A transformer heat dissipation performance prediction method based on digital twinning

Abstract

The application discloses a transformer heat dissipation performance prediction method based on digital twinning, which comprises the following steps: arranging a multi-point sensor on a transformer body, collecting real-time operation data to generate original operation data; preprocessing the original operation data to obtain preprocessed operation data; inputting the preprocessed operation data into a coupling model to output initial heat dissipation performance; comparing the initial heat dissipation performance with the preprocessed operation data to output corrected heat dissipation performance; completing feature extraction and learning operation to output second heat dissipation performance; fusing the corrected heat dissipation performance with the second heat dissipation performance to generate a digital twin; calculating temperature distribution by using the digital twin to obtain complete prediction results; and comparing the complete prediction results with a preset threshold, and triggering a risk prompt information when the early warning condition is met. The application realizes transformer heat dissipation performance prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation, and in particular relates to a transformer heat dissipation performance prediction method based on digital twinning. BACKGROUND

[0002] As a key device in the power system, the stability of the operation state of the transformer is directly related to the safety and reliability of the power grid. The heat dissipation performance, as an important indicator affecting the service life and operation efficiency of the transformer, has been the focus of research and engineering practice for a long time. In the prior art, the prediction of the heat dissipation performance of the transformer mainly relies on empirical formulas and single simulation models. The empirical formula is often based on idealized conditions and can provide approximate results under steady state or specific load, but its accuracy and applicability are obviously insufficient when facing complex and variable working conditions during operation. On the other hand, although the simulation model can describe the heat flow coupling relationship inside the transformer in detail, it has a large amount of calculation and slow operation speed, which is difficult to meet the real-time prediction demand. When the load fluctuates rapidly, the environmental temperature changes dramatically, or the cooling mode is switched, these traditional methods often lag behind and cannot reflect the actual operation state of the transformer in time, resulting in a deviation between the prediction results and the true situation.

[0003] In terms of data processing, the prior art usually only performs simple denoising or averaging processing on the collected operation data, lacks in-depth mining of spatiotemporal features, and thus the dynamic correlation and nonlinear laws contained in the data cannot be effectively utilized. In addition, the traditional model lacks a self-adaptive optimization mechanism, and when the input conditions change, the model parameters cannot be corrected in time, resulting in invalid prediction results. In this case, the prediction system not only cannot provide reliable basis for the operation and maintenance personnel, but may also lead to misleading judgments.

[0004] Therefore, how to provide a transformer heat dissipation performance prediction method based on digital twinning is a problem that those skilled in the art need to solve. SUMMARY

[0005] An object of the present application is to provide a transformer heat dissipation performance prediction method based on digital twinning. The present application collects operation data through a sensor and performs preprocessing, combines a coupling model to output an initial heat dissipation performance, and then corrects the result to obtain a more accurate result. Meanwhile, a second heat dissipation performance is generated by feature extraction and learning operation, and the corrected result is fused to form a digital twin. The temperature distribution and cooling efficiency are calculated based on the digital twin to obtain a complete prediction result, which is compared with a preset threshold. When the early warning condition is reached, a risk prompt is triggered, and the accurate prediction and risk prevention and control of the transformer heat dissipation performance are realized.

[0006] According to the transformer heat dissipation performance prediction method based on digital twinning, the following steps are included:

[0007] A multi-point sensor is arranged on the transformer body to collect real-time operation data including winding temperature, insulating oil temperature, oil flow rate, load current and ambient temperature, which is recorded in sequence of sampling time to generate a raw operation data sequence;

[0008] The raw operation data sequence is subjected to denoising, standardization and missing value compensation processing to obtain a pre-processed operation data sequence;

[0009] The pre-processed operation data sequence is input into a three-dimensional multi-physical field coupling model to calculate oil temperature distribution and winding temperature rise distribution, and output an initial heat dissipation performance prediction result;

[0010] The initial heat dissipation performance prediction result is compared with the pre-processed operation data sequence, and the model parameters are adjusted according to the comparison result, and a corrected heat dissipation performance prediction result is output;

[0011] The corrected heat dissipation performance prediction result and the pre-processed operation data sequence are input, and data-driven prediction is used for feature extraction and learning operation to output a second heat dissipation performance prediction result;

[0012] The corrected heat dissipation performance prediction result and the second heat dissipation performance prediction result are fused, and a Hamiltonian force field optimization method of adaptive spatio-temporal feature mapping is called in the fusion process to perform dynamic allocation of feature weights, and a real-time updated digital twin is generated;

[0013] The real-time updated digital twin is used to calculate the change rule of winding hot spot temperature distribution and insulating oil temperature with time, and cooling efficiency is solved to obtain a complete heat dissipation performance prediction result;

[0014] The heat dissipation performance prediction result is compared with a preset threshold, and when the comparison result meets the early warning condition, a running optimization strategy output step is executed, and a risk prompt information is triggered.

[0015] Optionally, the generation of the raw operation data sequence specifically includes:

[0016] A plurality of sensors are arranged on the transformer body, including different winding temperature sensors uniformly arranged in the winding area for continuously acquiring winding temperatures at different positions, a plurality of oil temperature sensors arranged in the insulating oil flow channel for monitoring oil temperatures at different flow channel sections, a plurality of flow rate sensors arranged at key positions of the oil channel for collecting local flow rates of the cooling oil, and a plurality of ambient temperature sensors arranged on the outside of the transformer tank for collecting ambient air temperatures, each sensor is assigned a unique identification number, and a one-to-one correspondence table of the sensor and the actual installation position is established, and a multi-point sensor arrangement list with unique number and installation position information is output;

[0017] The multi-point sensor deployment list is taken as input, a unified sampling frequency and sampling period are set for constraining all sensors to collect data in the same time interval, the starting sampling time of the system is recorded, at each sampling time, a set of temperature values is collected from the winding temperature sensor, a set of oil temperature values is collected from the oil temperature sensor, a set of oil flow rate values is collected from the flow rate sensor, the instantaneous value of the load current and the real-time value of the ambient temperature are obtained at the same time, the data collected at each time is formed into a complete sampling record, and an accurate time stamp is attached to the record, and a collection of multiple types of original sampling records arranged in chronological order is output;

[0018] Based on the collection of multiple types of original sampling records, the continuity of the sampling period is checked in chronological order, and missing and abnormal data are marked to generate an original running data sequence arranged in chronological order and attached with abnormality marks.

[0019] Optionally, the obtaining process of the pre-processed running data sequence specifically includes:

[0020] The original running data sequence is taken as input, the winding temperature, the insulating oil temperature, the oil flow rate, the load current and the ambient temperature original values in each record are read in chronological order, noise removal operations are performed on each sensor channel, including eliminating pulse-type abnormal points by using a fixed-length sliding window median method, weakening high-frequency jitter by using a low-pass smoothing method, performing consistency verification on detected long-time constant segments and retaining trend information, and a denoised running data sequence arranged in chronological order is output;

[0021] The denoised running data sequence is taken as input, standardization processing is performed on each sensor channel, including unit unification and value conversion according to the measurement unit and dimension of the channel, then shifting with the average level of the current channel in the current batch as the center, and scaling with the dispersion level of the current channel in the current batch as the scale, so that the data of all channels are transformed to a comparable dimensionless range, while the corresponding time stamp and channel identifier are retained, and a standardized running data sequence is output;

[0022] The standardized running data sequence is taken as input, the missing interval between adjacent time stamps of each channel is located, linear interpolation is performed between the front and rear two valid data according to the time proportion of the missing point located on both sides, the nearest valid data is used for forward or backward filling for the missing point at the beginning or end of the sequence, after completing the compensation of all missing positions, the sequence consistency is checked again according to the time stamp and each record is reconstructed, and a pre-processed running data sequence is output.

[0023] Optionally, the output process of the initial heat dissipation performance prediction result specifically includes:

[0024] reading a pre-processing operation data sequence, setting an ambient temperature as an external convection reference temperature, mapping a winding temperature as a solid region initial temperature field, mapping an insulating oil temperature as a fluid region initial temperature field, establishing a three-dimensional calculation domain containing the solid region and the fluid region, defining boundary conditions of the solid and the fluid, and outputting an initialized three-dimensional multi-physical field coupling calculation setting;

[0025] establishing a fluid mass conservation relationship, a momentum conservation relationship and an energy conservation relationship in the fluid region, introducing a buoyancy term caused by a temperature difference in the momentum conservation relationship to represent a natural convection effect, using the energy conservation relationship to describe a space-time variation of a fluid temperature field, establishing a heat conduction energy conservation relationship in the solid region, distributing a copper loss power corresponding to a load current as a heat source according to a volume, setting an iron loss as a fixed heat source, applying a no-slip velocity condition and a temperature and heat flow continuous condition on a solid-fluid contact boundary, and outputting an iterative solution format of a coupling equation group;

[0026] using the iterative solution format as an input, performing a coupling calculation on the solid region and the fluid region in a time stepping manner, solving a fluid velocity field and a pressure field in each time step, updating a fluid temperature field, and calculating a solid temperature field, continuously exchanging heat flow and temperature on a solid-fluid interface until an interface temperature difference and an interface heat flow difference are less than a preset threshold, while detecting that an equation residual satisfies a convergence condition, and outputting a three-dimensional oil temperature distribution and a three-dimensional winding temperature distribution at a current time step;

[0027] extracting a highest temperature of a winding region as a winding hot spot temperature through the three-dimensional oil temperature distribution and the three-dimensional winding temperature distribution, calculating an average temperature difference of an oil path inlet and an outlet as an oil temperature rise, combining a load current and an equivalent resistance to obtain a copper loss power, adding the copper loss power and an iron loss power to obtain a total loss power, multiplying an oil path mass flow, an oil specific heat and a temperature difference to obtain a heat power cooled away, defining a cooling efficiency as a ratio of the heat power cooled away to the total loss power, and outputting an initial heat dissipation performance prediction result containing the winding hot spot temperature, the oil temperature rise, the oil temperature distribution, the winding temperature distribution and the cooling efficiency.

[0028] Optionally, the outputting process of the corrected heat dissipation performance prediction result specifically includes:

[0029] comparing the winding hot spot temperature, the oil temperature rise and the cooling efficiency in the initial heat dissipation performance prediction result with a cooling efficiency calculated from a measured winding temperature, a measured oil temperature and a measured load current at a corresponding time in the pre-processing operation data sequence item by item to form an error sequence, wherein an error is defined as a difference between a predicted value and a measured value;

[0030] The overall error metric is calculated using the error sequence, the mean square error is obtained by averaging all error squares, the mean absolute error is obtained by averaging the absolute values of all errors, and the trend of the error metric is tracked, and the parameter adjustment operation is triggered when it is detected that the error metric does not decrease;

[0031] The parameter adjustment operation is the core, the trend of the error metric is taken as the input, and the key parameters in the three-dimensional multi-physical field coupling model are updated step by step, including the thermal conductivity of the solid region, the viscosity of the fluid region, the specific heat capacity, and the boundary heat exchange coefficient. After each parameter update, the prediction result is recalculated and a new error sequence is formed, until the error metric remains stable and decreases in the continuous iteration process, and the corrected heat dissipation performance prediction result is output.

[0032] Optionally, the output process of the second heat dissipation performance prediction result specifically includes:

[0033] Based on the corrected heat dissipation performance prediction result and the preprocessed running data sequence, the preprocessed record and the corrected result at the same time are selected, the feature basis vector is constructed, the winding temperature preprocessing vector, the insulating oil temperature preprocessing vector and the oil flow rate preprocessing vector are calculated in space respectively arithmetic average, maximum and minimum and connected in fixed order, recorded as the basis statistical feature of the current time, the load current preprocessing value and the environmental temperature preprocessing value are connected to the basis statistical feature according to the original value, the corrected winding hot spot temperature, the corrected oil temperature rise and the corrected cooling efficiency are connected to the basis statistical feature according to the original value, the time window length is set and the basis statistical features of continuous time are collected in time sequence to form a time series feature vector;

[0034] The parameter vector of the data-driven prediction model is set and the model output is set to the estimated value of the three-dimensional target quantity, corresponding to the estimated value of the winding hot spot temperature, the estimated value of the oil temperature rise and the estimated value of the cooling efficiency. The time series feature vector is selected as a batch training set, the reference target quantity of the current batch is set as the corrected winding hot spot temperature, the corrected oil temperature rise and the corrected cooling efficiency, the batch loss function is calculated, wherein the loss function is composed of three parts of linear weighted sum, respectively, the mean square error of the winding hot spot temperature, the mean absolute error of the oil temperature rise and the mean square error of the cooling efficiency. Perform batch update until the set condition is met, the set condition is that the absolute value of the difference between adjacent two loss values is less than the preset convergence difference value or the update times reach the set maximum update times, and the updated parameter vector is output;

[0035] Based on the updated parameter vector and the time series feature vector, the prediction values of the winding hot spot temperature, the oil temperature rise and the cooling efficiency are obtained by parameter operation and feature mapping of the data-driven prediction model, and are output as the second heat dissipation performance prediction result.

[0036] Optionally, the process of generating the real-time updated digital twin specifically includes:

[0037] Based on the corrected heat dissipation performance prediction results and the second heat dissipation performance prediction results, the two types of predicted values ​​of winding hot spot temperature, oil temperature rise and cooling efficiency are paired item by item under the same time index. The three-dimensional residual vector is calculated and combined in the order of physical model results minus data-driven results. The corresponding spatial location and time label are attached to each residual sample, and the residual sequence with spatial-time labels is obtained by sorting them in chronological order.

[0038] To perform adaptive spatiotemporal feature mapping on residual sequences with spatial-temporal labels, a fractional-order spatiotemporal kernel function is used for weighted integration. The kernel function value is calculated for each sample label in the sequence and the current label. The kernel function value is used as a weight to accumulate the corresponding residual vectors to obtain the spatiotemporal mapping feature vector at the current time.

[0039] A Hamiltonian force field optimization model is constructed using empty mapping eigenvectors. The feature weights are set as generalized coordinates, and the corresponding conjugate momentum is recorded separately. The total energy is defined as the sum of kinetic energy and potential energy. The kinetic energy is determined by the time-varying mass matrix and the conjugate momentum, while the potential energy is determined by the quadratic metric of the mapping features and the graph regularization term constructed based on the spatiotemporal adjacency relationship. The leapfrog scheme is used for discrete evolution. The conjugate momentum is updated half-step according to the potential energy gradient of the current coordinate. At the same time, the coordinates are updated in full step using the updated conjugate momentum. The conjugate momentum is updated half-step according to the potential energy gradient of the new coordinate. After each discrete step, the change in total energy and the norm of the two adjacent coordinate increments are monitored. When the number of evolution steps reaches the set upper limit, the evolution is terminated, and the dynamic feature weights at the current moment are output.

[0040] The dynamic feature weights are projected onto the fusion coefficient space corresponding to the three indicators through a fixed weight mapping matrix. At the same time, a bounded monotonic compression function is applied to the projection result to obtain a three-dimensional gating vector. Each component of the gating vector is used as the mixing ratio to combine each indicator between the two types of predicted values ​​to form a fusion result. The real-time updated digital twin composed of the three fusion results is output according to the time index.

[0041] Optionally, the process of obtaining the complete heat dissipation performance prediction result specifically includes:

[0042] Extract the temperature data and operating condition information at the current sampling moment to form a dataset for heat dissipation calculation;

[0043] The dataset is input into a real-time updated digital twin to calculate the temperature distribution at each location of the winding, and the maximum temperature value is determined in the temperature distribution to obtain the winding hot spot temperature and spatial distribution, and output the hot spot prediction result.

[0044] On the same data set, the average temperature of the insulating oil is calculated, the temperature difference of adjacent sampling time is recorded, the change rule of the oil temperature with time is obtained, and the oil temperature rise is determined by the difference between the outlet temperature and the inlet temperature, and the oil temperature prediction result is output;

[0045] The hotspot prediction result and the oil temperature prediction result are taken as inputs, energy balance calculation is carried out in combination with the total loss power, the ratio of the heat taken away by cooling to the loss is obtained, the current ratio is taken as the cooling efficiency, and the complete heat dissipation performance prediction result composed of the winding hotspot temperature, the oil temperature change rule and the cooling efficiency is output.

[0046] Optionally, the triggering of the risk prompt information specifically comprises:

[0047] The three prediction indexes of each moment of the complete heat dissipation performance prediction result are extracted in chronological order, the three prediction indexes are the winding hotspot temperature, the average value of the insulating oil temperature and the change rule thereof with time, and the cooling efficiency, and the three prediction indexes at the same moment are arranged as a record, and all the records are sequentially composed into a prediction result sequence;

[0048] Each record of the prediction result sequence is compared with a preset threshold set in each item, the preset threshold set comprises an upper limit threshold of the winding hotspot temperature, an upper limit threshold of the average value of the insulating oil temperature, and a lower limit threshold of the cooling efficiency, a comparison conclusion is given for each of the three prediction indexes at each moment, the comparison conclusion is divided into non-exceeding and exceeding, and a discrimination record containing the three comparison conclusions is output in chronological order;

[0049] Based on the discrimination record output in chronological order, each moment is judged, when any comparison conclusion is exceeding, it is marked that the current moment warning condition is established, otherwise it is marked that the current moment warning condition is not established, the specific index name leading to the warning is summarized, and the warning judgment result and the cause information are output;

[0050] When the warning condition is established, a corresponding strategy is selected from the operation optimization strategy library according to the cause information, and an operation optimization instruction is generated, including load distribution adjustment, cooling mode switching and operation time sequence adjustment, and a risk prompt information facing a monitoring system is generated.

[0051] The beneficial effects of the present application are:

[0052] 1、The present application significantly improves the prediction accuracy and real-time performance of the heat dissipation performance of the transformer by introducing the combination of digital twin technology and real-time sensor data, compared with the traditional prediction method relying on empirical formula or single simulation model, the present application can collect data of multiple key parameters including winding temperature, oil temperature, oil flow rate, load current, etc. in real time, through the fusion of three-dimensional multi-physical field coupling model and data driven algorithm, the heat dissipation state of the transformer under different working conditions can be more accurately reflected.

[0053] 2. The adaptive spatiotemporal feature mapping and Hamiltonian force field optimization algorithm adopted in this invention solve the feature extraction problem under multidimensional data interaction and realize dynamic adjustment and optimization under different load and environmental conditions. This innovation not only improves the accuracy of heat dissipation performance prediction, but also enhances the model's adaptability to complex operating states and changing environments, avoiding the problem of inaccurate prediction of non-steady-state conditions by traditional models.

[0054] 3. In the prediction process, the present invention achieves dual correction of heat dissipation performance and real-time operating data. By comparing the model output with the actual measurement data and adjusting the dynamic parameters, the reliability and accuracy of the prediction results are further improved. At the same time, the generation of the second heat dissipation performance prediction result is also optimized by the data-driven approach, which further enhances the adaptability to various heat dissipation conditions and makes the prediction results closer to the actual operating state of the transformer.

[0055] 4. This invention achieves dynamic monitoring and prediction of various heat dissipation performance indicators of transformers through real-time updated digital twins, and can provide risk warnings based on preset thresholds. This warning mechanism not only improves the safety of transformer operation but also provides decision support for maintenance personnel, enabling them to promptly identify potential overheating risks and take corresponding countermeasures, thereby significantly reducing the incidence of equipment failures and maintenance costs. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0057] Fig. 1 This is a flowchart of a transformer heat dissipation performance prediction method based on digital twin proposed in this invention;

[0058] Fig. 2 This is a schematic diagram of a transformer heat dissipation performance prediction method based on digital twin proposed in this invention. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0060] refer to Figs. 1-2 A method for predicting the heat dissipation performance of transformers based on digital twins includes the following steps:

[0061] The multi-point sensor is arranged on the transformer body to collect real-time operation data including winding temperature, insulating oil temperature, oil flow rate, load current and ambient temperature, and the operation data is recorded in sequence according to sampling time to generate an original operation data sequence;

[0062] The original operation data sequence is subjected to denoising, standardization and missing value compensation processing to obtain a preprocessed operation data sequence;

[0063] The preprocessed operation data sequence is input into a three-dimensional multi-physical field coupling model to calculate oil temperature distribution and winding temperature rise distribution, and output an initial heat dissipation performance prediction result;

[0064] The initial heat dissipation performance prediction result is compared with the preprocessed operation data sequence, and the model parameters are adjusted according to the comparison result to output a corrected heat dissipation performance prediction result;

[0065] The corrected heat dissipation performance prediction result and the preprocessed operation data sequence are input to perform feature extraction and learning operation by data-driven prediction to output a second heat dissipation performance prediction result;

[0066] The corrected heat dissipation performance prediction result and the second heat dissipation performance prediction result are fused, and a Hamiltonian force field optimization method of adaptive spatio-temporal feature mapping is called in the fusion process to perform dynamic allocation of feature weights to generate a real-time updated digital twin;

[0067] The real-time updated digital twin is used to calculate the change rule of winding hot spot temperature distribution and insulating oil temperature with time, and to perform cooling efficiency solving to obtain a complete heat dissipation performance prediction result;

[0068] The heat dissipation performance prediction result is compared with a preset threshold, and when the comparison result meets the early warning condition, a running optimization strategy output step is performed, and a risk prompt information is triggered.

[0069] The present application realizes correction and adaptive fusion by real-time collection and data preprocessing of the multi-point sensor, combines a three-dimensional multi-physical field coupling model and data-driven prediction, constructs a dynamically updated digital twin, can accurately represent winding hot spot temperature, oil temperature distribution and cooling efficiency under multiple working conditions, and significantly improves the accuracy and real-time performance of heat dissipation performance prediction.

[0070] In the embodiment, the generation of the original operation data sequence specifically includes:

[0071] A plurality of sensors are arranged on the transformer body, including different winding temperature sensors arranged uniformly in the winding area for continuously acquiring winding temperatures at different positions, a plurality of oil temperature sensors arranged in the insulation oil flow channel for monitoring oil temperatures at different flow channel sections, a plurality of flow rate sensors arranged at key positions of the oil channel for collecting local flow rates of the cooling oil, and a plurality of ambient temperature sensors arranged on the outside of the transformer tank for collecting ambient air temperatures, each sensor is assigned a unique identification number, and a one-to-one correspondence table of the sensors and actual installation positions is established, and a multi-point sensor arrangement list with unique numbers and installation position information is output;

[0072] The multi-point sensor arrangement list is taken as input, a unified sampling frequency and sampling period are set for constraining all sensors to collect data within the same time interval, the starting sampling time of the system is recorded, at each sampling time, a group of temperature values is collected from the winding temperature sensors, a group of oil temperature values is collected from the oil temperature sensors, a group of oil flow rate values is collected from the flow rate sensors, the instantaneous value of the load current and the real-time value of the ambient temperature are simultaneously acquired, the data collected at each time is formed into a complete sampling record, and an accurate time stamp is attached in the record, and a multi-type original sampling record set arranged in time sequence is output;

[0073] Based on the multi-type original sampling record set, the original operation data sequence arranged in time sequence and attached with abnormality markers is generated by arranging according to the time stamp order, checking the continuity of the sampling period, and marking the missing and abnormal data.

[0074] The present application ensures the consistency of multi-source data in space and time by reasonably arranging a plurality of sensors on the transformer body and its periphery, establishing a sensor list with unique numbers and position mapping, and combining a unified sampling frequency and strict time stamp management, so that the winding temperature, oil temperature, flow rate, load current and ambient temperature are comprehensively and continuously monitored, and the original operation data sequence with integrity and traceability is generated.

[0075] In the embodiment, the obtaining process of the preprocessed operation data sequence specifically includes:

[0076] The original values of the winding temperature, insulation oil temperature, oil flow rate, load current and ambient temperature in each record are read in time stamp order, noise removal operations are performed on each sensor channel, including eliminating pulse-type abnormal points by using a fixed-length sliding window median method, weakening high-frequency jitter by using a low-pass smoothing method, performing consistency verification on the detected long-time constant section and retaining trend information, and an operation data sequence after noise removal arranged in time sequence is output;

[0077] Taking the denoised running data sequence as input, standardization processing is performed on each sensor channel. This includes unifying the units and converting the values ​​according to the measurement units and dimensions of the channels, then shifting the current channel to the average level in the current batch as the center, and then scaling the current channel to the discrete level in the current batch as the scale. This transforms the data of all channels to a comparable dimensionless range, while retaining the corresponding timestamps and channel identifiers, and outputting the standardized running data sequence.

[0078] Using the standardized running data sequence as input, the missing intervals between adjacent timestamps of each channel are located. For missing points with valid data on both sides, linear interpolation is performed between the two valid data points according to the time ratio. For missing points at the beginning or end of the sequence, forward or backward filling is performed using the most recent valid data. After completing the compensation for all missing positions, the order consistency is checked again according to the timestamp and each record is reconstructed. The preprocessed running data sequence is then output.

[0079] This invention significantly improves the accuracy and consistency of multi-source sensor data by performing noise removal, standardization, and missing data compensation on the operational data sequence. This method effectively eliminates pulse anomalies and high-frequency interference, avoids information bias caused by constant value distortion, and simultaneously unifies the units of measurement and maintains the comparability of each channel, ensuring the reliability of subsequent modeling input data. The data sequence after complete preprocessing exhibits continuity and stability, providing high-quality support for heat dissipation performance prediction and digital twin construction.

[0080] In this embodiment, the output process of the initial heat dissipation performance prediction result specifically includes:

[0081] Read the preprocessed running data sequence, set the ambient temperature as the external convection reference temperature, map the winding temperature to the initial temperature field of the solid region, map the insulating oil temperature to the initial temperature field of the fluid region, establish a three-dimensional computational domain containing the solid region and the fluid region, define the boundary conditions of the solid and the fluid, and output the initialized three-dimensional multiphysics coupling calculation settings.

[0082] In the fluid region, fluid mass conservation, momentum conservation, and energy conservation relations are established. The momentum conservation relation introduces the buoyancy force term generated by the temperature difference to characterize the natural convection effect. The energy conservation relation is used to describe the spatiotemporal variation of the fluid temperature field. In the solid region, heat conduction energy conservation relations are established, and the copper loss power corresponding to the load current is allocated as a heat source according to volume. The iron loss is set as a fixed heat source. No-slip velocity condition and temperature and heat flow continuity condition are applied at the solid-fluid contact boundary. The iterative solution format of the coupled equation set is output.

[0083] With the iterative solution format as input, the solid region and the fluid region are coupled and calculated in a time-stepping manner, in each time step, the fluid velocity field and the pressure field are solved first, the fluid temperature field is updated, and the solid temperature field is calculated, and the heat flow and the temperature are continuously exchanged at the interface between the solid and the fluid, until the interface temperature difference and the interface heat flow difference are less than a preset threshold, while detecting that the equation residual satisfies the convergence condition, and outputting the three-dimensional oil temperature distribution and the three-dimensional winding temperature distribution at the current time step;

[0084] The highest temperature of the winding region is extracted as the winding hot spot temperature through the three-dimensional oil temperature distribution and the three-dimensional winding temperature distribution, the average temperature difference between the oil inlet and the outlet is calculated as the oil temperature rise, the copper loss power is obtained combined with the load current and the equivalent resistance, and the total loss power is obtained by adding the iron loss power, the heat power taken away by cooling is obtained by multiplying the oil mass flow, the specific heat of the oil and the temperature difference, the cooling efficiency is defined by the ratio of the heat power taken away by cooling to the total loss power, and the initial heat dissipation performance prediction result including the winding hot spot temperature, the oil temperature rise, the oil temperature distribution, the winding temperature distribution and the cooling efficiency is output.

[0085] The present application is based on multi-physical field coupling modeling, and comprehensively considers the effects of solid heat conduction, fluid convection and electromagnetic loss heat source, can truly reproduce the oil temperature distribution and winding temperature distribution inside the transformer, and through coupling solving the conservation equations of fluid and solid regions, realizes the dynamic exchange of heat flow and temperature, obtains high-precision winding hot spot temperature, oil temperature rise and cooling efficiency. The method effectively improves the physical consistency and reliability of the initial heat dissipation performance prediction, and provides a solid foundation for subsequent digital twin correction and optimization.

[0086] In the embodiment, the output process of the corrected heat dissipation performance prediction result specifically includes:

[0087] The winding hot spot temperature, the oil temperature rise and the cooling efficiency in the initial heat dissipation performance prediction result are compared with the cooling efficiency calculated by the measured winding temperature, the measured oil temperature and the measured load current at the corresponding time in the preprocessed operation data sequence, to form an error sequence, wherein the error is defined as the difference between the predicted value and the measured value;

[0088] The overall error measure is calculated using the error sequence, the mean square error is obtained by taking the average of the squares of all errors, the mean absolute error is obtained by taking the average of the absolute values of all errors, and the change trend of the error measure is tracked, and the parameter adjustment operation is triggered when it is detected that the error measure does not decrease;

[0089] The parameter adjustment operation is taken as the core, the change trend of the error metric is taken as the input, and key parameters in the three-dimensional multi-physical field coupling model are updated step by step, including the thermal conductivity of the solid region, the viscosity of the fluid region, the specific heat capacity, and the boundary heat exchange coefficient; after each time of updating the parameters, the prediction result is recalculated, and a new error sequence is formed again, until the error metric remains stable and decreases in the continuous iteration process, and the corrected heat dissipation performance prediction result is output.

[0090] The application introduces an error metric driven parameter adaptive updating mechanism, dynamically corrects the key parameters of the multi-physical field coupling model, effectively reduces the deviation between the prediction value and the measured value, and through the joint measurement of the mean square error and the mean absolute error, can sensitively capture the model performance change and timely trigger the adjustment, so that the corrected prediction result remains continuously convergent and the accuracy is improved under different working conditions, thereby enhancing the reliability and stability of the heat dissipation performance prediction, and providing protection for the safety of equipment operation.

[0091] In the embodiment, the output process of the second heat dissipation performance prediction result specifically includes:

[0092] Based on the corrected heat dissipation performance prediction result and the preprocessed running data sequence, the preprocessed record and the correction result at the same time are selected, a feature basis vector is constructed, the winding temperature preprocessing vector, the insulating oil temperature preprocessing vector and the oil flow rate preprocessing vector are respectively calculated in space arithmetic average, maximum value and minimum value and connected in fixed order, recorded as the basis statistical feature of the current time, the load current preprocessing value and the environmental temperature preprocessing value are connected to the basis statistical feature according to the original value, the corrected winding hot spot temperature, the corrected oil temperature rise and the corrected cooling efficiency are connected to the basis statistical feature according to the original value, the length of the time window is set, and the basis statistical features of continuous time are collected in time sequence to form a time sequence feature vector by connecting them in order;

[0093] The parameter vector of the data-driven prediction model is set, and the model output is set to be the estimated value of the three-dimensional target quantity, which respectively corresponds to the estimated value of the winding hot spot temperature, the estimated value of the oil temperature rise and the estimated value of the cooling efficiency, the time sequence feature vector is selected as a batch training set, the reference target quantity of the current batch is set to be the corrected winding hot spot temperature, the corrected oil temperature rise and the corrected cooling efficiency, the batch loss function is calculated, wherein the loss function is composed of three parts of linear weighted summation, which are the mean square error of the winding hot spot temperature, the mean absolute error of the oil temperature rise and the mean square error of the cooling efficiency, and the batch update is executed until the set condition is met, the set condition is that the absolute value of the difference between adjacent two loss values is less than the preset convergence difference value or the update times reach the set maximum update times, and the updated parameter vector is output.

[0094] Based on the updated parameter vector and the time sequence feature vector, the parameter operation and feature mapping of the data-driven prediction model are performed to obtain the predicted values of the winding hot spot temperature, oil temperature rise and cooling efficiency, and output as the second heat dissipation performance prediction result.

[0095] The winding hot spot temperature, oil temperature rise and cooling efficiency are predicted with high precision by constructing the time sequence feature vector containing the multidimensional statistics and the environment, load and correction results, and performing adaptive training in the data-driven model combined with the multi-index weighted loss.

[0096] In the embodiment, the generation process of the real-time updated digital twin specifically includes:

[0097] Based on the corrected heat dissipation performance prediction result and the second heat dissipation performance prediction result, the two types of predicted values of the winding hot spot temperature, oil temperature rise and cooling efficiency are paired under the same time index, and a three-dimensional residual vector is calculated and combined in the order of physical model result minus data-driven result, and the corresponding spatial position and time label are added to each residual sample, and a residual sequence with space-time label is obtained by arranging in time sequence;

[0098] The residual sequence with space-time label is subjected to adaptive space-time feature mapping, and a fractional order space-time kernel function is used for weighted integration:

[0099] ;

[0100] wherein, denotes the fractional order space-time kernel function, and denote the space-time label of the current residual sample and the space-time label of another residual sample compared with the current residual sample, denotes the sum of squares of differences in spatial position and time stamp, denotes the fractional order, denotes the time decay coefficient, denotes the gamma function, denotes the current time, denotes the exponential function with natural base, denotes the integral variable, representing the potential space-time scale, is used to introduce the fractional order effect, so that the kernel function has memory in multiple scales, The dynamic attenuation characteristic evolving over time is embodied, the kernel function value is calculated for each sample label in the sequence and the current time label, the kernel function value is used as a weight to weight and accumulate the corresponding residual vector to obtain a space-time mapping feature vector at the current time;

[0101] A Hamilton force field optimization model is constructed by the space-time mapping feature vector, the feature weight is set as a generalized coordinate, the corresponding conjugate momentum is recorded separately, and the total energy is defined as the sum of kinetic energy and potential energy, wherein the kinetic energy is determined by the time-varying mass matrix and the conjugate momentum, the potential energy is determined by the quadratic metric of the mapping feature and the graph regularization term constructed based on the space-time adjacency relationship, the leapfrog format is used for discrete evolution, the conjugate momentum is updated by half step according to the potential energy gradient of the current coordinate, the coordinate is updated by whole step using the updated conjugate momentum, and the conjugate momentum is updated by half step according to the potential energy gradient of the new coordinate, and the change amplitude of the total energy and the norm of the coordinate increment of the adjacent two times are monitored after each discrete step, the evolution is terminated when the evolution step number reaches the set upper limit, and the dynamic feature weight at the current time is output.

[0102] The dynamic feature weight is projected into the fusion coefficient space corresponding to the three indicators through a fixed weight mapping matrix, a bounded and monotonic compression function is applied to the projection result to obtain a three-dimensional gating vector, and the components of the gating vector are used as mixing proportions to combine each indicator between the two prediction values to form a fusion result, and the real-time updated digital twin composed of the three fusion results is output according to the time index.

[0103] The application realizes multi-scale weighted mapping of the residual sequence through fractional order space-time kernel function, introduces memory and dynamic attenuation characteristics, so that the residual information under different working conditions is fully retained and expressed, combines Hamilton force field optimization to maintain energy consistency, ensures the stability and controllability of the feature weight evolution process, and finally realizes adaptive proportional combination of the physical model and the data-driven result by using the gating vector in the fusion stage, which not only significantly improves the accuracy and robustness of hot spot temperature, oil temperature rise and cooling efficiency prediction, but also enhances the real-time updating ability and multi-working condition adaptability of the digital twin.

[0104] In the embodiment, the process of obtaining the complete heat dissipation performance prediction result specifically includes:

[0105] Extracting temperature data and running condition information at the current sampling time to form a data set for heat dissipation calculation;

[0106] Inputting the data set into the real-time updated digital twin, calculating the temperature distribution of each position of the winding, determining the maximum temperature in the temperature distribution, obtaining the hot spot temperature and spatial distribution of the winding, and outputting the hot spot prediction result;

[0107] On the same data set, the average temperature of the insulating oil is calculated, the temperature difference between adjacent sampling time points is recorded, the change rule of the oil temperature with time is obtained, and the oil temperature rise is determined by the difference between the outlet temperature and the inlet temperature, and the oil temperature prediction result is output;

[0108] The hotspot prediction result and the oil temperature prediction result are taken as inputs, energy balance calculation is performed in combination with the total loss power, the ratio of the heat taken away by cooling to the loss is obtained, the current ratio is taken as the cooling efficiency, and a complete heat dissipation performance prediction result composed of the winding hotspot temperature, the oil temperature change rule and the cooling efficiency is output.

[0109] The present application can accurately calculate the winding temperature distribution, oil temperature change and cooling efficiency of each position of the transformer through real-time data input and dynamic updating of the digital twin, predict the cooling efficiency according to the winding hotspot temperature and the oil temperature rise, and perform energy balance in combination with the total loss power, so as to ensure that the heat dissipation performance of the transformer is comprehensively evaluated; through real-time monitoring and optimization of the temperature distribution and the cooling efficiency, the heat dissipation capacity of the transformer is significantly improved, the risk of overheating is reduced, and the safety and operating efficiency of the transformer are improved.

[0110] In the embodiment, the triggering of the risk prompt information specifically includes:

[0111] The three prediction indexes of each time point of the complete heat dissipation performance prediction result are extracted in chronological order, the three prediction indexes are the winding hotspot temperature, the average value of the insulating oil temperature and its change rule with time, and the cooling efficiency, and the three prediction indexes at the same time point are arranged as a record, and all records are arranged in chronological order to form a prediction result sequence;

[0112] Each record of the prediction result sequence is compared with a preset threshold set in each item, the preset threshold set includes an upper limit threshold of the winding hotspot temperature, an upper limit threshold of the average value of the insulating oil temperature, and a lower limit threshold of the cooling efficiency, a comparison conclusion is given for each of the three prediction indexes at each time point, the comparison conclusion is divided into non-exceeding and exceeding, and a discrimination record containing the three comparison conclusions is output in chronological order;

[0113] Based on the discrimination record output in chronological order, each time point is judged, when any comparison conclusion is exceeding, it is marked that the current time point warning condition is established, otherwise it is marked that the current time point warning condition is not established, and the specific index name leading to the warning is summarized and the warning judgment result and the cause information are output;

[0114] When the warning condition is established, a corresponding strategy is selected from the operation optimization strategy library according to the cause information, and an operation optimization instruction is generated, including load distribution adjustment, cooling mode switching and operation time sequence adjustment, and a risk prompt information facing the monitoring system is generated.

[0115] The application can accurately judge the abnormal change of the heat dissipation performance of the transformer by monitoring and comparing key indicators such as winding hot spot temperature, insulating oil temperature and cooling efficiency in real time, and timely triggering a warning when exceeding the preset threshold, and through dynamic updating of the warning condition and cause information, the system can automatically adjust the operation strategy, such as load distribution, cooling mode or operation timing, to ensure that the transformer operates efficiently within a safe range.

[0116] Embodiment 1

[0117] In order to verify the feasibility of the application in implementation, the application is applied to the operation monitoring and heat dissipation performance prediction of a typical large-capacity oil-immersed power transformer. The load of this type of transformer fluctuates frequently in actual power grid operation, and the cooling system is significantly affected by environmental temperature. The traditional heat performance prediction method based on fixed empirical parameters is prone to problems such as insufficient prediction accuracy, response lag and untimely risk identification when facing complex operating conditions. The application forms a real-time updateable digital twin by deploying multiple types of sensors, constructing a three-dimensional multi-physical field coupling model, combining data-driven prediction and adaptive fusion algorithm, effectively solving the shortcomings of existing technology in heat dissipation performance prediction.

[0118] In this scenario, first, the winding temperature, oil temperature, flow rate, environmental temperature and load current data are collected by multiple point sensors, and the original data are denoised, standardized and missing compensated to make them smoother and more consistent with actual operating conditions. The preprocessed data are input into the three-dimensional multi-physical field coupling model to obtain the oil temperature distribution and winding temperature distribution. There is a certain deviation between the initial prediction and the measured data, and the thermal conductivity, viscosity and heat transfer coefficient are dynamically updated through the error-driven parameter correction mechanism, the prediction accuracy is obviously improved, further the correction results and preprocessed data are used to train the data-driven model, the time series features are extracted and optimized by combining the multi-index loss function, the prediction error of the model under independent operating conditions is reduced to within 2%, finally the real-time updateable digital twin is constructed by adaptive spatiotemporal feature mapping and Hamiltonian force field optimization fusion of physical and data-driven results, realizing second-level prediction update, issuing overheat risk warning 15 minutes in advance and providing cooling optimization strategy.

[0119] Table 1 Comparison of the effect of the application and the traditional method in heat dissipation performance prediction

[0120]

[0121] From table 1, it can be seen that the application is superior to the traditional experience model in various performance indicators. The average error of winding hot spot temperature is reduced from 4.2 DEG C to 1.4 DEG C after correction, and is only 1.1 DEG C after further fusion. The average error of oil temperature rise is reduced from 0.9 DEG C to 0.3 DEG C, and is further reduced to 0.2 DEG C after fusion. The cooling efficiency error is reduced from 4.0% to 1.2%, and is only 0.9% after fusion. In terms of real-time, the prediction update delay is shortened from 12 seconds to 3 seconds, and the fusion prediction only needs 1 second. The early warning advance is improved from 2 minutes to 10 minutes, and the fusion method can prompt 15 minutes in advance. At the same time, the abnormal detection accuracy is improved from 82% to 91%, and reaches 95% after fusion. The optimization strategy realizes an 8% reduction in energy consumption, fully proving the significant advantages of the application in precision, real-time and risk prevention and control.

[0122] The above is only a preferred embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and the inventive concept of the application within the technical range disclosed by the application, which should be covered in the protection scope of the application.

Claims

1. A transformer heat dissipation performance prediction method based on digital twinning, characterized in that, The method comprises the following steps: A multi-point sensor is arranged on the transformer body to collect real-time operation data and generate an original operation data sequence; The original operation data sequence is standardized to obtain a pre-processed operation data sequence; The pre-processed operation data sequence is input into a three-dimensional multi-physical field coupling model to calculate oil temperature distribution and winding temperature rise distribution, and output an initial heat dissipation performance prediction result; The initial heat dissipation performance prediction result is compared with the pre-processed operation data sequence, and the model parameters are adjusted according to the comparison result, and a corrected heat dissipation performance prediction result is output; The corrected heat dissipation performance prediction result and the second heat dissipation performance prediction result are fused, and a Hamiltonian force field optimization of adaptive space-time feature mapping is called to generate a real-time updated digital twin; The winding hot spot temperature distribution and the insulating oil temperature variation law are calculated by using the real-time updated digital twin, and the cooling efficiency is solved to obtain a complete heat dissipation performance prediction result; The heat dissipation performance prediction result is compared with a preset threshold value, an operation optimization strategy output step is executed, and a risk prompt information is triggered. The generation of the original operation data sequence specifically comprises:

2. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, A plurality of sensors are arranged on the transformer body, each sensor is assigned a unique identification number, and a one-to-one correspondence table of the sensors and the actual installation positions is established to output a multi-point sensor arrangement list; The multi-point sensor arrangement list is input, a unified sampling frequency and sampling period are set, the starting sampling time of the system is recorded, the data collected at each time is constituted into a complete sampling record, and an accurate time stamp is attached in the record, and a plurality of original sampling record sets are output; Based on the plurality of original sampling record sets, the sampling period continuity is checked in the order of time stamps, missing and abnormal data are marked, and an original operation data sequence is generated. The process of obtaining the pre-processed operation data sequence specifically comprises:

3. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The original operation data sequence is input, the original values in each record are read in the order of time stamps, noise removal operations are performed on each sensor channel respectively, and a denoised operation data sequence is output; The denoised operation data sequence is input, standardization processing is performed on each sensor channel respectively, and the corresponding time stamps and channel identifications are retained, and a standardized operation data sequence is output; The standardized operation data sequence is input, the missing intervals of each channel between adjacent time stamps are located, the missing positions are compensated, the time stamp order consistency is checked again, and each record is reconstructed, and a pre-processed operation data sequence is output. The output process of the initial heat dissipation performance prediction result specifically comprises:

4. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The pre-processed operation data sequence is read, the environmental temperature is set as an external convection reference temperature, the winding temperature is mapped as an initial temperature field of a solid region, the insulating oil temperature is mapped as an initial temperature field of a fluid region, a three-dimensional calculation domain containing the solid region and the fluid region is established, the boundary conditions of the solid and the fluid are defined, and an initialized three-dimensional multi-physical field coupling calculation setting is output. ​ Fluid mass conservation equation, momentum conservation equation and energy conservation equation are established in the fluid region, copper loss power corresponding to the load current is distributed as a heat source according to volume, iron loss is set as a fixed heat source, no-slip velocity condition and temperature and heat flow continuity condition are applied on the solid-fluid contact boundary, and an iterative solution format of the coupled equation set is output; The iterative solution format is taken as input, and the solid region and the fluid region are calculated in a time-stepping manner, in each time step, the fluid velocity field and the pressure field are solved first, the fluid temperature field is updated, and the solid temperature field is calculated, and the heat flow and the temperature are continuously exchanged at the solid-fluid interface until the interface temperature difference and the interface heat flow difference are less than a preset threshold; The average temperature difference between the oil path inlet and the outlet is calculated as the oil temperature rise, the copper loss power is obtained combined with the load current and the equivalent resistance, and the total loss power is obtained by adding the iron loss power, the heat power taken away by cooling is obtained by multiplying the oil mass flow, the specific heat of the oil and the temperature difference, the cooling efficiency is defined as the ratio of the heat power taken away by cooling to the total loss power, and the initial heat dissipation performance prediction result is output.

5. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The output process of the modified heat dissipation performance prediction result specifically includes: The initial heat dissipation performance prediction result is compared with the preprocessed running data sequence item by item to form an error sequence; The overall error measure is calculated using the error sequence, and the trend of the error measure is tracked, and the parameter adjustment operation is triggered when it is detected that the error measure does not decrease; Taking the parameter adjustment operation as the core and taking the trend of the error measure as the input, the key parameters in the three-dimensional multi-physical field coupling model are updated step by step, the prediction result is recalculated after each parameter update, and a new error sequence is formed, until the error measure remains stable and decreases in the continuous iteration process, and the modified heat dissipation performance prediction result is output.

6. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The output process of the second heat dissipation performance prediction result specifically includes: Based on the modified heat dissipation performance prediction result and the preprocessed running data sequence, the preprocessed record and the modified result at the same time are selected to construct a feature basis vector, the length of the time window is set and the basis statistical features of continuous time are collected in time sequence to form a time sequence feature vector; The parameter vector of the data-driven prediction model is set and the model output is set as the estimated value of the three-dimensional target quantity, the time sequence feature vector is selected as a batch training set, and batch updating is performed until the set condition is met, and the updated parameter vector is output; Based on the updated parameter vector and the time sequence feature vector, the prediction values of the winding hot spot temperature, the oil temperature rise and the cooling efficiency are obtained through parameter operation and feature mapping of the data-driven prediction model, and the second heat dissipation performance prediction result is output.

7. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The generation process of the real-time updated digital twin specifically includes: Based on the modified heat dissipation performance prediction result and the second heat dissipation performance prediction result, the two types of prediction values of the winding hot spot temperature, the oil temperature rise and the cooling efficiency are paired item by item at the same time index, a three-dimensional residual vector is calculated and combined, and each residual sample is attached with a corresponding spatial position and time label to obtain a residual sequence with space-time labels. The residual sequence with space-time labels is used to perform adaptive space-time feature mapping, and a fractional order space-time kernel function is used for weighted integration. The kernel function value is calculated for each sample label in the sequence, and the kernel function value is used as the weight for the corresponding residual vector. The space-time mapping feature vector at the current time is obtained by weighted accumulation; A Hamiltonian force field optimization model is constructed based on the space mapping feature vector. The feature weight is set as the generalized coordinate, and the corresponding conjugate momentum is recorded separately. The total energy is defined as the sum of kinetic energy and potential energy. Leapfrog format is used for discrete evolution. The conjugate momentum is updated by half step according to the potential energy gradient of the current coordinate. The coordinate is updated by whole step using the updated conjugate momentum, and the conjugate momentum is updated by half step according to the potential energy gradient of the new coordinate. The change amplitude of the total energy and the norm of the coordinate increment of the adjacent two times are monitored after each discrete step. When the evolution step reaches the set upper limit, the evolution is terminated, and the dynamic feature weight at the current time is output. The dynamic feature weight is projected to the fusion coefficient space corresponding to the three indicators through a fixed weight mapping matrix. The fusion result is obtained by combining each indicator between the two predicted values. The real-time updated digital twin composed of three fusion results is output according to the time index.

8. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The process of obtaining the complete heat dissipation performance prediction result specifically includes: Extracting temperature data and operating condition information at the current sampling time to form a data set for heat dissipation calculation; Input the data set into the real-time updated digital twin to calculate the temperature distribution of each position of the winding, determine the maximum temperature in the temperature distribution, obtain the winding hot spot temperature and spatial distribution, and output the hot spot prediction result; On the same data set, calculate the average temperature of the insulating oil, record the temperature difference between adjacent sampling times, obtain the change rule of the oil temperature with time, and determine the oil temperature rise by the difference between the outlet temperature and the inlet temperature. The oil temperature prediction result is output; The hot spot prediction result and the oil temperature prediction result are input, and the energy balance calculation is performed combined with the total loss power to obtain the ratio of the heat carried away to the loss. The current ratio is taken as the cooling efficiency, and the complete heat dissipation performance prediction result composed of the winding hot spot temperature, the oil temperature change rule and the cooling efficiency is output.

9. The transformer heat dissipation performance prediction method based on digital twinning of claim 1, wherein, The triggering of the risk prompt information specifically includes: Extracting the three prediction indicators of each time of the complete heat dissipation performance prediction result in time sequence, and arranging the three prediction indicators of the same time into a record. The prediction result sequence is composed of all records in time order; Compare each record of the prediction result sequence with the preset threshold set item by item, and give a comparison conclusion for each time of the three prediction indicators. The comparison conclusion is divided into non-overrun and overrun. The discrimination record containing the three comparison conclusions is output in time sequence; Based on the discrimination record output in time sequence, the warning judgment is made for each time, and the specific indicator name leading to the warning is summarized and the warning judgment result and the cause information are output; When the warning condition is established, the corresponding strategy is selected from the operation optimization strategy library according to the cause information, and the operation optimization instruction is generated, and the risk prompt information facing the monitoring system is generated.

Images