediction.

[0072] Specifically, the prediction error between the physical prediction result and the data-driven prediction result is calculated in real time. At each prediction time point, the actual measured temperature of the high-voltage electrical equipment from the sensor is obtained as the true value benchmark. The absolute difference between the physical prediction result and the actual measured temperature is defined as the first error; the absolute difference between the data-driven prediction result and the actual measured temperature is defined as the second error. These two error values ​​together constitute the comprehensive prediction error at the current moment. Based on these two error values, fusion weights are dynamically assigned to the physical prediction result and the data-driven prediction result. The weight allocation follows a core principle: the smaller the prediction error of the model, the higher its reliability, and it should occupy a larger proportion in the final prediction. Therefore, the allocation of fusion weights is inversely proportional to the prediction error. Specifically, the fusion weights of the physical model and the data-driven model can be calculated in the following way:

[0073] ;

[0074] ;

[0075] In the formula, For the fusion weights assigned to the physical prediction results, The fusion weights assigned to the data-driven prediction results; This is the first error, which is the prediction error of the physical model; The second error is the prediction error of the data-driven model. This allocation method ensures that the sum of the two weights is always 1, and the smaller the error of either model, the larger its corresponding weight. The assigned fusion weights will be used to linearly combine the physical prediction results and the data-driven prediction results. The final prediction value is given by the following formula:

[0076] ;

[0077] In the formula, This is the final temperature rise trend prediction generated; It is a physical prediction result generated by the physical temperature rise model; This is the data-driven prediction result generated by the data-driven prediction model. See the dual-model fusion effect for details. Figure 2 When the load suddenly increases, the data-driven weight increases to 0.86 to correct the deviation of the physical model. After the physical model is updated, the weight rises back to 0.71. The fusion prediction always closely matches the actual temperature rise, which confirms the effectiveness of the dynamic weight allocation.

[0078] Optionally, the method further includes:

[0079] Determine whether the prediction error exceeds a threshold. If it does, update the data-driven model parameters by re-acquiring data and generate updated data-driven model parameters.

[0080] The data-driven prediction results are regenerated based on the updated data-driven model parameters and then weighted and fused with the physical prediction results.

[0081] Specifically, the generated final temperature rise trend prediction is continuously compared with the actual measured temperature of the high-voltage electrical equipment acquired in real time, and the deviation between the two, i.e., the prediction error, is calculated. Simultaneously, a reasonable prediction error threshold is preset, determined based on the equipment's safe operation requirements and prediction accuracy standards. The absolute value of the prediction error is checked in real time to see if it exceeds the preset threshold. If the prediction error is within the threshold range, it indicates that the current data-driven model parameters can still capture the equipment's operating patterns well, and the existing model continues to be used for prediction. When the prediction error exceeds the threshold, it is considered that the operating characteristics of the high-voltage electrical equipment may have changed, such as changes in thermal characteristics due to equipment aging or fundamental shifts in load patterns. These changes render the original data-driven model unable to accurately fit the current state. At this point, an update mechanism is triggered. Historical operating data and historical temperature data from the most recent period are automatically reacquired to form a new training sample set. This new training sample set is used to incrementally train the existing data-driven prediction model, adjusting the model's internal network weights through optimization algorithms, thereby generating an updated set of data-driven model parameters. The updated data-driven model parameters are used to regenerate the data-driven prediction results. These regenerated data-driven prediction results are then combined with the physical prediction results generated by the physical temperature rise model using a dynamic weighted fusion method to generate the final temperature rise trend prediction.

[0082] Optionally, the method further includes:

[0083] Determine whether the prediction error exceeds a threshold. If it does, correct the thermodynamic parameters in the physical temperature rise model and generate updated physical model parameters.

[0084] The physical prediction results are regenerated based on the updated physical model parameters and then weighted and fused with the data-driven prediction results.

[0085] Specifically, the system continuously monitors and determines whether the prediction error generated by the final fusion result exceeds a pre-set threshold. This threshold is set based on the acceptable average error level during long-term operation. If the prediction error consistently exceeds this range, it is determined that the thermodynamic parameters in the current physical temperature rise model can no longer accurately reflect the true thermophysical state of the equipment. Once the prediction error is determined to exceed the threshold, a correction procedure for the thermodynamic parameters in the physical temperature rise model is triggered. Using statistically significant prediction errors and corresponding operational data, key parameters in the physical model are corrected in reverse. For example, if the predicted temperature is lower than the actual temperature, it may mean that the actual equivalent thermal resistance of the equipment has increased, and the heat dissipation capacity has decreased. In this case, the thermal resistance coefficient in the thermal balance equation is adjusted according to the deviation, generating corrected thermodynamic parameters. The corrected thermodynamic parameters are used to reconstruct the physical temperature rise model, especially the thermal balance equation. Subsequently, based on this updated physical model and real-time preprocessed data, calculations are re-performed to generate new physical prediction results. Finally, this corrected physical prediction result is weighted and fused again with the data-driven prediction result generated by the data-driven prediction model using a dynamically allocated fusion weight method to produce the final temperature rise trend prediction.

[0086] Optionally, the generated updated physical model parameters include:

[0087] Based on the prediction error, the thermal resistance coefficient correction value is calculated in reverse to generate the corrected thermodynamic parameters;

[0088] The heat balance equation in the physical temperature rise model is updated based on the corrected thermodynamic parameters to generate the adjusted physical model.

[0089] Updated physical model parameters are generated based on the adjusted physical model.

[0090] Specifically, when the prediction error consistently exceeds a preset threshold, indicating that the thermodynamic parameters of the physical model have deviated from the actual thermal characteristics of the equipment, the following correction process will be initiated. First, the thermal resistance coefficient correction value is calculated in reverse based on the accumulated prediction error. Under steady-state or quasi-steady-state conditions, there is an approximately linear relationship between the equipment temperature rise, heating power, and equivalent thermal resistance. Based on the actual temperature rise calculated from the actual measured temperature and the currently input heating power, an equivalent thermal resistance target value that reflects the true heat dissipation capacity of the current equipment is derived in reverse. A smooth update algorithm is then used to calculate the new thermal resistance coefficient, thereby generating the corrected thermodynamic parameters. This update process can be expressed as:

[0091] ;

[0092] in, It is the updated equivalent thermal resistance, i.e., the generated corrected thermodynamic parameters; It is the old equivalent thermal resistance used in the physical model before the revision; It is an update factor between 0 and 1, used to control the speed of correction and avoid drastic parameter fluctuations caused by instantaneous disturbances; It is the actual temperature rise calculated based on the actual measured temperature and the ambient temperature. This is the internal heat generation power of the device calculated based on real-time operating electrical variables. Based on the generated corrected thermodynamic parameters, i.e., the new equivalent thermal resistance, the system updates the core heat balance equation in the physical temperature rise model. This means that in the terms describing the device's heat dissipation process, the old equivalent thermal resistance is replaced by the updated equivalent thermal resistance, thus generating an adjusted physical model. Based on the adjusted physical model, a complete set of parameters included in the adjusted physical model is solidified, including the newly corrected equivalent thermal resistance and other unadjusted heat capacity parameters. This complete set of parameters constitutes the updated physical model parameters. See [link to physical model correction results] for details. Figure 3 On a certain day, due to dust accumulation during heat dissipation, the error reached 3.5℃. After correction, the thermal resistance increased by 8%, and the prediction error decreased to 0.7℃, which confirms the adaptability of online calibration to changes in the physical characteristics of the equipment.

[0093] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides an online prediction system for the temperature rise trend of high-voltage electrical equipment, the system comprising:

[0094] The multi-dimensional sensing and data preparation module is used to acquire the operating electrical variables and environmental variables of high-voltage electrical equipment, and to preprocess the operating electrical variables and environmental variables to generate preprocessed data.

[0095] The physical temperature rise model prediction module is used to construct a physical temperature rise model based on the preprocessed data and generate physical prediction results based on the physical temperature rise model.

[0096] The data-driven prediction model module is used to construct a data-driven prediction model based on the preprocessed data and generate data-driven prediction results based on the data-driven prediction model.

[0097] The prediction error calculation module is used to obtain the actual measured temperature of the high-voltage electrical equipment and calculate the prediction error by combining the physical prediction result and the data-driven prediction result.

[0098] The intelligent fusion temperature rise trend prediction module is used to dynamically assign fusion weights to the physical prediction result and the data-driven prediction result based on the prediction error, and to perform weighted fusion of the physical prediction result and the data-driven prediction result based on the fusion weights to generate the final temperature rise trend prediction.

[0099] Example 1:

[0100] To verify the feasibility of this invention in practice, it was applied to the online condition monitoring of a critical power transformer in a 220kV main substation in a certain region. This transformer bears the core responsibility of supplying power to an important industrial area, and its operating temperature is a key indicator for assessing its health status and safety margin. Traditional temperature monitoring methods only provide real-time temperature values ​​and lack effective prediction of future temperature rise trends, making it difficult to provide early warnings and make maintenance decisions before faults occur. This substation aims to use this invention to achieve high-precision, adaptive online prediction of transformer temperature rise trends.

[0101] In this embodiment, the substation uses the prediction system proposed in this invention to monitor the target transformer in all aspects. First, the multi-dimensional sensing and data preparation module acquires operating electrical variables through the transformer's built-in current transformers and voltage transformers, and collects environmental variables such as ambient temperature and humidity through sensors deployed around the transformer. The collected raw data undergoes moving average filtering and max-min normalization to generate standardized preprocessed data.

[0102] Two prediction models run in parallel. The physical temperature rise model prediction module is built based on the transformer's thermal balance equation and uses historical data to calibrate its physical model parameters, such as equivalent thermal resistance and heat capacity, to generate physical prediction results. Simultaneously, the data-driven prediction model module employs a long short-term memory network algorithm, using historical temperature data and historical operating data archived at the substation from the past two years for training, generating data-driven model parameters, and producing data-driven prediction results based on real-time preprocessed data.

[0103] The prediction error calculation module obtains the actual measured value of the transformer top oil temperature and calculates the absolute error between the physical prediction result and the data-driven prediction result and the actual value. Finally, the intelligent fusion temperature rise trend prediction module dynamically allocates fusion weights based on these two errors, performs a weighted average of the two prediction results, and generates the final temperature rise trend prediction. The system also has a built-in prediction error threshold, such as continuously exceeding 3°C. Once triggered, it will initiate an online update of the physical model parameters or the data-driven model parameters.

[0104] To verify the beneficial effects of this invention, the transformer underwent a six-month continuous online test, recording prediction data under different seasons and load conditions, and comparing it with traditional single physical model prediction methods. The following are the data analysis and effect verification results during the test.

[0105] Under normal, stable load conditions, the physical temperature rise model can accurately reflect the basic thermodynamic laws of the transformer with a small prediction error. At this time, the intelligent fusion module automatically assigns a high weight of approximately 0.7 to 0.8 to the physical prediction results, making the final prediction results stable and highly interpretable.

[0106] Under conditions of sudden load changes, such as at 2:00 PM one afternoon when downstream factories simultaneously started production, the load current rapidly increased from 800A to 1200A. Due to a certain time lag, the physical model's predictions were inaccurate. However, the data-driven model, trained on a large amount of historical data, was better able to capture such nonlinear and dynamic changes, resulting in higher prediction accuracy. At this point, the fusion module detected the increased error in the physical model and immediately increased the weight of the data-driven model to 0.65, effectively correcting the prediction bias. Ultimately, the fused prediction result matched the actual temperature change curve.

[0107] During the later stages of testing, increased environmental dust due to the arrival of summer led to dust accumulation on the transformer radiator surface, reducing heat dissipation efficiency. The fused prediction results were consistently lower than the actual measured temperature, and the prediction error exceeded the 3°C threshold. This triggered the adaptive update mechanism of the physical model. The system calculated a correction value for the thermal resistance coefficient based on this prediction error, increasing the model's equivalent thermal resistance parameter by approximately 8%. The updated physical model regenerated the prediction results, effectively correcting the systematic bias and ensuring the accuracy of the fused prediction.

[0108] Data shows that the fusion prediction method proposed in this invention exhibits high accuracy and robustness throughout the entire testing period. Its mean absolute prediction error is controlled within 1.2°C, while the average error of the single physics model prediction method in comparison is approximately 3.5°C. The adaptive update mechanism of this invention successfully identifies and corrects model mismatch issues caused by changes in equipment state, ensuring the reliability of the prediction system during operation.

[0109] In summary, the method of this invention demonstrates high accuracy and strong adaptability under various operating conditions. The final fused prediction value is closer to the actual temperature rise than any single model under stable, abrupt, and updated conditions. Under stable load conditions, the physical model has a higher weight; however, under abrupt load conditions, the weight of the data-driven model rapidly increases, dominating the prediction results, demonstrating the effectiveness of intelligent decision-making and the dynamic weight allocation mechanism. After detecting persistent prediction bias caused by dust accumulation on the radiator, the system successfully reduced the prediction error by correcting the physical model parameters. This indicates that the invention can effectively cope with slow changes in the physical characteristics of equipment, ensuring the long-term effectiveness and reliability of the prediction model and the value of the adaptive update mechanism. These data results fully demonstrate the advantages of this invention in improving the accuracy and robustness of temperature rise prediction for high-voltage electrical equipment.

[0110] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0111] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. An online prediction method of temperature rise trend of high-voltage electrical equipment, characterized in that, The method includes: The operating electrical variables and environmental variables of high-voltage electrical equipment are acquired, and the operating electrical variables and environmental variables are preprocessed to generate preprocessed data; A physical temperature rise model is constructed based on the preprocessed data, and physical prediction results are generated based on the physical temperature rise model. A data-driven prediction model is constructed based on the preprocessed data, and data-driven prediction results are generated based on the data-driven prediction model. The actual measured temperature of the high-voltage electrical equipment is obtained, and the prediction error is calculated by combining the physical prediction result with the data-driven prediction result. Based on the prediction error, a fusion weight is dynamically assigned to the physical prediction result and the data-driven prediction result. Based on the fusion weight, the physical prediction result and the data-driven prediction result are weighted and fused to generate the final temperature rise trend prediction. Determine whether the prediction error exceeds a threshold. If it does, update the data-driven model parameters by re-acquiring data and generate updated data-driven model parameters. The data-driven prediction results are regenerated based on the updated data-driven model parameters and then weighted and fused with the physical prediction results. If the prediction error exceeds a threshold, the thermodynamic parameters in the physical temperature rise model are corrected to generate updated physical model parameters. Generating updated physical model parameters includes: calculating the thermal resistance coefficient correction value based on the prediction error to generate corrected thermodynamic parameters; updating the heat balance equation in the physical temperature rise model based on the corrected thermodynamic parameters to generate an adjusted physical model; and generating updated physical model parameters based on the adjusted physical model. The updating process can be represented as follows: In the formula, is the updated equivalent thermal resistance, i.e. the generated corrected thermodynamic parameter; is the old equivalent thermal resistance used in the physical model before correction; is an update factor between 0 and 1, used to control the speed of correction, to avoid the parameter from fluctuating sharply due to transient disturbance; is the actual temperature rise value calculated according to the actual measured temperature and the ambient temperature; is the internal heating power of the device calculated according to the real-time operating electrical variables; The physical prediction results are regenerated based on the updated physical model parameters and then weighted and fused with the data-driven prediction results.

2. The online prediction method for the temperature rise trend of high-voltage electrical equipment according to claim 1, characterized in that, The generated preprocessed data includes: Obtain the current and voltage parameters as operating electrical variables, and the ambient temperature and humidity parameters as environmental variables; The current parameter, the voltage parameter, the ambient temperature parameter, and the ambient humidity parameter are filtered to generate filtered data; The filtered data is then normalized to generate preprocessed data.

3. The online prediction method for the temperature rise trend of high-voltage electrical equipment according to claim 1, characterized in that, The generated physical prediction results include: Thermodynamic parameters are extracted based on the preprocessed data; Based on the aforementioned thermodynamic parameters, a thermal equilibrium equation is constructed, and physical model parameters are generated. The physical prediction results are generated by calculating based on the physical model parameters and the preprocessed data.

4. The online prediction method for the temperature rise trend of high-voltage electrical equipment according to claim 1, characterized in that, The generated data-driven prediction results include: Acquire historical temperature data and historical operation data, and integrate the historical temperature data and historical operation data to generate a training dataset; An adaptive prediction algorithm is trained based on the training dataset to generate data-driven model parameters. The data-driven prediction results are generated by calculating the data-driven model parameters and the preprocessed data.

5. The online prediction method for the temperature rise trend of high-voltage electrical equipment according to claim 4, characterized in that, The method further includes: The broadband ultra-high frequency electromagnetic wave signal generated by the internal partial discharge of the high-voltage electrical equipment is collected by a preset ultra-high frequency partial discharge detection technology. The broadband ultra-high frequency electromagnetic wave signal is converted into a logarithmic energy characteristic signal, and non-power frequency noise is filtered out by harmonic analysis to obtain the partial discharge characteristic value; The partial discharge feature values ​​are integrated into the training dataset as additional features characterizing insulation degradation.

6. The online prediction method for the temperature rise trend of high-voltage electrical equipment according to claim 4, characterized in that, The generation of the final temperature rise trend prediction includes: The absolute difference between the physical prediction result and the actual measured temperature is calculated as the first error, and the absolute difference between the data-driven prediction result and the actual measured temperature is calculated as the second error. The first error and the second error are combined to form the prediction error. Based on the proportional relationship between the first error and the second error, dynamically assign fusion weights to the physical prediction result and the data-driven prediction result; The fusion weights are used to perform a weighted average of the physical prediction results and the data-driven prediction results to generate the final temperature rise trend prediction.

7. An online prediction system for the temperature rise trend of high-voltage electrical equipment, applied to the online prediction method for the temperature rise trend of high-voltage electrical equipment as described in any one of claims 1-6, characterized in that, The system includes: The multi-dimensional sensing and data preparation module is used to acquire the operating electrical variables and environmental variables of high-voltage electrical equipment, and to preprocess the operating electrical variables and environmental variables to generate preprocessed data. The physical temperature rise model prediction module is used to construct a physical temperature rise model based on the preprocessed data and generate physical prediction results based on the physical temperature rise model. The data-driven prediction model module is used to construct a data-driven prediction model based on the preprocessed data and generate data-driven prediction results based on the data-driven prediction model. The prediction error calculation module is used to obtain the actual measured temperature of the high-voltage electrical equipment and calculate the prediction error by combining the physical prediction result and the data-driven prediction result. The intelligent fusion temperature rise trend prediction module is used to dynamically assign fusion weights to the physical prediction result and the data-driven prediction result based on the prediction error, and to perform weighted fusion of the physical prediction result and the data-driven prediction result based on the fusion weights to generate the final temperature rise trend prediction.

Images