Intelligent adjusting system and method for transformer

Abstract

The invention belongs to the technical field of power system control, and discloses an intelligent adjusting system and method for a transformer. The method comprises the following steps: inputting transformer operation data and weather data collected in real time into a pre-trained load prediction model to obtain predicted load data; inputting the acquired physical parameters and operation data and preset calibration parameters into a pre-constructed digital twinborn model to obtain a cooling capacity index; calculating the total heat source power of the copper loss and the iron loss according to the predicted load data and the electrical parameters of the transformer collected in real time, and dynamically quantifying the balance state of the total heat source power and the cooling capacity index through a matching degree evaluation function in combination with the cooling capacity index to obtain a matching degree evaluation index; and the problem of dynamic unbalance between the heat dissipation capability of the transformer and the load demand in an extreme environment is systematically solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system control, and more specifically, to an intelligent transformer regulation system and method. Background Art

[0002] With the deep integration of smart grids and new power systems, the operational reliability of transformers faces severe challenges in complex environments. Taking a power grid as an example, it is characterized by high load density and high frequency of extreme weather. Meteorological conditions such as typhoons and persistent high temperatures affect more than 2,000 transformers annually on average. The traditional transformer cooling system relies on static model control with fixed thresholds and exposes core contradictions in extreme weather (including high temperature, high voltage, and typhoons, where high temperature, high voltage, and typhoons are all weather conditions exceeding the preset corresponding thresholds): high temperature leads to a decrease in air density and an increase in the thermal resistance of heat sinks, and strong typhoon winds cause a decline in the efficiency of cooling fans and an acceleration of the fouling deposition on heat sinks, directly resulting in a decrease in the effectiveness of the cooling system; at the same time, the sudden load changes accompanied by extreme weather cause the heat dissipation demand to soar in a short period, forming a dynamic imbalance of "decrease in heat dissipation capacity and surge in load demand".

[0003] The existing technology lacks real-time coupling analysis of meteorological parameters, equipment status, and load fluctuations. The adjustment strategy lags behind the actual demand, and a dynamic evaluation model for the attenuation of heat dissipation capacity has not been established. Adjusting solely based on manual experience is likely to lead to insufficient cooling or overcooling, thus resulting in a dynamic imbalance between the heat dissipation capacity and load demand of transformers in extreme environments.

[0004] In view of this, the present invention proposes an intelligent transformer regulation system and method to solve the above problems. Summary of the Invention

[0005] To overcome the above-mentioned defects of the prior art and achieve the above object, the present invention provides the following technical solution: An intelligent transformer regulation method, comprising:

[0006] Input the real-time collected transformer operation data and weather data into a pre-trained load prediction model to obtain predicted load data;

[0007] Input the collected physical parameters and operation data, as well as the preset calibration parameters, into a pre-constructed digital twin model to obtain a cooling capacity index;

[0008] Calculate the total heat source power of copper loss and iron loss based on the predicted load data and the real-time collected electrical parameters of the transformer, and combine with the cooling capacity index. Dynamically quantify the balance state between the total heat source power and the cooling capacity index through a matching degree evaluation function to obtain a matching degree evaluation index;

[0009] If the matching degree evaluation index exceeds the set threshold, the device status, operation data, and weather data collected in real time are input into a pre-trained cooling strategy model to obtain a cooling strategy;

[0010] After verifying that the device status is normal, adjustment instructions are generated and executed according to the cooling strategy; during the execution of the adjustment instructions, the calibration parameters of the digital twin model are dynamically corrected based on the device status and operation data collected in real time.

[0011] Furthermore, the training method of the load prediction model includes:

[0012] Collect the transformer operation data and weather data during historical extreme weather periods and mark them as training data;

[0013] After performing normalization preprocessing on the training data, it is used as the model input sample, and the actual load data during the historical extreme weather period is used as the supervision signal;

[0014] Use a long short-term memory neural network to construct a load prediction model; the long short-term memory neural network includes an input layer, multiple hidden layers, and a fully connected output layer; the dimension of the input layer is consistent with the feature dimension of the preprocessed model input sample; the fully connected output layer outputs the predicted load data for a future preset time period and marks the load surge points in the predicted load data;

[0015] With the goal of minimizing the mean square error between the predicted load and the supervision signal, the parameters of the load prediction model are iteratively optimized through the backpropagation algorithm to obtain the final load prediction model.

[0016] Furthermore, the method for determining whether a data point in the predicted load data is a load surge point includes:

[0017] Construct a multi-dimensional feature space including time features, load historical trend features, and extreme meteorological features; the time features include periodicity indicators of daily, weekly, and monthly cycles; the load historical trend features include statistical indicators of historical load mean, historical load standard deviation, and historical load change rate; the extreme meteorological features are weather data and its change rate;

[0018] Based on the multi-dimensional feature space, calculate the feature values of each data point in the predicted load data in sequence to construct a set of feature values; the feature values include load growth rate, deviation from the historical load mean, and correlation with weather data;

[0019] Calculate the matching degree between each eigenvalue in the eigenvalue set and the preset surge point eigenvalue; if the matching degree exceeds the set matching degree threshold, determine the corresponding data point as a load surge point and mark the corresponding extreme scenario type; if the matching degree does not exceed the set matching degree threshold, determine the corresponding data point as a normal data point; the extreme scenario types include high temperature, high pressure, and strong wind;

[0020] For the data points determined as load surge points in the predicted load data, verify whether they meet the persistence condition of load surge.

[0021] Furthermore, the construction method of the digital twin model includes:

[0022] Based on the collected three-dimensional geometric parameters inside the transformer, construct a three-dimensional geometric model;

[0023] Take the collected physical parameters and historical operation data of the transformer as the input of the digital twin model; the physical parameters include material thermal characteristic parameters, oil viscosity-temperature relationship, and air density-temperature relationship;

[0024] Adopt the computational fluid dynamics method, couple the Navier-Stokes equation and the Fourier heat conduction equation, and establish a multi-physics field coupling model;

[0025] Introduce the particle swarm optimization algorithm, and calibrate the calibration parameters in the multi-physics field coupling model with the optimization goal of minimizing the root mean square error between the oil temperature distribution predicted by the multi-physics field coupling model and the measured oil temperature distribution in the historical operation data; the calibration parameters include the convective heat transfer coefficient and the oil circulation resistance coefficient;

[0026] The digital twin model is jointly composed of the multi-physics field coupling model and the three-dimensional geometric model.

[0027] Furthermore, the method for calibrating the calibration parameters in the multi-physics field coupling model includes:

[0028] Construct a particle swarm optimization space; each particle in the particle swarm optimization space is defined as a two-dimensional vector composed of the convective heat transfer coefficient and the oil circulation resistance coefficient;

[0029] Input the parameter vector of each particle into the multi-physics field coupling model, conduct numerical simulation in combination with historical operation data, and obtain the predicted oil temperature distribution; compare the predicted oil temperature distribution with the measured oil temperature distribution in the same period of history, and calculate the root mean square error as the particle fitness value;

[0030] Each particle in the particle swarm optimization space adjusts its speed and position according to its own searched minimum root mean square error parameter combination and the minimum root mean square error parameter combination in the current population, and searches for a better solution area in the parameter space;

[0031] When the change in the global optimal fitness value for a continuously preset positive integer number of iterations is less than the set fitness threshold, or when the maximum number of iterations is reached, the optimization process is terminated;

[0032] Embed the convective heat transfer coefficient and the oil circulation resistance coefficient that minimize the root mean square error obtained by particle swarm optimization into the multi-physics coupling model to complete the calibration of the calibration parameters in the multi-physics coupling model.

[0033] Furthermore, the method for obtaining the matching degree evaluation index includes:

[0034] Calculate the total heat source power of copper loss and iron loss based on the predicted load data and the electrical parameters of the transformer;

[0035] Calculate the ratio of the cooling capacity index obtained from the digital twin model to the total heat source power to form a basic matching degree index;

[0036] Monitor the temperature distribution of the key parts of the transformer. If the temperature difference between each part is greater than the set temperature difference, a penalty correction is made to the basic matching degree index; the key parts include windings, iron cores, and the top layer of the oil tank;

[0037] Dynamically adjust the basic matching degree index in combination with real-time weather data to obtain a dimensionless matching degree evaluation index.

[0038] Furthermore, the training method of the cooling strategy model includes:

[0039] Construct a multi-dimensional state space including the equipment status, operation data, and weather data of the transformer;

[0040] Construct an action space of discrete adjustment methods including fan speed regulation, standby equipment switching, and oil pump power regulation;

[0041] Use the proximal policy optimization algorithm to construct a policy network; the policy network takes the multi-dimensional state space as input and the action probability distribution of the action space as output;

[0042] Design a composite reward function including safety reward, energy consumption reward, and life reward;

[0043] Based on the composite reward function, iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the adjustment method selected by the policy network to obtain the cooling strategy model.

[0044] Furthermore, the method for iteratively optimizing the parameters of the policy network based on the composite reward function to maximize the cumulative reward value of the adjustment method selected by the policy network includes:

[0045] Use the digital twin model to simulate extreme weather scenarios and generate equipment statuses under different extreme weather scenarios;

[0046] The policy network selects a regulation method from the action space according to the current policy state;

[0047] The digital twin model simulates the operation of the device according to the selected regulation method, outputs the device state at the next unit time, and calculates the reward value of the corresponding composite reward function;

[0048] Store the device state at the next unit time and the reward value of the corresponding composite reward function in the experience pool;

[0049] Evaluate the relative value of each selected regulation method through the Generalized Advantage Estimation method;

[0050] Based on the data in the experience pool and the relative value of the regulation method, use the gradient descent algorithm to iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the regulation method selected by the policy network.

[0051] Furthermore, the method for dynamically correcting the calibration parameters of the digital twin model includes:

[0052] During the execution of the regulation instruction of the cooling strategy, collect the transformer operation data and device state in real time as the feedback data for dynamic correction;

[0053] Compare the real-time prediction result of the digital twin model with the feedback data, and calculate the deviation between the real-time prediction result and the feedback data;

[0054] If the deviation exceeds the set deviation, identify the calibration parameters that affect the prediction accuracy, and use the gradient descent algorithm to iteratively adjust the calibration parameters according to the direction and magnitude of the deviation;

[0055] After adjustment, verify the prediction accuracy of the corrected digital twin model in the subsequent set period. If the deviation converges to the set deviation threshold in the subsequent set period, confirm that the parameter correction is effective and update the digital twin model; if the deviation does not converge to the set deviation threshold in the subsequent set period, enter the next round of correction until the deviation converges to the set deviation threshold.

[0056] Furthermore, the transformer operation data includes the vibration amplitude of the cooling equipment, the state of the transformer body, electrical parameters, oil temperature distribution, and load information; the weather data includes environmental temperature, humidity, wind speed, precipitation, and weather warning type; the electrical parameters include load current and winding resistance.

[0057] The present invention discloses a transformer intelligent regulation system, including:

[0058] A load prediction module, configured to input the real-time collected transformer operation data and weather data into a pre-trained load prediction model to obtain predicted load data;

[0059] A digital twin module, which is used to input the collected physical parameters, operation data, and preset calibration parameters into a pre-constructed digital twin model to obtain a cooling capacity index;

[0060] A matching degree evaluation module, which calculates the total heat source power of copper loss and iron loss based on the predicted load data and the electrical parameters of the transformer collected in real time, and combines the cooling capacity index. Through a matching degree evaluation function, it dynamically quantifies the balance state between the total heat source power and the cooling capacity index to obtain a matching degree evaluation index;

[0061] An intelligent decision-making module. If the matching degree evaluation index exceeds the set threshold, it inputs the device status, operation data, and weather data collected in real time into a pre-trained cooling strategy model to obtain a cooling strategy;

[0062] A closed-loop control module, which is used to generate and execute adjustment instructions according to the cooling strategy after verifying that the device status is normal. During the execution of the adjustment instructions, based on the device status and operation data collected in real time, it dynamically corrects the calibration parameters of the digital twin model.

[0063] Compared with the prior art, the technical effects and advantages of an intelligent adjustment system and method for a transformer of the present invention are as follows:

[0064] Predictive load data is generated by collecting the operation data and weather data of the transformer in real time; the cooling capacity index of the transformer is obtained through the constructed digital twin model; the balance state between the current total heat source power and the cooling capacity index of the transformer is dynamically evaluated by combining the predictive load data and the cooling capacity index to obtain a matching degree evaluation index; according to the matching degree evaluation index and the device status, operation data, and weather data collected in real time, the corresponding cooling strategy is selected through the cooling strategy model; adjustment instructions are generated and executed according to the cooling strategy; the device status is verified before execution, and the parameters of the digital twin model are dynamically corrected according to the feedback data after execution. The entire system performs real-time coupling analysis on weather data, device status, and load fluctuations, establishes a dynamic evaluation model for the attenuation of heat dissipation capacity, solves the problem of dynamic imbalance between the heat dissipation capacity and load demand of the transformer in extreme environments, realizes precise perception of the device status, intelligent generation of cooling strategies, and continuous optimization of the adjustment effect, and significantly improves the safety, economy, and reliability of the transformer operation. Description of the Drawings

[0065] Figure 1 It is a schematic diagram of an intelligent adjustment system for a transformer according to Embodiment 1 of the present invention;

[0066] Figure 2 It is a flowchart of an intelligent adjustment method for a transformer according to Embodiment 3 of the present invention;

[0067] Figure 3Flowchart of the method for determining whether it is a load surge point in Embodiment 1 of the present invention;

[0068] Figure 4 Flowchart of the method for obtaining the matching degree evaluation index in Embodiment 1 of the present invention. Detailed implementation manners

[0069] The following will combine the accompanying drawings in the embodiments of the present invention to describe the technical solutions in the embodiments of the present invention in detail, clearly and completely. It should be specifically noted that the following described specific embodiments are only used to better explain and illustrate the technical solutions of the present invention, aiming to enable those skilled in the art to better understand and implement the present invention, and should not be construed as a limitation on the protection scope of the present invention. Without departing from the spirit and essence of the present invention, those skilled in the art can modify, adjust or equivalently replace it according to the content disclosed in the present invention, and these should all be regarded as the protection scope of the present invention.

[0070] Embodiment 1

[0071] Please refer to Figure 1 As shown, this embodiment discloses a transformer intelligent regulation system, including a load prediction module, a digital twin module, a matching degree evaluation module, an intelligent decision-making module and a closed-loop control module. Each module is connected by wire and / or wirelessly to achieve data transmission.

[0072] The load prediction module is used to input the real-time collected transformer operation data and weather data into a pre-trained load prediction model to obtain predicted load data.

[0073] The transformer operation data includes the vibration amplitude of the cooling equipment, the state of the transformer body, electrical parameters, oil temperature distribution and load information; the weather data includes ambient temperature, humidity, wind speed, precipitation and weather warning type.

[0074] The training method of the load prediction model includes:

[0075] Collect the transformer operation data and weather data during the historical extreme weather period and mark them as training data;

[0076] Perform normalization preprocessing on the training data and use it as the model input sample, and use the actual load data during the historical extreme weather period as the supervision signal;

[0077] Construct a load prediction model using a long short-term memory neural network; the long short-term memory neural network includes an input layer, multiple hidden layers and a fully connected output layer; the dimension of the input layer is consistent with the feature dimension of the preprocessed model input sample; the fully connected output layer outputs the predicted load data for a future preset time period and marks the load surge points in the predicted load data;

[0078] With the goal of minimizing the mean square error between the predicted load and the supervision signal, the parameters of the load prediction model are iteratively optimized through the backpropagation algorithm to obtain the final load prediction model.

[0079] The methods for collecting the transformer operation data and weather data during historical extreme weather periods and labeling them as training data include:

[0080] Obtain the transformer operation data and the corresponding weather data during extreme weather periods for three consecutive years, and label them as training data; the extreme weather periods include high-temperature orange warning days, typhoon landing periods, and cold wave passing days. Synchronously sample the training data at 15-minute intervals to form an original data sequence containing timestamps.

[0081] The methods for using the preprocessed training data as model input samples and using the actual load data during historical extreme weather periods as supervision signals include:

[0082] For the numerical features in the transformer operation data and weather data, use the min-max scaling method to uniformly map them to the [0,1] interval; the numerical features include temperature, current, and wind speed; for categorical features such as weather warning types, use one-hot encoding to convert them into binary vectors. Each processed sample forms an input vector containing 20-dimensional features, and at the same time, extract the actual load data corresponding to the time period as the supervision signal to form a training data set containing 100,000 samples.

[0083] The methods for constructing a load prediction model using a long short-term memory neural network (LSTM) and outputting the predicted load data for a future preset time period and marking the load surge points in the predicted load data include:

[0084] The load prediction model includes an input layer, 3 LSTM hidden layers, and a fully connected output layer. The number of neurons in the input layer is the same as the dimension of the preprocessed features, which is 20, and it is used to receive the normalized input vector. Each LSTM hidden layer is set with 128 memory units, and the dropout technique is used between layers to prevent overfitting, and the activation function uses the tanh function; the inactivation rate of the dropout technique is 0.2. The fully connected output layer is used to output the predicted load data at 15-minute intervals within the next 24 hours, with a total of 96 neurons.

[0085] During the model training stage, the mean square error between the predicted load data and the supervision signal is used as the main loss function, and the binary cross-entropy loss is used as the auxiliary loss function. The total loss function is the weighted sum of the two. The weights of the main loss function and the auxiliary loss function are 0.8 and 0.2 respectively. The optimization algorithm uses the Adam algorithm, the initial learning rate is set to 0.001, the batch size is 64, and the training period is 200. During the training process, the input samples are input into the model according to the time series. The LSTM hidden layer captures the time dependence of the load data and the abnormal fluctuation characteristics under extreme weather effects through the gating mechanism. The fully connected layer generates the predicted load sequence and the probability distribution of the surge points based on the output of the hidden layer. After each training period, the loss value on the validation set is calculated. When the validation loss does not decrease for 10 consecutive periods, the early stopping mechanism is triggered to terminate the training.

[0086] It should be noted that during the above normalization preprocessing process, the missing data is filled using the adjacent moment mean interpolation method to ensure the integrity of the input samples. When constructing the LSTM hidden layer, the number of memory units in each layer can be adjusted according to the computing resources and model complexity, and the typical range is from 64 to 256. During the training process, a dynamic learning rate adjustment strategy can be adopted. When the decrease amplitude of the loss value is less than 1%, the learning rate is decayed by a factor of 0.9 to improve the optimization efficiency.

[0087] A method for iteratively optimizing the load prediction model parameters through the backpropagation algorithm with the goal of minimizing the mean square error between the predicted load and the supervision signal to obtain the final load prediction model includes:

[0088] Iteratively optimizing the model parameters through the backpropagation algorithm; the load prediction model parameters include the weight matrix, bias vector of the LSTM layer, and the connection parameters of the fully connected layer; when the total loss function converges to less than 0.05, the final load prediction model is obtained.

[0089] Refer to Figure 3 As shown, a method for determining whether a data point in the predicted load data is a load surge point includes:

[0090] Construct a multi-dimensional feature space including time features, load historical trend features, and extreme meteorological features; the time features include periodic indicators of daily cycle, weekly cycle, and monthly cycle; the load historical trend features include statistical indicators of historical load mean, historical load standard deviation, and historical load change rate; the extreme meteorological features are weather data and its change rate;

[0091] Based on the multi-dimensional feature space, calculate the feature values of each data point in the predicted load data in sequence to construct a feature value set; the feature values include load growth rate, deviation from the historical mean, and correlation with weather data;

[0092] Calculate the matching degree between each eigenvalue in the eigenvalue set and the preset surge point eigenvalue; if the matching degree exceeds the set matching degree threshold, determine the corresponding data point as a load surge point and mark the corresponding extreme scenario type; if the matching degree does not exceed the set matching degree threshold, determine the corresponding data point as a normal data point; the extreme scenario types include high temperature, high pressure, high humidity, and strong wind;

[0093] For the data points determined as load surge points in the predicted load data, further verify whether they meet the persistence condition of load surge.

[0094] The method for constructing a multi-dimensional feature space including time features, load historical trend features, and extreme meteorological features includes:

[0095] Regarding the time series characteristics of the predicted load data, extract three types of features to construct a feature space: Time features calculate the load mean and the periodic index of the peak occurrence time within the daily cycle (24 hours), weekly cycle (7 days), and monthly cycle (30 days) through the sliding window technique, which are used to identify the time pattern of load fluctuations; Load historical trend features are based on the load data of the past three years, calculate the historical load mean, standard deviation, and 15-minute sliding window change rate corresponding to each time point, where the historical mean reflects the normal load level, the standard deviation characterizes the load fluctuation range, and the change rate is used to measure the load change speed; Extreme meteorological features collect the real-time environmental temperature, wind speed, precipitation, and their hourly change rates. For example, the temperature change rate is defined as the temperature difference between the current moment and the previous hour, and the wind speed change rate is defined as the wind speed difference between the current moment and the previous hour, which are used to quantify the impact of extreme weather on the load.

[0096] The method for constructing an eigenvalue set by calculating the eigenvalue of each data point in the predicted load data based on the multi-dimensional feature space includes:

[0097] For time features, calculate the deviation degree between the load mean in the same time period within the daily cycle corresponding to the current data point and the historical daily mean. The calculation method is:

[0098] ;

[0099] In the formula, is the deviation degree between the load mean in the same time period within the daily cycle corresponding to the current data point and the historical daily mean; is the current load value; is the historical load mean in the same time period on the same day; is the historical load standard deviation in the same time period on the same day.

[0100] The calculation method of the load growth rate in the load historical trend features is:

[0101] ;

[0102] In the formula, is the load growth rate; is the load value in the previous sampling period.

[0103] The deviation degree between the load historical trend feature and the historical mean value is calculated as:

[0104] ;

[0105] In the formula, is the deviation degree between the load historical trend feature and the historical mean value; is the historical full - time period load mean value.

[0106] The correlation of extreme meteorological features is calculated through the Pearson correlation coefficient. For example, the calculation method of temperature correlation is:

[0107] ;

[0108] In the formula, is the temperature correlation; is the historical temperature data; is the corresponding load data; is the mean value of temperature data; is the mean value of load data; is the standard deviation of temperature data; is the standard deviation of load data; i is the index of the data, representing the i - th sample. For example, the first temperature and load data are and ; n is the number of data, that is, the number of temperature and load data. A five - dimensional eigenvalue set including time deviation degree, load growth rate, historical deviation degree, temperature correlation and wind speed correlation is formed.

[0109] The method of calculating the matching degree between each eigenvalue in the eigenvalue set and the preset surge point eigenvalue, and if the matching degree exceeds the set matching degree threshold, then determining the corresponding data point as a load surge point and marking the corresponding extreme scenario type includes:

[0110] The preset eigenvalue is statistically obtained based on historical extreme scenario data: for example, the load growth rate threshold in a high - temperature scenario is 20%, the historical deviation degree threshold is 1.5, and the temperature correlation threshold is 0.8; in a high - pressure scenario, the load growth rate threshold is 15%, the historical deviation degree threshold is 1.3, and the wind speed correlation threshold is 0.6; in a strong - wind scenario, the load growth rate threshold is 18%, the historical deviation degree threshold is 1.4, and the wind speed correlation threshold is 0.7.

[0111] The matching degree calculation adopts the weighted summation method, and the weight distribution can be: load growth rate 0.4, historical deviation 0.3, temperature correlation 0.15, wind speed correlation 0.15. The load growth rate directly reflects the surge amplitude; the historical deviation reflects the degree of abnormality; the temperature correlation is a sensitive factor for high-temperature scenarios; the wind speed correlation is a sensitive factor for strong-wind scenarios. When the matching degree is greater than or equal to 0.7, the data point is determined as a candidate load surge point, and the corresponding extreme scenario type is marked according to the maximum correlation feature.

[0112] The method for further verifying whether the data points determined as load surge points in the predicted load data meet the persistence condition of load surge includes:

[0113] Check whether the load growth rates of this data point and the next two consecutive sampling periods, that is, a total of three periods, 45 minutes, are all greater than 15%, and the historical deviations are all greater than 1.2. Setting this condition is to exclude short-term noise interference. For example, instantaneous pulse loads may cause individual data points to be abnormal, but if the duration is insufficient, it does not belong to the real surge scenario. If the persistence condition is met, it is finally determined as a load surge point; if not, the candidate point is removed.

[0114] It should be noted that in the construction of the above multi-dimensional feature space, the length of the time window can be adjusted according to the actual operation cycle of the transformer. The typical daily cycle window is 24 hours, the weekly cycle is 7 days, and the monthly cycle is 30 days to cover the daily fluctuations, weekend effects, and seasonal change laws of the load; when calculating the feature values, the historical data sample size should be no less than 3 years to ensure the stability of statistical parameters; the matching degree weight distribution is determined through machine learning training of historical surge events, and a logistic regression model is used to optimize the weights of each feature to make the misjudgment rate less than 5%; the number of sampling periods for the persistence condition is set to 3, corresponding to a duration of 45 minutes, which meets the minimum response cycle requirements of the power system load regulation, can avoid misjudgment, and ensure the timeliness of early warning.

[0115] It should be noted that through the above method, it is possible to systematically identify load surge points from three dimensions: time law, load fluctuation characteristics, and extreme meteorological impacts, and improve the determination accuracy by combining the persistence verification mechanism, providing a reliable decision-making basis for the pre-adjustment of the transformer cooling strategy.

[0116] The digital twin module is used to input the collected physical parameters, operation data, and pre-set calibration parameters into the pre-constructed digital twin model to obtain the cooling capacity index.

[0117] The construction method of the digital twin model includes:

[0118] Based on the collected three-dimensional geometric parameters inside the transformer, construct a three-dimensional geometric model;

[0119] The collected physical parameters and historical operation data of the transformer are used as the input of the digital twin model; the physical parameters include material thermal characteristic parameters, oil viscosity-temperature relationship, and air density-temperature relationship.

[0120] The computational fluid dynamics method is adopted to couple the Navier-Stokes equation and the Fourier heat conduction equation to establish a multi-physics field coupling model.

[0121] The particle swarm optimization algorithm is introduced, and the calibration parameters in the multi-physics field coupling model are calibrated with the goal of minimizing the root mean square error between the oil temperature distribution predicted by the multi-physics field coupling model and the measured oil temperature distribution in the historical operation data; the calibration parameters include the convective heat transfer coefficient and the oil circulation resistance coefficient.

[0122] The digital twin model is jointly composed of a multi-physics field coupling model and a three-dimensional geometric model.

[0123] The method for constructing the three-dimensional geometric model based on the collected three-dimensional geometric parameters of the transformer inside includes:

[0124] Based on laser scanning technology, the three-dimensional geometric parameters of the winding arrangement, oil duct distribution, and radiator structure inside the transformer are obtained, and the scanning accuracy is controlled at the millimeter level to ensure that the model can accurately reflect the internal space structure of the transformer. The computer-aided design software is used to construct the three-dimensional geometric model, and the obtained point cloud data is meshed to generate a three-dimensional solid model containing at least 100,000 grid cells, where the grid density of the key heat-generating components is not less than 20 cells per square centimeter to ensure the accuracy of subsequent thermal analysis; the key heat-generating components include the high-voltage winding and the low-voltage winding.

[0125] It should be noted that during the construction process of the above three-dimensional geometric model, the accuracy of mesh division will affect the accuracy and calculation efficiency of the calculation results. Finer meshes can provide more accurate calculation results, but will increase the calculation amount and calculation time; coarser meshes have faster calculation speeds, but may lose a certain amount of accuracy. Therefore, it is necessary to make a trade-off according to the actual situation and select an appropriate mesh division accuracy. In this embodiment, after multiple tests and verifications, a mesh division scheme that can both ensure the calculation accuracy and meet the calculation efficiency requirements is determined.

[0126] The method for using the collected physical parameters and historical operation data of the transformer as the input of the digital twin model includes:

[0127] Collect the thermal characteristic parameters of the transformer materials, including the thermal conductivity of the winding copper material and the specific heat capacity of the core silicon steel sheet; the typical value of the thermal conductivity of the winding copper material is 385 W / (m·K); the typical value of the specific heat capacity of the core silicon steel sheet is 460 J / (kg·K); obtain the viscosity-temperature relationship curve of the transformer oil; the viscosity-temperature relationship curve is obtained through experimental measurement, covering the temperature range from -20°C to 120°C, and is used to describe the influence of the oil temperature change on the oil fluidity; collect the air density-temperature relationship data; the air density-temperature relationship conforms to the ideal gas state equation and is used to calculate the heat exchange characteristics of the air around the radiator. At the same time, collect the historical operation data of the past three years, including the oil temperature distribution and the cooling equipment power under different load conditions, with a sampling frequency of 15 minutes per time, to form a historical data set containing at least 100,000 records.

[0128] It should be noted that when collecting physical parameters and historical operation data, the accuracy and integrity of the data are crucial for the performance of the model. To ensure the accuracy of the data, high-precision measuring instruments and equipment are used for data collection, and the collected data is verified and corrected multiple times. At the same time, to ensure the integrity of the data, as much historical operation data as possible is collected, including data under different seasons and different load conditions, so that the model can better adapt to various actual operation situations.

[0129] The method of establishing a multi-physics field coupling model by using the computational fluid dynamics method and coupling the Navier-Stokes equation and the Fourier heat conduction equation includes:

[0130] Use the computational fluid dynamics method to couple the Navier-Stokes equation and the Fourier heat conduction equation. Among them, the Navier-Stokes equation is used to describe the flow characteristics of the transformer oil, considering the characteristics of the oil viscosity and density changing with temperature; the Fourier heat conduction equation is used to calculate the heat transfer process inside the transformer, including the heat conduction of solid components and the convective heat transfer of fluids. At least 5 monitoring points are set in the model, distributed at key positions such as the winding hot spot, the middle of the core, and the oil conservator, for subsequent comparison with the measured data.

[0131] It should be noted that in the multi-physics field coupling model, appropriate numerical methods and algorithms are required to solve the Navier-Stokes equation and the Fourier heat conduction equation. In this embodiment, the finite volume method is used to discretize these two equations, and an iterative solution algorithm is used to solve the discretized equations. The finite volume method is a commonly used numerical calculation method, which has the advantages of high calculation accuracy and good stability. The iterative solution algorithm can effectively handle complex non-linear equations, and by continuously iterating and updating the numerical solution, it finally converges to a solution that meets the accuracy requirements.

[0132] Introduce the particle swarm optimization algorithm. The method for calibrating the calibration parameters in the multi-physics field coupling model with the goal of minimizing the root mean square error between the oil temperature distribution predicted by the multi-physics field coupling model and the measured oil temperature distribution in historical operation data includes:

[0133] Construct a particle swarm optimization space; each particle in the particle swarm optimization space is defined as a two-dimensional vector composed of the convective heat transfer coefficient and the oil circulation resistance coefficient;

[0134] Input the parameter vectors of each particle into the multi-physics field coupling model, perform numerical simulation in combination with historical operation data to obtain the predicted oil temperature distribution; compare the predicted oil temperature distribution with the measured oil temperature distribution in the same historical period, and calculate the root mean square error as the particle fitness value;

[0135] Each particle in the particle swarm optimization space adjusts its velocity and position according to the combination of the minimum root mean square error parameters searched by itself and the combination of the minimum root mean square error parameters in the current population, and searches for a better solution area in the parameter space according to a preset rule;

[0136] When the change amount of the global optimal fitness value in continuously preset positive integer iterations is less than the set fitness threshold, or when the maximum number of iterations is reached, terminate the optimization process;

[0137] Embed the convective heat transfer coefficient and the oil circulation resistance coefficient with the minimized root mean square error obtained by particle swarm optimization into the multi-physics field coupling model to complete the calibration of the calibration parameters in the multi-physics field coupling model.

[0138] The method for constructing the particle swarm optimization space includes:

[0139] Define the key parameters to be optimized as a two-dimensional parameter vector , where represents the convective heat transfer coefficient on the radiator surface, which is used to characterize the heat exchange efficiency between air and the radiator. Its initial value range is set to 50 to 200, and this range covers the change in convective heat transfer capacity during normal operation and extreme environments of the transformer; represents the oil circulation resistance coefficient of the transformer, which is used to describe the resistance characteristics of the oil flow in the circulation path. The initial value range is set to 0.1 to 0.3, and this range is determined based on the transformer oil duct structure and fluid mechanics principles. Each particle randomly generates an initial position and an initial velocity within the above parameter range at the beginning, and the initial velocity range is set to 20% of the position range, that is, -30 to 30 and -0.04 to 0.04, to ensure that the particle has sufficient initial exploration ability within the search space.

[0140] The parameter vector of each particle is input into the multi-physics field coupling model, and numerical simulation is performed in combination with historical operating data to obtain the predicted oil temperature distribution. The predicted oil temperature distribution is then compared with the historical measured oil temperature distribution during the same period. The root mean square error is calculated as the particle fitness value. The method includes:

[0141] The parameter vector of each particle is input into the established multi-physics coupling model, and numerical simulation is performed in combination with historical operating data. The historical operating data includes historical data of at least 10 typical operating conditions, such as high temperature, low temperature, rated load, and overload. Each operating condition contains no less than 500 sets of sampling data, and the sampling interval is 10 minutes. During the simulation process, the multi-physics coupling model calculates the temperature distribution of the transformer winding, core, and oil based on the input convective heat transfer coefficient and oil circulation resistance coefficient. The predicted oil temperature distribution is compared with the measured oil temperature data of the same period, and the root mean square error formula is used to calculate the fitness value. The measured oil temperature data of the same period is collected by the transformer's built-in temperature sensor, with an accuracy of ±0.5°C, and is arranged at five key locations, such as the winding hotspot, the middle of the core, and the oil surface.

[0142] In the particle swarm optimization space, each particle adjusts its speed and position according to the preset rules based on the minimum error parameter combination it has searched for and the minimum error parameter combination in the current population. The method of searching for a better solution region in the parameter space includes:

[0143] Each particle tracks two optimal positions during the search process: one is the minimum error parameter combination searched by the particle itself, that is, the individual optimal position; the other is the minimum error parameter combination currently searched by the entire particle swarm, that is, the global optimal position. The particle's speed and position are updated according to the following rules:

[0144] ;

[0145] ;

[0146] Where, is the current velocity of the particle; is the current position of the particle; is the updated velocity of the particle; is the updated position of the particle; is the optimal position of the individual; is the global optimal position; is the inertia weight, which is initially set to 0.9 and linearly decreases to 0.4 with the number of iterations to balance the global search and local development capabilities; and is the learning factor, which is set to 2.0 and is used to adjust the learning intensity of particles toward the individual optimal and global optimal positions; 、 is a random number in the range of 0 to 1, which is used to increase the randomness of the search process. Through the above update rules, the particles continuously move towards the region of better solutions in the parameter space, realizing the collaborative optimization of individual experience and group information.

[0147] When the change amount of the global optimal fitness value in consecutive preset positive integer iterations is less than the set fitness threshold, or when the maximum number of iterations is reached, the methods for terminating the optimization process include:

[0148] Set the maximum number of iterations to 200 times to avoid excessive consumption of computing resources; at the same time, when the change amount of the global optimal fitness value in consecutive 10 iterations is less than 0.1, it is determined that the algorithm has converged and the optimization process is terminated in advance. The setting of this termination condition not only ensures that the algorithm has sufficient search time but also avoids invalid repeated calculations, taking into account both calibration accuracy and efficiency.

[0149] The method for calibrating the calibration parameters in the multi-physics field coupling model by embedding the minimum root mean square error convective heat transfer coefficient and oil circulation resistance coefficient obtained by particle swarm optimization includes:

[0150] After the optimization process is terminated, the parameter combination with the minimum root mean square error obtained by particle swarm search, that is, the optimal convective heat transfer coefficient and oil circulation resistance coefficient, is embedded into the multi-physics field coupling model. After embedding, the accuracy of the model is verified: under normal operating conditions, the root mean square error between the predicted oil temperature and the measured oil temperature of the model should not exceed 1.5; under extreme conditions, such as when the ambient temperature suddenly rises by 10°C and the load suddenly increases by 30%, the error should not exceed 2.0. If the verification passes, the calibration is completed; if the verification fails, the initial parameter range or the parameters of the optimization algorithm can be adjusted and the calibration can be performed again.

[0151] It should be noted that the parameter settings of the above particle swarm optimization algorithm are determined based on the balance between computational complexity and search accuracy. Through a large number of simulation experiments, it is verified that this quantity can not only ensure the diversity of the population but also avoid excessive computing time; the selection of historical operation data needs to cover typical working conditions within the entire life cycle of the transformer to ensure that the calibrated model has wide applicability; the layout of monitoring points needs to conform to the thermal distribution characteristics of the transformer, and areas with larger temperature gradients are preferably selected to improve the accuracy of fitness calculation. Through the above methods, the key parameters of the multi-physics field coupling model can be effectively optimized, the prediction accuracy of the digital twin model for the thermal state of the transformer can be improved, and a reliable physical model support can be provided for the subsequent intelligent generation of cooling strategies.

[0152] The matching degree evaluation module calculates the total heat source power of copper loss and iron loss based on the predicted load data and the electrical parameters of the transformer collected in real time, and combines the cooling capacity index. Through the matching degree evaluation function, the balance state between the total heat source power and the cooling capacity index is dynamically quantified to obtain the matching degree evaluation index.

[0153] Refer to Figure 4 As shown, the method for obtaining the matching degree evaluation index includes:

[0154] According to the predicted load data and the electrical parameters of the transformer, calculate the total heat source power of the winding copper loss and the core iron loss;

[0155] Calculate the ratio of the cooling capacity index obtained from the digital twin model to the total heat source power to form a basic matching degree index;

[0156] Monitor the temperature distribution of the key parts of the transformer. If the temperature difference between each part is greater than the set temperature difference, a penalty correction is made to the basic matching degree index; the key parts include the winding, the core, and the top layer of the oil tank;

[0157] Dynamically adjust the basic matching degree index in combination with the real-time weather data to obtain a dimensionless matching degree evaluation index.

[0158] The method for calculating the total heat source power of the winding copper loss and the core iron loss according to the predicted load data and the electrical parameters of the transformer includes:

[0159] The predicted load data includes real-time monitoring data of load current and voltage within the next 15 minutes to 1 hour. The electrical parameters of the transformer include winding resistance, core material characteristic parameters, and rated frequency. Among them, the winding copper loss is calculated by Joule's law; the core iron loss is calculated according to the hysteresis and eddy current loss characteristics of the core material, and the calculation method is:

[0160] ;

[0161] In the formula, is the hysteresis loss coefficient; is the eddy current loss coefficient; the hysteresis loss coefficient and the eddy current loss coefficient are determined by testing the magnetization curve of the core material; is the power grid frequency, taking 50Hz; is the maximum magnetic flux density of the core, which can be calculated from the rated voltage of the transformer and the cross-sectional area of the core. The total heat source power is the sum of the winding copper loss and the core iron loss, and is used to characterize the internal heat generation rate of the transformer under the predicted load.

[0162] The method for calculating the ratio of the cooling capacity index obtained from the digital twin model to the total heat source power to form a basic matching degree index includes:

[0163] Extract the cooling capacity index from the digital twin model in real time; the cooling capacity index includes oil circulation flow rate, fan effective heat dissipation power, and radiator surface convective heat transfer coefficient. Substitute the cooling capacity index into the heat exchange model to calculate the actual heat dissipation power; the heat exchange model comprehensively considers oil convection, fan forced heat dissipation, and radiator natural convection, and the calculation method is:

[0164] ;

[0165] In the formula, is the heat exchange model; is the oil circulation flow rate, measured by an oil flow relay; ; h is the convective heat transfer coefficient on the radiator surface, obtained by simulating the ambient wind speed and temperature through a digital twin model; is the specific heat capacity of transformer oil, taking 1900 J / (kg·K); is the temperature difference between the inlet and outlet of the oil; is the heat dissipation area of the radiator; is the temperature difference between the radiator surface and the environment. The basic matching degree index is calculated by the ratio of the heat dissipation power to the total heat source power. This index reflects the direct matching relationship between the cooling capacity and the heat dissipation demand. A value greater than 1 indicates sufficient heat dissipation capacity, and a value less than 1 indicates insufficient heat dissipation capacity.

[0166] The method of monitoring the temperature distribution of key parts of the transformer and performing a penalty correction on the basic matching degree index if the temperature difference between parts is greater than the set temperature difference includes:

[0167] The temperature sensors arranged at the hot spot position of the transformer winding, the middle column position of the iron core, and the top oil surface position of the oil tank are used to monitor the temperature of key parts in real time, and the difference between the temperature of each part and the hot spot temperature of the winding is calculated. If the temperature difference between any two parts exceeds the set temperature difference threshold, such as the temperature difference between the winding and the iron core is greater than 15°C, and the temperature difference between the winding and the top oil of the oil tank is greater than 20°C, it is determined that the temperature distribution is uneven and there is a risk of local overheating. At this time, a penalty correction is performed on the basic matching degree index. The correction formula is:

[0168] ;

[0169] In the formula, is the basic matching degree index after penalty correction; is the basic matching degree index; is the correction coefficient, with a value of 0.02 °C⁻², determined through a thermal stability experiment; is the variance of the temperature of key parts, used to quantify the unevenness of the temperature distribution. The larger the variance, the greater the correction amplitude.

[0170] The method of dynamically adjusting the basic matching degree index by combining real-time weather data to obtain a dimensionless matching degree evaluation index includes:

[0171] Corresponding adjustment strategies are adopted for different weather types: when the ambient temperature is greater than 35°C, the air density decreases, resulting in a decrease in the fan heat dissipation efficiency. Calculate the temperature correction factor according to the formula In the formula is the real-time ambient temperature; is the temperature correction factor; when the wind speed is greater than 10 m / s, strong wind enhances the convective heat transfer on the radiator surface, and the wind speed correction factor is calculated; in the formula, is the real-time wind speed; is the wind speed correction factor; when the relative humidity is greater than 85%, the change in the thermal conductivity of the insulating material affects the internal heat conduction, and the humidity correction factor is calculated; in the formula, is the relative humidity; is the humidity correction factor. The comprehensive correction factor , and the final matching degree evaluation index is the ratio of the basic matching degree index after penalty correction to the comprehensive correction factor. This index is dimensionless, and it unifies and quantifies the heat dissipation matching degree, temperature uniformity, and environmental impact.

[0172] It should be noted that the layout positions of the above key part temperature sensors are determined according to the transformer thermal design specifications to ensure that the highest temperature points and typical heat dissipation nodes can be captured; the set temperature difference threshold is determined by referring to the standard of GB / T 1094.2 - 2013 "Power Transformers - Part 2: Temperature Rise of Liquid-Immersed Transformers" and combining the thermal stability parameters provided by the equipment manufacturer; the acquisition frequency of weather data is synchronized with the predicted load data (1 minute / time) to ensure the real-time nature of dynamic adjustment. The coefficients in each correction formula are verified through historical operation data fitting and simulation experiments. For example, the temperature correction factor is trained with 1000 groups of high-temperature working condition data, so that the correlation degree between the corrected matching degree index and the actual overheating fault reaches more than 92%. The matching degree evaluation index obtained by this method can accurately reflect the real-time state of the transformer heat dissipation system and provide a quantitative basis for the intelligent adjustment of the cooling strategy.

[0173] Intelligent decision-making module: If the matching degree evaluation index exceeds the set threshold, the device status, operation data, and weather data collected in real time are input into the pre-trained cooling strategy model to obtain the cooling strategy.

[0174] The training method of the cooling strategy model includes:

[0175] Construct a multi-dimensional state space including transformer device status, operation data, and weather data;

[0176] Construct an action space of discrete adjustment methods including fan speed regulation, standby equipment switching, and oil pump power regulation;

[0177] Adopt the proximal policy optimization algorithm to construct a policy network; the policy network takes the multi-dimensional state space as the input and the action probability distribution of the action space as the output;

[0178] Design a composite reward function that includes safety rewards, energy consumption rewards, and lifespan rewards;

[0179] Based on the composite reward function, iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the adjustment method selected by the policy network.

[0180] The method for constructing a multi-dimensional state space that includes transformer equipment status, operation data, and weather data includes:

[0181] The state space contains three types of feature dimensions: transformer equipment status, operation data, and weather data. Among them, the transformer equipment status includes winding hot spot temperature, core column temperature, top oil temperature of the oil tank, and operation status of the cooling equipment; the operation data includes load current, winding voltage, oil circulation flow rate, and pressure difference between the inlet and outlet of the radiator; the weather data includes ambient temperature, wind speed, relative humidity, and precipitation. The data collection frequency is synchronized with the cooling strategy adjustment cycle and is set to 1 minute / time to ensure the real-time nature of the state information.

[0182] The method for constructing an action space of discrete adjustment methods that includes fan speed adjustment, switching of standby equipment, and oil pump power adjustment includes:

[0183] The action space is defined as a finite set that includes discrete adjustment methods, specifically including: ① Fan speed adjustment, setting three adjustment modes, low speed 500 r / min, medium speed 1000 r / min, and high speed 1500 r / min; ② Switching of standby equipment, including standby fans and standby oil pumps; There are 2 groups of standby fans in total, and the status is "0 - off" or "1 - on"; There is 1 group of standby oil pumps in total, and the status is "0 - off" or "1 - on"; ③ Oil pump power adjustment, divided into 5 levels according to the output frequency of the oil pump frequency converter, which are 20 Hz, 30 Hz, 40 Hz, 50 Hz, and 60 Hz, corresponding to powers of 10 kW, 15 kW, 20 kW, 25 kW, and 30 kW. The above discretization process facilitates the precise execution of control equipment in engineering practice and covers the main adjustment means of the transformer cooling system.

[0184] The method for constructing a policy network using the Proximal Policy Optimization algorithm includes:

[0185] The policy network is a three-layer fully connected neural network. The number of neurons in the input layer is the same as the dimension of the state space, a total of 15 dimensions. The hidden layer is set with two network layers with 128 neurons each. The activation function uses the ReLU function to introduce non-linear mapping ability. The number of neurons in the output layer is the same as the dimension of the action space, a total of 10 discrete actions, and the probability distribution of each action is output through the Softmax function. The network parameters are initialized using the Xavier initialization method to ensure that the variances of the activation values of each layer are consistent and avoid the problems of gradient disappearance or explosion.

[0186] The method for designing a composite reward function including safety reward, energy consumption reward, and lifespan reward includes:

[0187] The composite reward function is specifically defined as:

[0188] ;

[0189] In the formula, is the composite reward function; is the safety reward; is the energy consumption reward; is the lifespan reward; , and are weight coefficients; the safety reward is designed based on the matching degree evaluation index and satisfies the following relationship:

[0190] ;

[0191] It should be noted that when the matching degree evaluation index is less than 0.8, it indicates insufficient heat dissipation capacity; when the matching degree evaluation index is greater than 1.2, it indicates redundant heat dissipation capacity; this design prompts the policy network to prioritize ensuring the safe operation of the transformer. The energy consumption reward is calculated based on the real-time power consumption of the cooling equipment and is the product of the real-time power consumption of the cooling equipment and -0.01, guiding the policy network to select a regulation method with lower energy consumption. The lifespan reward considers the number of starts and stops of the equipment and the operating duration, sets a single start-stop penalty of -5 for the switching operation of standby equipment, and sets a penalty of -2 per hour for the high-speed operating state of the oil pump and fan to delay equipment aging. The weight coefficients takes 0.6, takes 0.3, takes 0.1, which is determined through expert experience and historical fault data statistics to ensure that the priority of the safety goal is higher than that of the energy consumption and lifespan goals.

[0192] It should be noted that the layout positions of the temperature sensors in the state space are determined according to the standard JB / T 8450-2016 "Winding Temperature Controllers for Transformers" to ensure accurate monitoring of the temperatures of key parts; the adjustment accuracy of the action space matches the actual control accuracy of the cooling equipment to ensure the executability of the policy output; the thresholds and coefficients in the composite reward function are optimized through 1000 hours of hardware-in-the-loop simulation testing, enabling the energy consumption of the trained policy to be reduced by more than 15% under safety constraints and the equipment lifespan to be extended by 20%.

[0193] The method for iteratively optimizing the parameters of the policy network based on the composite reward function to maximize the cumulative reward value of the regulation method selected by the policy network includes:

[0194] Use a digital twin model to simulate extreme weather scenarios and generate device states under different extreme weather scenarios;

[0195] Based on the current policy state, the policy network selects a regulation method from the action space;

[0196] The digital twin model simulates the operation of the device according to the selected regulation method, outputs the device state at the next unit time, and calculates the reward value of the corresponding composite reward function;

[0197] Store the device state at the next unit time and the reward value of the corresponding composite reward function in the experience pool;

[0198] Evaluate the relative value of each selected regulation method through the Generalized Advantage Estimation method;

[0199] Based on the data in the experience pool and the relative value of the regulation method, use the gradient descent algorithm to iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the regulation method selected by the policy network.

[0200] The method for using a digital twin model to simulate extreme weather scenarios and generate device states under different extreme weather scenarios includes:

[0201] Input extreme weather parameters, such as ambient temperature of 40°C, wind speed of 15 m / s, relative humidity of 90%, and initial device state, such as load current of 1000 A, winding hot spot temperature of 85°C, and fan speed of 1000 r / min. The digital twin model calculates the internal oil flow distribution, temperature field change, and cooling equipment power consumption of the transformer by solving the Navier-Stokes equation and Fourier heat conduction equation, outputs a device state vector containing 15-dimensional state parameters, and sets the simulation time step to 1 minute, which is consistent with the cooling strategy adjustment period.

[0202] The method for the policy network to select a regulation method from the action space based on the current policy state includes:

[0203] The policy network adopts the ε-greedy policy to select specific regulation actions: select the action with the highest probability with a 90% probability and randomly select other actions with a 10% probability, which not only ensures the exploration of the policy but also takes into account the exploitation. For example, when the calculated selection probability of the high-speed fan (1500 r / min) is 60%, the action may be directly selected according to the policy or other actions may be randomly selected.

[0204] The method for the digital twin model to simulate the operation of the device according to the selected regulation method, output the device state at the next unit time, and calculate the reward value of the corresponding composite reward function includes:

[0205] Input the selected adjustment actions, such as starting a group of standby fans and adjusting the oil pump power to 40 Hz, as boundary conditions into the digital twin model, update the operating parameters of the cooling equipment, and re-solve the multi-physics coupling equations to obtain the equipment state vector for the next unit time. At the same time, calculate the composite reward function value based on the real-time equipment power consumption, matching degree evaluation index, and equipment start-stop times.

[0206] The method for evaluating the relative value of each selected adjustment method through the Generalized Advantage Estimation (GAE) method includes:

[0207] Store the next state and the reward value in the experience pool. The experience pool is a circular buffer with a capacity of 100,000 entries. Each data record contains the current state, selected action, reward value, and next state. When storing, adopt the prioritized experience replay strategy, giving higher priority to samples with larger absolute reward values, such as negative rewards when the heat dissipation capacity is insufficient, to increase the training weight of important data and avoid the gradient update being dominated by ordinary samples.

[0208] Evaluate the relative value of the adjustment method through the Generalized Advantage Estimation (GAE) method. Calculate the advantage function to measure the quality of the current action compared to the average policy. The formula is:

[0209] ;

[0210] where, is the advantage function; s is the current time step, representing the current moment for evaluating the advantage; m is the future step offset, representing the offset of the future number of steps, traversing each step from s + 1 to f; f is the truncated future step, representing the farthest future moment considered when calculating the advantage; is the discount factor, with a value of 0.99, reflecting the present value decay of future rewards; is the advantage estimation parameter, with a value of 0.95, used to balance bias and variance; is the state value function at state and can be estimated by a separately trained value network. is the immediate reward, that is, the instantaneous feedback obtained at the next time step s + m + 1 after performing the action at time step s + m; is the state value function at state This method effectively reduces the variance of reward estimation and improves the stability of policy evaluation by accumulating future rewards and subtracting the current state value.

[0211] Based on the data in the experience pool and the relative value of the adjustment method, the method for iteratively optimizing the parameters of the policy network using the gradient descent algorithm to maximize the cumulative reward value of the adjustment method selected by the policy network includes:

[0212] Use the gradient descent algorithm to minimize the objective function :

[0213] ;

[0214] in, is the current policy network; For the old policy network, it is updated every 10 iterations; For the clipping parameter, we set it to 0.2 to prevent the policy update from being too large and causing performance degradation; As the clipping function, the probability ratio of the new and old policy actions is limited to [ ] within the scope; is the adjustment method in the action space; is the state feature in the state space; is the expectation operator, which represents the state-action pair sampled from the experience pool D Calculate the expected value. The optimizer uses the Adam algorithm with an initial learning rate of 3×10⁻ 4 The reward decays by a factor of 0.95 every 50 training cycles to ensure rapid convergence in the early stages and fine-tuning in the later stages. Termination conditions are when the cumulative reward value fluctuation is less than 1% for 100 consecutive cycles or when the maximum number of training cycles is 2000, balancing training accuracy and computational efficiency.

[0215] It should be noted that the simulation accuracy of the digital twin model is calibrated using historical operating data, and the oil temperature prediction error in extreme scenarios does not exceed ±2°C, ensuring the engineering reliability of the generated equipment state. The exploration rate of the ε-greedy strategy is set at 10% at the beginning of training and linearly decays to 5% with increasing iterations, gradually transitioning from the exploration phase to the development phase. The discount factor and advantage parameter in the generalized advantage estimation are determined through 1000 hours of hardware-in-the-loop testing, keeping the reward evaluation error within 5%. The priority calculation of the prioritized experience replay uses TD error absolute value sorting, increasing the training frequency of key samples by more than 30%. Through the above method, the policy network can gradually learn a cooling adjustment strategy with lower energy consumption and longer equipment life under safety constraints. The cumulative reward value of the trained model in typical extreme scenarios is significantly improved compared to the initial random strategy, achieving optimal control of the cooling system.

[0216] The closed-loop control module is used to verify that there are no abnormalities in the equipment status and then generate and execute adjustment instructions based on the cooling strategy. During the execution of the adjustment instructions, the calibration parameters of the digital twin model are dynamically corrected based on the real-time collected equipment status and operating data.

[0217] Methods for dynamically correcting the calibration parameters of the digital twin model include:

[0218] During the execution of the cooling strategy adjustment instructions, transformer operation data and equipment status are collected in real time as feedback data for dynamic correction;

[0219] Compare the real-time prediction results of the digital twin model with the feedback data, and calculate the deviation between the real-time prediction results and the feedback data;

[0220] If the deviation exceeds the set deviation, identify the calibration parameters that affect the prediction accuracy, and use the gradient descent algorithm to iteratively adjust the calibration parameters according to the direction and magnitude of the deviation;

[0221] After adjustment, verify the prediction accuracy of the corrected digital twin model within the subsequent set period. If the deviation converges to the set deviation threshold within the subsequent set period, confirm that the parameter correction is effective and update the digital twin model; if the deviation does not converge to the set deviation threshold within the subsequent set period, enter the next round of correction until the deviation converges to the set deviation threshold.

[0222] The method of comparing the real-time prediction results of the digital twin model with the feedback data and calculating the deviation between the real-time prediction results and the feedback data includes:

[0223] The digital twin model outputs the predicted oil temperature distribution in real time, including 3 key positions: winding, iron core, and oil surface, oil circulation flow rate, and power consumption of cooling equipment. The deviation calculation uses the root mean square error method. The set deviation threshold is: the oil temperature prediction error exceeds 2°C, the oil circulation flow rate error exceeds 15%, and the cooling equipment power error exceeds 10%. This threshold is determined according to the transformer thermal protection setting value and the equipment operation accuracy requirements, and is used to judge whether the model prediction deviates from the actual operation state.

[0224] If the deviation exceeds the set deviation, the method of identifying the calibration parameters that affect the prediction accuracy and using the gradient descent algorithm to iteratively adjust the calibration parameters according to the direction and magnitude of the deviation includes:

[0225] The calibration parameters are the convective heat transfer coefficient and the oil circulation resistance coefficient directly related to the heat exchange efficiency in the digital twin model. The two affect the prediction of oil temperature distribution and oil flow state through the multi-physics field coupling model. The gradient descent algorithm is used to adjust the parameters, the objective function is the deviation value of the current period, and the parameter update formula is:

[0226] ;

[0227] In the formula, is the value of the parameter to be adjusted at iteration step t; for the value at iteration step 1, that is, is the convective heat transfer coefficient; the value at iteration step 2, that is, is the oil circulation resistance coefficient; is the learning rate, set to 0.01, which can be determined through simulation experiments to avoid parameter oscillation; is the gradient of the objective function with respect to the parameters. The adjustment direction is determined according to the sign of the deviation: if the model predicts that the oil temperature is higher than the measured value, that is, the heat dissipation capacity is underestimated, then increase the convective heat transfer coefficient or decrease the oil circulation resistance coefficient; if the model predicts that the oil temperature is lower than or equal to the measured value, then decrease the convective heat transfer coefficient or increase the oil circulation resistance coefficient. The adjustment amplitude is positively correlated with the absolute value of the deviation, and the single adjustment amount does not exceed 10% of the current value of the parameter to ensure the stability of the adjustment process.

[0228] After adjustment, verify the prediction accuracy of the corrected digital twin model in the subsequent set period. Set the verification period to 3 consecutive control periods. If the deviations of all monitored parameters converge to the set threshold within 3 periods, it is confirmed that the parameter correction is effective, and the optimal parameter combination is updated to the digital twin model; if not up to the standard, keep the original parameters and enter the next round of correction until the deviation converges. This verification mechanism ensures the effectiveness of parameter adjustment and avoids mis-correction caused by short-term fluctuations.

[0229] It should be noted that before calculating the deviation, the measured data needs to be filtered by moving average to eliminate high-frequency noise interference; the gradient descent algorithm uses batch gradient descent, and all monitored data of the current period is used for each iteration to ensure the accuracy of gradient calculation; setting the parameter adjustment amplitude limit is to prevent the oscillation of the simulation results caused by the drastic change of the model parameters, which conforms to the principle of giving priority to stability in engineering control. Through the above methods, the parameter drift in the digital twin model caused by equipment aging and environmental changes can be corrected in real time, ensuring that the model prediction accuracy always meets the requirements of engineering applications and providing a reliable model support for the precise adjustment of the cooling strategy.

[0230] Embodiment 2

[0231] This embodiment provides a transformer intelligent regulation system, further including:

[0232] A pre-response and adaptive module, which is used to start the equipment health detection and redundancy strategy adjustment in advance in combination with meteorological warnings, and at the same time regularly update the time-varying parameters of the model and optimize the algorithm strategy library to dynamically balance the heat dissipation capacity and the equipment life.

[0233] The method of starting the equipment health detection and redundancy strategy adjustment in advance in combination with meteorological warnings includes:

[0234] Obtain the meteorological forecast data for the next 72 hours through the meteorological warning interface of the meteorological station. The meteorological forecast data includes extreme weather types, expected arrival times, environmental temperature peaks, and wind speed peaks;

[0235] When an extreme weather warning is detected, start the equipment health detection process; if it is determined that there are health hazards in the cooling equipment, generate a redundancy strategy trigger signal;

[0236] Based on the redundant strategy trigger signal and the meteorological warning level, call the redundant strategy library to generate pre-adjustment instructions;

[0237] During the pre-response execution process, collect device response data in real time and calculate the pre-adjustment effect index; if the pre-adjustment effect index is less than the preset value, automatically upgrade the redundant strategy.

[0238] The device health detection process includes:

[0239] Exemplarily, call the historical data of the vibration amplitude of the cooling device per minute in the past 7 days, calculate the mean value and standard deviation of the vibration amplitude. If the sum of the mean value of the vibration amplitude and twice the standard deviation is greater than the bearing wear warning threshold, it is determined that there is a health hazard in the cooling device, and a redundant strategy trigger signal is generated.

[0240] The method for generating the pre-adjustment instructions is as follows:

[0241] Exemplarily, if the meteorological warning is a high-temperature warning and the device health status is abnormal, start the standby fan rotation strategy 3 hours in advance to transfer the main fan load to the standby device; if the meteorological warning is a typhoon warning, lock the fan blade angle at 0 2 hours in advance and increase the oil pump power to bipolar operation to reduce the physical impact of strong winds on the cooling system.

[0242] The calculation of the pre-adjustment effect index The method is:

[0243] ;

[0244] In the formula, is the oil temperature before adjustment; is the peak environmental temperature; is the oil temperature after pre-adjustment; is the current environmental temperature.

[0245] It should be noted that this index reflects the suppression ability of the pre-response to the temperature rise in extreme weather, and its value range is from 0 to 1. The closer it is to 1, the better the pre-adjustment effect.

[0246] The method for regularly updating the time-varying parameters of the model and optimizing the algorithm strategy library to dynamically balance the heat dissipation capacity and the device life includes:

[0247] Obtain the daily average environmental temperature and the average load rate, and update the air density and transformer oil viscosity in the digital twin model;

[0248] Based on the adjustment records in the past preset time, use the online learning algorithm to update the weight of the composite reward function of the policy network;

[0249] Divide historical extreme scenarios into a preset number of typical operating condition clusters through cluster analysis, evaluate the policy effects of the adjustment data within each operating condition cluster, extract the optimal action sequence, and update it to the policy library;

[0250] Output an adaptive adjustment report containing the execution results of the pre-response policy, the update records of the model time-varying parameters, and the policy library optimization log.

[0251] The method for updating the composite reward function weights of the policy network includes:

[0252] Exemplarily, based on the adjustment records of the past 24 hours, if the emergency policy is frequently triggered under high-temperature scenarios, the safety reward weight is automatically increased to 0.7 to strengthen the safety finite logic.

[0253] The method for dividing historical extreme scenarios into a preset number of typical operating condition clusters through cluster analysis, evaluating the policy effects of the adjustment data within each operating condition cluster, extracting the optimal action sequence, and updating it to the policy library includes:

[0254] Exemplarily, the algorithm policy library is optimized once a month. Through cluster analysis, historical extreme scenarios, such as "high temperature + high load" and "typhoon + sudden load recovery", are divided into 5 typical operating condition clusters, and the optimal action sequence is extracted; for the "typhoon + sudden load" operating condition, if based on historical data, the strategy of "starting the standby fan first and then gradually increasing the main fan speed" has lower energy consumption and comparable oil temperature control effect than the strategy of "turning on all equipment", then this sequence is set as the default strategy for this operating condition.

[0255] It should be noted that the pre-response and adaptive module transforms passive response into active defense through the full-chain mechanism of "meteorological warning - health detection - redundant adjustment - parameter adaptation - policy evolution". While improving the adaptability of equipment to extreme environments, it continuously enhances the long-term operation efficiency of the system through periodic parameter updates and policy optimizations.

[0256] Embodiment 3

[0257] Please refer to Figure 2 As shown, this embodiment provides a transformer intelligent adjustment method, including:

[0258] Input the real-time collected transformer operation data and weather data into a pre-trained load prediction model to obtain predicted load data;

[0259] Input the collected physical parameters and operation data, as well as the preset calibration parameters, into a pre-constructed digital twin model to obtain the cooling capacity index;

[0260] Calculate the total heat source power of copper loss and iron loss based on the predicted load data and the electrical parameters of the transformer collected in real time, and combine the cooling capacity index. Dynamically quantify the balance state between the total heat source power and the cooling capacity index through a matching degree evaluation function to obtain a matching degree evaluation index;

[0261] If the matching degree evaluation index exceeds the set threshold, input the device status, operation data, and weather data collected in real time into the pre-trained cooling strategy model to obtain a cooling strategy;

[0262] After verifying that the device status is normal, generate and execute an adjustment instruction according to the cooling strategy; during the execution of the adjustment instruction, dynamically correct the calibration parameters of the digital twin model based on the device status and operation data collected in real time.

[0263] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0264] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A transformer intelligent regulation method, characterized in that, Including: Input the real-time collected operation data and weather data of the transformer into a pre-trained load prediction model to obtain predicted load data; Input the collected physical parameters and operation data, as well as pre-set calibration parameters, into a pre-constructed digital twin model to obtain the cooling capacity index; Calculate the total heat source power of copper loss and iron loss based on the predicted load data and the electrical parameters of the transformer collected in real time, and combine with the cooling capacity index. Through a matching degree evaluation function, dynamically quantify the balance state between the total heat source power and the cooling capacity index to obtain a matching degree evaluation index; If the matching degree evaluation index exceeds the set threshold, input the device state, operation data, and weather data collected in real time into a pre-trained cooling strategy model to obtain a cooling strategy; After verifying that the device state is normal, generate and execute adjustment instructions according to the cooling strategy; During the execution of the adjustment instructions, dynamically correct the calibration parameters of the digital twin model based on the device state and operation data collected in real time.

2. The intelligent adjustment method of a transformer according to claim 1, wherein, The training method of the load prediction model includes: Collect the operation data and weather data of the transformer during historical extreme weather periods and mark them as training data; Perform normalization preprocessing on the training data and use it as the model input sample, and use the actual load data during the historical extreme weather period as the supervision signal; Construct a load prediction model using a long short-term memory neural network; the long short-term memory neural network includes an input layer, multiple hidden layers, and a fully connected output layer; the dimension of the input layer is consistent with the feature dimension of the preprocessed model input sample; the fully connected output layer outputs the predicted load data for a future preset time period and marks the load surge points in the predicted load data; Aim at minimizing the mean square error between the predicted load and the supervision signal, and iteratively optimize the parameters of the load prediction model through the backpropagation algorithm to obtain the final load prediction model.

3. A method for intelligent adjustment of a transformer according to claim 2, characterized in that, The method for judging whether a data point in the predicted load data is a load surge point includes: Construct a multi-dimensional feature space including time features, load historical trend features, and extreme meteorological features; the time features include periodicity indicators of daily cycle, weekly cycle, and monthly cycle; the load historical trend features include statistical indicators of historical load mean, historical load standard deviation, and historical load change rate; the extreme meteorological feature is weather data and its change rate; Based on the multi-dimensional feature space, calculate the feature values of each data point in the predicted load data in turn to construct a feature value set; the feature values include load growth rate, deviation from the historical load mean, and correlation with weather data; Calculate the matching degree between each feature value in the feature value set and the preset surge point feature value; if the matching degree exceeds the set matching degree threshold, determine the corresponding data point as a load surge point and mark the corresponding extreme scenario type; if the matching degree does not exceed the set matching degree threshold, determine the corresponding data point as a normal data point; the extreme scenario types include high temperature, high pressure, and strong wind; Verify whether the data points determined as load surge points in the predicted load data meet the persistence conditions of load surge.

4. A method for intelligent adjustment of a transformer according to claim 1, characterized in that, The construction method of the digital twin model includes: Based on the collected three-dimensional geometric parameters inside the transformer, a three-dimensional geometric model is constructed; The collected physical parameters and historical operation data of the transformer are used as the input of the digital twin model; the physical parameters include material thermal characteristic parameters, oil viscosity-temperature relationship, and air density-temperature relationship; Using the computational fluid dynamics method, coupling the Navier-Stokes equation and the Fourier heat conduction equation, a multi-physics field coupling model is established; Introduce the particle swarm optimization algorithm, and minimize the root mean square error between the oil temperature distribution predicted by the multi-physics field coupling model and the measured oil temperature distribution in the historical operation data as the optimization goal, and calibrate the calibration parameters in the multi-physics field coupling model; the calibration parameters include the convective heat transfer coefficient and the oil circulation resistance coefficient; The digital twin model is jointly composed of the multi-physics field coupling model and the three-dimensional geometric model.

5. The intelligent adjustment method of a transformer according to claim 4, characterized in that, The method for calibrating the calibration parameters in the multi-physics field coupling model includes: Construct a particle swarm optimization space; each particle in the particle swarm optimization space is defined as a two-dimensional vector composed of the convective heat transfer coefficient and the oil circulation resistance coefficient; Input the parameter vectors of each particle into the multi-physics field coupling model, perform numerical simulation in combination with historical operation data, and obtain the predicted oil temperature distribution; compare the predicted oil temperature distribution with the measured oil temperature distribution in the same historical period, and calculate the root mean square error as the particle fitness value; Each particle in the particle swarm optimization space adjusts its speed and position according to the combination of the minimum root mean square error parameters searched by itself and the combination of the minimum root mean square error parameters in the current population, and searches in the parameter space towards the region of the better solution; When the change amount of the global optimal fitness value in continuous preset positive integer iterations is less than the set fitness threshold, or when the maximum number of iterations is reached, the optimization process is terminated; Embed the convective heat transfer coefficient and the oil circulation resistance coefficient with the minimized root mean square error obtained by particle swarm optimization into the multi-physics field coupling model to complete the calibration of the calibration parameters in the multi-physics field coupling model.

6. A transformer intelligent adjustment method according to claim 1, characterized in that, The method for obtaining the matching degree evaluation index includes: According to the predicted load data and the electrical parameters of the transformer, calculate the total heat source power of the copper loss and the iron loss; Calculate the ratio of the cooling capacity index obtained from the digital twin model to the total heat source power to form a basic matching degree index; Monitor the temperature distribution of the key parts of the transformer. If the temperature difference between each part is greater than the set temperature difference, a penalty correction is performed on the basic matching degree index; the key parts include windings, iron cores, and the top layer of the oil tank; Dynamically adjust the basic matching degree index in combination with real-time weather data to obtain a dimensionless matching degree evaluation index.

7. A transformer intelligent adjustment method according to claim 1, characterized in that The training method of the cooling strategy model includes: Construct a multi-dimensional state space including the equipment state, operation data, and weather data of the transformer; Construct an action space of discrete adjustment methods including fan speed adjustment, standby equipment switching, and oil pump power adjustment; Use the proximal policy optimization algorithm to construct a policy network; the policy network takes the multi-dimensional state space as the input and the action probability distribution of the action space as the output; Design a composite reward function including safety reward, energy consumption reward, and life reward; Based on the composite reward function, iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the adjustment method selected by the policy network, and obtain the cooling policy model.

8. A transformer intelligent adjustment method according to claim 7, characterized in that The method for iteratively optimizing the parameters of the policy network based on the composite reward function to maximize the cumulative reward value of the adjustment method selected by the policy network includes: Use the digital twin model to simulate extreme weather scenarios and generate the device states under different extreme weather scenarios; The policy network selects an adjustment method from the action space according to the current policy state; The digital twin model simulates the device operation according to the selected adjustment method, outputs the device state at the next unit time, and calculates the reward value of the corresponding composite reward function; Store the device state at the next unit time and the reward value of the corresponding composite reward function in the experience pool; Evaluate the relative value of each selected adjustment method through the Generalized Advantage Estimation method; Based on the data in the experience pool and the relative value of the adjustment method, use the gradient descent algorithm to iteratively optimize the parameters of the policy network to maximize the cumulative reward value of the adjustment method selected by the policy network.

9. A transformer intelligent adjustment method according to claim 1, characterized in that, The method for dynamically correcting the calibration parameters of the digital twin model includes: During the execution of the adjustment instruction of the cooling policy, collect the transformer operation data and device state in real time as the feedback data for dynamic correction; Compare the real-time prediction result of the digital twin model with the feedback data and calculate the deviation between the real-time prediction result and the feedback data; If the deviation exceeds the set deviation, identify the calibration parameters that affect the prediction accuracy, and use the gradient descent algorithm to iteratively adjust the calibration parameters according to the direction and magnitude of the deviation; After adjustment, verify the prediction accuracy of the corrected digital twin model in the subsequent set period. If the deviation converges to the set deviation threshold in the subsequent set period, confirm that the parameter correction is effective and update the digital twin model; if the deviation does not converge to the set deviation threshold in the subsequent set period, enter the next round of correction until the deviation converges to the set deviation threshold.

10. A transformer intelligent adjustment method according to claim 1, characterized in that: The transformer operation data includes the vibration amplitude of the cooling equipment, the state of the transformer body, electrical parameters, oil temperature distribution, and load information; the weather data includes environmental temperature, humidity, wind speed, precipitation, and weather warning type; the electrical parameters include load current and winding resistance.

11. A transformer intelligent regulation system for implementing the transformer intelligent regulation method according to any one of claims 1-10, characterized in that, It includes: A load prediction module for inputting the real-time collected transformer operation data and weather data into a pre-trained load prediction model to obtain predicted load data; A digital twin module for inputting the collected physical parameters and operation data, as well as the pre-set calibration parameters, into a pre-constructed digital twin model to obtain the cooling capacity index; A matching degree evaluation module calculates the total heat source power of copper loss and iron loss based on the predicted load data and the real-time collected electrical parameters of the transformer, and combines the cooling capacity index to dynamically quantify the balance state between the total heat source power and the cooling capacity index through a matching degree evaluation function to obtain a matching degree evaluation index. The intelligent decision-making module, if the matching degree evaluation index exceeds the set threshold, inputs the device status, operation data, and weather data collected in real time into a pre-trained cooling strategy model to obtain a cooling strategy; The closed-loop control module is used to generate and execute adjustment instructions according to the cooling strategy after verifying that the device status is normal; During the execution of the adjustment instructions, the calibration parameters of the digital twin model are dynamically corrected based on the device status and operation data collected in real time.

Images