Method for predicting micro-water content in transformer oil based on artificial neural network

Abstract

The invention discloses a method for predicting the micro-water content in transformer oil based on an artificial neural network, and the method comprises the following steps: obtaining monitoring data in the operation process of a transformer, and obtaining a monitoring data sequence; a heat flow field simulation model is constructed, and the change trend of the oil temperature gradient and the flow field distribution is obtained; on the basis of output of the heat flow field simulation model, dynamic change characteristics of the oil liquid flow field are obtained; obtaining a time-space evolution rule of moisture migration according to the dynamic change characteristics of the oil liquid flow field; and constructing a water migration-micro water content prediction model. According to the method, a three-dimensional heat flow field is reconstructed in real time according to operation parameters such as oil temperature and load current in a transformer, oil flow is finely simulated, under the constraint of a physical rule, a migration distribution rule of micro-water is accurately predicted by using a data-driven artificial neural network, a prediction model capable of adapting to complex working conditions is constructed, and a prediction result is obtained. Therefore, accurate prediction of the micro-water content in the transformer oil is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment monitoring, and in particular to a method for predicting the micro water content in transformer oil based on an artificial neural network. Background Art

[0002] Accurate prediction of the micro water content in transformer oil is a key issue to ensure the safe operation of transformers. Traditional prediction methods rely on empirical models and statistical analysis, and it is difficult to cope with the moisture migration law under complex working conditions. During the long-term operation of transformers, the internal oil will form a thermally driven convective cycle. The dynamic changes in the temperature gradient and flow field distribution lead to complex and variable moisture migration. The frequent fluctuations of the transformer load will cause sudden changes in the oil temperature, further increasing the uncertainty of moisture migration. Existing prediction models usually assume that the temperature field and flow field are constant, ignoring the drastic oil flow changes caused by load fluctuations and seasonal alternations, resulting in large deviations in prediction results, especially a serious decline in accuracy under strong convection conditions. For example, a method for predicting the micro water content in transformer oil based on an improved TRANSFORMER disclosed in CN115598327A.

[0003] In addition, Chinese patent document CN115754249A, publication (announcement) date: March 7, 2023, discloses an on-line monitoring system for the micro water content in transformer oil, including a detection module and a control module. The detection module includes an oil temperature acquisition unit, a micro water acquisition unit, and a parameter acquisition unit; the oil temperature acquisition unit is used to acquire the real-time temperature of the transformer oil; the micro water acquisition unit is used to acquire the moisture content of the transformer oil; the parameter acquisition unit is used to detect the voltage and current on the low-voltage side of the transformer; the control module includes a signal transmission unit for receiving the acquisition signals of the oil temperature acquisition unit, the micro water acquisition unit, and the parameter acquisition unit, and also includes a main control unit for processing the signals, and an adjustment unit for controlling the working state of the transformer. Its characteristics are: by real-time detecting the operating transformer equipment, the maintenance efficiency of the transformer is improved, and then the specified transformer can be repaired quickly and accurately, avoiding unnecessary dangers; its disadvantages are: First, since the oil inside the transformer is in a dynamic change environment of the oil convection field, directly collecting through the micro water acquisition unit has poor accuracy; Second, this document does not disclose how the micro water acquisition unit collects the micro water content. Summary of the Invention

[0004] To solve the current technical problems, the main purpose of the present invention is to provide a method for predicting the micro water content in transformer oil based on an artificial neural network. By analyzing the thermal flow field distribution inside the transformer and combining multi-source monitoring data, a prediction model capable of adapting to complex working conditions is constructed, so as to achieve accurate prediction of the micro water content in transformer oil.

[0005] The technical solution adopted by the present invention is: a method for predicting the micro - water content in transformer oil based on an artificial neural network, comprising the following steps: S1. Obtain the monitoring data during the operation of the transformer to generate a monitoring data sequence; the monitoring data includes oil temperature, load current, and ambient temperature; S2. Based on the monitoring data sequence, construct a heat - flow field simulation model on the three - dimensional geometric model of the transformer to obtain the changing trends of the oil - temperature gradient and the flow - field distribution; S3. Based on the output of the heat - flow field simulation model, obtain the flow velocity and flow direction of the oil at different temperature gradients to obtain the dynamic change characteristics of the oil flow field; S4. According to the dynamic change characteristics of the oil flow field, obtain the migration path and concentration distribution of water under the action of oil convection to obtain the spatio - temporal evolution law of water migration; S5. Obtain the historical monitoring data of the micro - water content in the transformer, and conduct a comparative analysis with the spatio - temporal evolution law of water migration to obtain a water - migration - micro - water - content prediction model; S6. Input the real - time monitoring data into the water - migration - micro - water - content prediction model to obtain the predicted value of the micro - water content at the current moment.

[0006] In S1: Obtain the oil - temperature value, current value, and temperature value during the operation of the transformer through sensors to form the original monitoring values; Perform data cleaning on the original monitoring values. If there are outliers, filter them using a preset threshold. If there are missing values, fill them using an interpolation algorithm; Establish a standardization processing flow based on the cleaned data values and convert the processed values into standard values; Use time - series analysis methods to perform sequence - value modeling on the standard values to generate a monitoring data sequence; If the sequence values fluctuate, use the Isolation Forest algorithm to detect outliers; According to the detection results, use a regression algorithm to predict the operating state of the transformer; Generate a monitoring data sequence.

[0007] In S2: Obtain the transformer monitoring data sequence and the three - dimensional geometric model, and use the finite - element analysis algorithm to process the data sequence to construct a heat - flow field simulation model; On the three - dimensional geometric model, based on the finite - element analysis results, initialize the boundary conditions and initial parameters of the heat - flow field simulation model to determine the solution domain of the numerical simulation; Through the numerical simulation algorithm, calculate the oil - temperature gradient distribution in the heat - flow field simulation model to obtain the spatial change trend of the oil - temperature gradient; According to the results of the oil temperature gradient distribution, the flow field analysis algorithm is used to calculate the numerical changes in the flow field distribution and obtain the spatial characteristics of the flow field distribution; Establish a correlation model between the oil temperature gradient and the flow field distribution to obtain the change trends of the oil temperature gradient and the flow field distribution.

[0008] In S3: Obtain the temperature distribution data output by the thermal flow field simulation model and establish the initial boundary conditions for the oil flow; Adopt the hydrodynamic equation to calculate the pressure field and velocity field distributions of the oil under the action of the temperature gradient; According to the pressure field and velocity field distributions, determine the flow velocity value and flow direction angle of the oil flow; Analyze the change trends of the oil flow velocity and flow direction to obtain the motion law of the oil flow in the flow field; If the temperature gradient changes, recalculate the pressure field and velocity field distributions of the oil flow; Through iterative calculation, obtain the dynamic change characteristics of the oil flow in the flow field; Adopt the machine learning algorithm to classify and predict the dynamic change characteristics of the oil flow.

[0009] In S4: Obtain the dynamic change data of the oil flow in the flow field, and combine with the physical mechanism of water migration to establish the water diffusion equation in the oil; Adopt the diffusion equation to calculate the water migration path under the action of convection and determine the spatial distribution of water in the oil; According to the results of the water spatial distribution, combined with the dynamic change characteristics of the flow field, obtain the law of the change of water concentration over time; Adopt the machine learning algorithm to classify the water migration path and determine the main migration mode; According to the migration mode and the concentration distribution data, obtain the spatio-temporal evolution law of water migration.

[0010] In S5: Obtain the historical monitoring data of the micro water content in the transformer, and use the time series analysis method to process the monitoring data to obtain the time distribution characteristics of the water content; Process the monitoring data through the spatial interpolation algorithm to determine the spatial distribution characteristics of the water content, and combine with the time distribution characteristics to analyze the spatio-temporal evolution law of water migration; Adopt the convolutional neural network to extract the characteristics of the spatio-temporal evolution law of water migration to obtain the dynamic feature vector of water migration; Compare and analyze the historical monitoring data of the micro water content with the dynamic feature vector of water migration to determine their correlation; Use the convolutional neural network to establish a non-linear mapping relationship between water migration and micro water content to obtain the water migration-micro water content prediction model.

[0011] The parameters of the moisture migration - micro - water content prediction model are optimized using the gradient descent algorithm, and the weights and biases of the convolutional neural network are adjusted to improve the prediction accuracy of the model; The optimized prediction model is applied to new monitoring data, and the prediction results of moisture migration and micro - water content are output.

[0012] In S6: Obtain real - time monitoring data and input the monitoring data into the pre - established moisture migration - micro - water content prediction model; Using the gradient boosting algorithm, calculate for the input data to obtain the predicted value of the micro - water content at the current moment; After S6, it also includes the step of iterating the prediction model: Obtain the initial predicted value of the moisture migration - micro - water content prediction model and compare it with the preset value; If the difference between the initial predicted value and the preset value exceeds the preset threshold, start the adaptive learning algorithm for model adjustment; Adopt an online adjustment method to obtain the current model parameter set and optimize the parameter combination through the iteration method; According to the optimized parameter set, update the structure of the moisture migration - micro - water content prediction model; Using the gradient descent algorithm, calculate the model accuracy value and determine whether it reaches the preset standard; If it does not reach the preset standard, readjust the migration value and water content parameters; Obtain the updated moisture migration - micro - water content prediction model.

[0013] It also includes the step of providing decision - making support for transformer operation and maintenance: Obtain the monitoring data during the operation of the transformer and standardize the data; Adopt the moisture migration - micro - water content prediction model, input the standardized operation data, and calculate the predicted value of the micro - water content; According to the output result of the prediction model, generate a trend curve of the micro - water content changing with time; If the predicted value of the micro - water content exceeds the preset threshold, trigger the early - warning mechanism and generate an abnormal report; According to the micro - water content trend curve, judge the operation state of the transformer and analyze the change law; Adopt machine - learning algorithms, combine historical data with real - time predicted values, and optimize the parameters of the prediction model; Integrate the predicted value, trend curve, and early - warning information to form an operation and maintenance decision - making support report.

[0014] The present invention has the following beneficial effects: The present invention first obtains multi-source monitoring data during the operation of the transformer and obtains a standardized data sequence through data preprocessing. Then, a thermal fluid field simulation model is constructed based on finite element analysis, the law of oil convection movement is deduced by combining the fluid dynamics equation, and a water diffusion equation is established to calculate the law of water migration. Next, a water migration-micro water content prediction model is constructed using historical micro water content data and the law of water migration by means of a convolutional neural network.

[0015] According to the operating parameters such as the internal oil temperature and load current of the transformer, this method reconstructs the three-dimensional thermal fluid field in real time, finely simulates the oil flow, and under the constraint of physical laws, accurately predicts the migration distribution law of micro water by using a data-driven artificial neural network, and constructs a prediction model that can adapt to complex working conditions, so as to realize the accurate prediction of the micro water content in transformer oil. Description of the Drawings

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a flowchart of the present invention. Detailed Embodiments

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0019] See Figure 1 , a method for predicting the micro water content in transformer oil based on an artificial neural network, comprising the following steps: S1. Obtain the monitoring data during the operation of the transformer to obtain a monitoring data sequence; the monitoring data includes oil temperature, load current, and ambient temperature.

[0020] Obtain monitoring data such as oil temperature values, current values, and temperature values during the operation of the transformer through sensors to form original monitoring values. Perform data cleaning on the original monitoring values. If there are outliers, filter them using preset thresholds. If there are missing values, fill them using interpolation algorithms. Establish a standardization processing flow based on the cleaned data values and convert the processed values into standard values. Use time series analysis methods to model the sequence values of the standard values and generate a monitoring data sequence. If the sequence values fluctuate, use the Isolation Forest algorithm to detect outliers. According to the detection results, use regression algorithms to predict the operating state of the transformer. Finally, generate a standardized monitoring data sequence for subsequent analysis.

[0021] Specifically, during the operation of the transformer, the acquisition of multi-source monitoring data such as oil temperature, load current, and ambient temperature is the key to ensuring the safe operation of the equipment. By deploying high-precision sensors, oil temperature data is collected in real time. For example, at a certain moment, the measured oil temperature is 75°C, the load current is 420A, and the ambient temperature is 28°C. These data are transmitted to the data processing center through the Internet of Things platform. In the data preprocessing module, first, the box plot algorithm is used to identify outliers. Set the upper and lower limit thresholds. For example, if the oil temperature exceeds 90°C or is lower than 30°C, it is regarded as an outlier. Use the 3σ rule to detect outliers in the load current. If the current value deviates from the mean by more than three standard deviations, it is regarded as an outlier. For missing values, the linear interpolation method is used to fill them. For example, if the oil temperature data is missing at a certain time point, based on the oil temperature values of 75°C and 78°C at the previous and subsequent time points, the missing value is calculated to be 75°C through linear calculation. After cleaning and filling, the data sequence is standardized. Using the Z-score standardization method, the oil temperature, load current, and ambient temperature are respectively converted into a standard normal distribution sequence with a mean of 0 and a standard deviation of 1.

[0022] S2. According to the monitoring data sequence, construct a heat flow field simulation model on the three-dimensional geometric model of the transformer to obtain the changing trends of the oil temperature gradient and the flow field distribution.

[0023] Obtain the transformer monitoring data sequence and the three-dimensional geometric model, use the finite element analysis algorithm to process the data sequence, and construct a heat flow field simulation model. On the three-dimensional geometric model, based on the finite element analysis results, initialize the boundary conditions and initial parameters of the heat flow field simulation model to determine the solution domain of the numerical simulation. Through the numerical simulation algorithm, calculate the oil temperature gradient distribution in the heat flow field simulation model to obtain the spatial changing trend of the oil temperature gradient. According to the oil temperature gradient distribution results, use the flow field analysis algorithm to calculate the numerical changes in the flow field distribution and obtain the spatial characteristics of the flow field distribution.

[0024] If the numerical simulation results reach the preset accuracy, extract the key parameters of the oil temperature gradient and the flow field distribution, and store the calculation results. According to the calculation results, use the regression analysis method in machine learning algorithms to establish an association model between the oil temperature gradient and the flow field distribution. Based on the association model, use the clustering algorithm to perform pattern recognition on the oil temperature gradient and the flow field distribution, and obtain the change trends of the oil temperature gradient and the flow field distribution.

[0025] Specifically, in the process of constructing the transformer thermal fluid field simulation model, first, based on the standardized monitoring data sequence, key parameters such as the transformer operating environment temperature and load current are extracted. For example, the ambient temperature is 25°C and the load current is 1500 A. Subsequently, using the finite element analysis method, the three-dimensional geometric model of the transformer is meshed, and about 500,000 mesh nodes are generated using tetrahedral elements to ensure the model accuracy. In the thermal fluid field simulation, by setting the viscosity of the transformer oil to 0.3 Pa·s and the density to 850 kg / m³, and using the Navier-Stokes equation and the energy conservation equation for numerical simulation calculations. During the calculation process, the ANSYS Fluent software solver is used, and the time step is set to 1 second and the number of iterations is set to 1000 times to ensure convergence. Through the simulation, the distribution of the oil temperature gradient inside the transformer is obtained. For example, the highest oil temperature reaches 75°C in the winding area, while the lowest oil temperature is 45°C at the bottom of the oil tank. At the same time, the flow field distribution results show that the oil flow velocity reaches a maximum of 2 m / s near the winding, while the flow velocity drops to 0.5 m / s in the oil tank edge area. Based on these results, the influence of the oil temperature gradient and the flow field distribution on the transformer's heat dissipation performance is further analyzed, providing data support for optimizing the transformer design and improving the operation efficiency.

[0026] S3. Based on the output of the thermal fluid field simulation model, obtain the flow velocity and flow direction of the oil at different temperature gradients, and obtain the dynamic change characteristics of the oil on the flow field.

[0027] Obtain the temperature distribution data output by the thermal fluid field simulation model and establish the initial boundary conditions for the oil flow. Use the fluid dynamics equation to calculate the pressure field and velocity field distributions of the oil under the action of the temperature gradient. According to the pressure field and velocity field distributions, determine the flow velocity value and flow direction angle of the oil flow. Analyze the change trends of the oil flow velocity and flow direction to obtain the motion law of the oil on the flow field. If the temperature gradient changes, recalculate the pressure field and velocity field distributions of the oil flow. Through iterative calculations, obtain the dynamic change characteristics of the oil on the flow field. Use machine learning algorithms to classify and predict the dynamic change characteristics of the oil flow.

[0028] Specifically, based on the output results of the heat flow field simulation model, first, by solving the Navier-Stokes equations and the energy equation, combined with the continuity equation, the velocity field and temperature field distributions of the oil at a specific temperature gradient are obtained. Assuming the initial oil temperature is 300 K, the left wall is heated to 350 K, and the right wall remains at 300 K. The finite volume method is used to discretize the computational domain, with a grid size of 0.1 m and a time step of 0.01 s. The flow field distribution in the steady state is calculated. Under the action of the temperature gradient, the density of the oil changes. The density of the oil near the hot wall decreases, generating buoyancy that drives the oil to move upward, forming natural convection. By calculating the Reynolds number and Grashof number, the flow state is determined to be laminar. The SIMPLE algorithm is used to solve the velocity-pressure coupling problem, and the maximum oil flow velocity is obtained as 5 m / s, with the flow direction being a clockwise cycle. Further analyzing the flow field characteristics under different temperature gradients, when the temperature of the left wall is increased to 400 K, the maximum oil flow velocity increases to 8 m / s, and the convection intensity is significantly enhanced. By analyzing the variation characteristics of the flow field with time through Fourier transform, it is found that the flow field reaches a steady state within 10 s, and the periodic pulsation frequency is 1 Hz. Combining with the vorticity equation to calculate the vortex intensity, the maximum vorticity value is obtained as 2 s⁻¹, indicating that obvious vortex structures are formed during the oil flow process. These analysis results provide a theoretical basis for optimizing the convective heat transfer performance of the oil.

[0029] S4. According to the dynamic change characteristics of the oil convection field, obtain the migration path and concentration distribution of water under the action of oil convection, and obtain the spatio-temporal evolution law of water migration.

[0030] Obtain the dynamic change data of the oil convection field, combine the physical mechanism of water migration, and establish the water diffusion equation in oil. Use the diffusion equation to calculate the water migration path under the action of convection, and determine the spatial distribution of water in the oil. According to the spatial distribution results of water, combined with the dynamic change characteristics of the convection field, obtain the law of water concentration change with time.

[0031] If the change in water concentration exceeds the preset threshold, recalculate the diffusion equation and adjust the migration path. According to the adjusted path, update the water concentration distribution data and obtain a new spatio-temporal evolution law. Use machine learning algorithms to classify the water migration paths and determine the main migration patterns. According to the migration patterns and concentration distribution data, obtain the spatio-temporal evolution law of water migration.

[0032] Specifically, according to the dynamic change characteristics of the oil convection field, combined with the physical mechanism of water migration, the migration process of water in the oil can be described by establishing the water diffusion equation in oil.

[0033] First, based on the theory of convective diffusion, the partial differential equation for water diffusion in oil is established by combining Fick's second law with the Navier-Stokes equation. The formula is as follows: ; ; In the formula: represents the water concentration in oil c with respect to time t rate of change; represents the diffusion coefficient of water in oil; represents the Laplace operator; represents the velocity vector of the oil phase; represents the velocity vector divergence of; represents the density of the oil phase; oil phase velocity vector with respect to time t rate of change; represents the pressure gradient; represents the dynamic viscosity of the oil phase; represents the external force acting on the oil phase.

[0034] Assume the oil flow rate is 1 m / s, the diffusion coefficient is 1×10⁻ 9 m² / s, and the initial water concentration is 1%. The partial differential equation is discretized by the finite difference method, with a time step of 1 s and a spatial step of 1 m, and the equation is solved using an iterative algorithm. During the calculation, the boundary condition is that the water concentration on the oil surface remains constant. The numerical simulation results show that after 100 s, the water concentration gradually increases along the flow direction in the oil, forming a gradient distribution, and the maximum concentration reaches 15%. Further analyzing the water migration path, the particle tracking algorithm is used to simulate the movement trajectories of 1000 water particles in the oil. The results show that the water particles mainly migrate along the flow direction, and the migration distance increases linearly with time. By comparing the simulation results under different flow rates and diffusion coefficients, it is found that the flow rate has a significant impact on the water migration path. When the flow rate increases to 2 m / s, the water migration speed increases by about 50%. Finally, combining the water concentration distribution and the migration path, the spatio-temporal evolution law of water in the oil is obtained, providing a theoretical basis for oil water control.

[0035] S5. Obtain the historical monitoring data of the micro water content of the transformer, and conduct a comparative analysis with the spatio-temporal evolution law of water migration to obtain a water migration-micro water content prediction model.

[0036] Obtain the historical monitoring data of the micro water content of the transformer, and use the time series analysis method to process the monitoring data to obtain the time distribution characteristics of the water content. The formula is as follows: ; In the formula: T represents the length of the time series; represents t the value of the micro water content at the moment, represents the average value, represents the variance, which is used to describe the time distribution characteristics of the micro water content.

[0037] Process the monitoring data through the spatial interpolation algorithm to determine the spatial distribution characteristics of the moisture content, and combine the time distribution characteristics to analyze the spatio-temporal evolution law of moisture migration; the formula is as follows: ; In the formula: Z(x,y) represents the interpolation result of any point; represents the weight coefficient; represents the value of the known point; represents the distance; h represents the influence radius, which is used for spatial interpolation to calculate the moisture distribution.

[0038] Use the convolutional neural network to extract the characteristics of the spatio-temporal evolution law of moisture migration to obtain the dynamic feature vector of moisture migration; the formula is as follows: ; In the formula: C(t) represents the convolutional output; represents the weight; represents the input feature; represents the convolutional kernel; represents the bias term, which is used to extract the moisture migration characteristics.

[0039] Compare and analyze the historical monitoring data of the micro water content with the dynamic feature vector of moisture migration to determine the correlation between the two; the formula is as follows: ; In the formula: R represents the correlation coefficient; represents the migration feature value; represents the water content value; and respectively represent the corresponding average values, which are used to analyze the correlation between the two.

[0040] Use the convolutional neural network to establish a non-linear mapping relationship between moisture migration and micro water content to obtain a prediction model; the formula is as follows: ; In the formula: P(t) represents the prediction output, represents the weight of the output layer, represents the weight matrix of the hidden layer, Represents the input feature vector; Represents the bias vector, which is used to establish a non - linear mapping relationship.

[0041] Furthermore, the gradient descent algorithm is used to optimize the parameters of the prediction model, adjust the weights and biases of the convolutional neural network, and improve the prediction accuracy of the model.

[0042] Apply the optimized prediction model to new monitoring data, output the prediction results of moisture migration and micro - water content, and complete the establishment of the moisture migration - micro - water content prediction model.

[0043] Specifically, the historical monitoring data of the micro - water content in the transformer can be collected by humidity sensors deployed inside the transformer. For example, a certain type of sensor records data every 10 minutes, with a recording period of one year, generating approximately 52,560 data points in total. These data include timestamps and micro - water content values, with the unit of ppm. The micro - water content usually fluctuates between 10 and 100 ppm. The spatio - temporal evolution law of moisture migration can be obtained by analyzing the humidity data at different positions inside the transformer. For example, the inside of the transformer is divided into multiple regions, and multiple sensors are set in each region to record the humidity distribution at different time points. When using a convolutional neural network (CNN) to extract the non - linear mapping relationship between moisture migration and micro - water content, first convert the spatio - temporal humidity data into a two - dimensional matrix. The rows of the matrix represent the time dimension, the columns represent the space dimension, and each element represents the humidity value at a specific time and position. Use the TensorFlow framework to build a CNN model. The input layer is a two - dimensional humidity matrix. The convolutional layer uses a 3×3 convolutional kernel, the activation function is ReLU, the pooling layer uses max - pooling, and the fully - connected layer outputs the predicted value of the micro - water content. During the training process, the mean squared error (MSE) is used as the loss function, the optimizer is selected as Adam, the learning rate is set to 0.01, and it is trained for 100 epochs. The final MSE of the moisture migration - micro - water content prediction model on the test set is 0.05, and the R² is 0.95, indicating that the model has a high prediction accuracy. Through this model, the changing trend of the micro - water content inside the transformer over time can be accurately predicted, providing a scientific basis for the maintenance and fault prevention of the transformer.

[0044] S6. Input the real - time monitoring data into the moisture migration - micro - water content prediction model to obtain the predicted value of the micro - water content at the current moment.

[0045] Obtain the oil temperature and load current data of real-time monitoring, and input them into the pre-established moisture migration - micro water content prediction model. Use the gradient boosting algorithm to calculate for the input data to obtain the predicted value of the micro water content at the current moment. Obtain the actual value of the micro water content at the same moment from the monitoring system, and conduct a comparative analysis with the predicted value. Through the mean square error algorithm, calculate the difference value between the predicted value and the actual value to determine the error range. If the error value is less than the preset threshold, it is judged that the accuracy of the prediction model meets the requirements at this moment. If the error value is greater than the preset threshold, use the random forest algorithm to optimize and adjust the model parameters. According to the optimized model parameters, recalculate the predicted value of the micro water content at the current moment to obtain the final prediction result.

[0046] Specifically, during the real-time monitoring process, the system collects the transformer oil temperature of 65°C and the load current of 120A. These data are input into the moisture migration - micro water content prediction model based on the LSTM neural network. The model is trained through historical data and can capture the dynamic influence of oil temperature and load current on the micro water content. The model first normalizes the input data, converts the oil temperature to 65, and the load current to 12, and then conducts feature extraction and time series analysis through a three-layer LSTM network. The model outputs the predicted value of the micro water content at the current moment as 25 ppm, while the system measures the actual value as 26 ppm through an online micro water sensor. The error analysis module uses the root mean square error (RMSE) as the evaluation index and calculates that the prediction error is 1 ppm and the error rate is 85%. To further improve the prediction accuracy, the system feeds the error data back to the adaptive learning module of the model to adjust the weight parameters of the LSTM network and optimize the model performance. At the same time, the system conducts a comparative analysis of the prediction result and the actual value, generates a trend chart, and shows that the change trends of the predicted value and the actual value are basically the same, indicating that the model has strong generalization ability. Based on the current error analysis result, the system automatically adjusts the sampling frequency to once per minute to ensure a high prediction accuracy even under large load fluctuations.

[0047] After S6, it also includes the step of predicting model iteration: Obtain the initial predicted value of the moisture migration - micro water content prediction model and compare it with the preset value. If the difference between the initial predicted value and the preset value exceeds the preset threshold, start the adaptive learning algorithm to adjust the model. Adopt the online adjustment method to obtain the current model parameter set and optimize the parameter combination through the iteration method. According to the optimized parameter set, update the structure of the moisture migration - micro water content prediction model. Use the gradient descent algorithm to calculate the model accuracy value and judge whether it reaches the preset standard. If it does not reach the preset standard, readjust the migration value and water content parameters for secondary optimization. Through multiple iterations of optimization, obtain the updated prediction model and complete the improvement of the model accuracy.

[0048] Specifically, in the moisture migration - micro - water content prediction model, if the accuracy of the prediction result is lower than the preset threshold, an adaptive learning algorithm is used to adjust the model parameters online. First, the weight matrix of the model is optimized by the gradient descent method. The initial learning rate is set to 1, and the number of iterations is 100. After each iteration, the value of the loss function is calculated. If the loss value does not converge, the learning rate is adjusted to 5 for further optimization. During the optimization process, regularization techniques are adopted to prevent the model from overfitting. Then, the cross - validation method is used to evaluate the optimized model to ensure its generalization ability. If the accuracy of the model on the validation set is still lower than the preset threshold, the Bayesian optimization algorithm is further used to tune the hyperparameters. The number of iterations is 50, and finally, the optimal combination of hyperparameters is selected.

[0049] Through the above iterative optimization process, the prediction accuracy of the model is gradually improved, and finally, an updated prediction model is obtained. In the whole process, the data pre - processing link uses standardization methods, such as Z - score standardization, to normalize the input features to ensure the stability of model training.

[0050] It also includes steps to provide decision - making support for transformer operation and maintenance: The prediction model is applied to the real - time prediction of the micro - water content of the transformer, and the predicted value of the micro - water content and its change trend are output to provide decision - making support for transformer operation and maintenance.

[0051] Obtain the transformer operation data, including temperature, oil pressure, and current parameters, and standardize the data. Use the updated prediction model, input the standardized operation data, and calculate the predicted value of the micro - water content. According to the output result of the prediction model, generate a trend curve of the micro - water content changing with time. If the predicted value of the micro - water content exceeds the preset threshold, trigger the early - warning mechanism and generate an exception report. According to the micro - water content trend curve, judge the operation state of the transformer and analyze the change law. Use machine - learning algorithms, combine historical data with real - time predicted values, and optimize the model parameters. Integrate the predicted values, trend curves, and early - warning information to form an operation and maintenance decision - making support report.

[0052] Specifically, in the real-time prediction of the micro water content of the transformer, key parameters such as the temperature, humidity, and oil temperature of the transformer are first collected through sensors. For example, the temperature value is 65°C, the humidity is 45%, and the oil temperature is 70°C. These data are uploaded to the cloud server through the Internet of Things gateway, and the pre-trained LSTM neural network model is used for real-time prediction. The input layer of the model receives the above parameters. After being processed by two layers of LSTM hidden layers, the predicted value of the micro water content of the transformer is output. For example, the prediction result is 15 ppm. At the same time, the model analyzes the micro water content data in the past 10 minutes through the sliding window technology and calculates its change trend. For example, the trend value is +2 ppm / min, indicating that the micro water content shows an upward trend. The system automatically generates operation and maintenance suggestions based on the predicted value and its change trend, combined with the preset operation and maintenance threshold, such as 15 ppm as the warning line, such as "It is recommended to perform oil treatment within 24 hours", and pushes the results to the mobile terminal of the operation and maintenance personnel to provide real-time decision-making support for the operation and maintenance of the transformer.

[0053] The above embodiments are only used to illustrate the present invention. The structures, connection methods, manufacturing processes, etc. of each component can all be changed. Any equivalent transformation and improvement based on the technical solution of the present invention should not be excluded from the protection scope of the present invention.

Claims

1. A method for predicting the water content in transformer oil based on artificial neural network, characterized in that: The following steps are involved: S1. Obtain monitoring data during transformer operation and generate a monitoring data sequence; the monitoring data includes oil temperature, load current and ambient temperature; S2. Based on the monitoring data sequence, a thermal flow field simulation model is constructed on the three-dimensional geometric model of the transformer to obtain the change trend of the oil temperature gradient and flow field distribution; S3. Based on the output of the thermal flow field simulation model, the flow velocity and flow direction of the oil under different temperature gradients are obtained to obtain the dynamic change characteristics of the oil flow field; S4. According to the dynamic change characteristics of the oil convection field, the migration path and concentration distribution of water under the action of oil convection are obtained, and the temporal and spatial evolution law of water migration is obtained; S5. Obtain historical monitoring data of transformer micro-water content, compare and analyze it with the temporal and spatial evolution law of water migration, and obtain a water migration-micro-water content prediction model; S6. Input the real-time monitoring data into the water migration-micro-water content prediction model to obtain the micro-water content prediction value at the current moment.

2. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S1: The oil temperature, current value and temperature value of the transformer during operation are obtained through sensors to form the original monitoring value; Data cleaning is performed on the original monitoring values. If there are abnormal values, they are filtered using the preset threshold. If there are missing values, they are filled using the interpolation algorithm. Establish a standardized processing flow based on the cleaned data values ​​to convert the processed values ​​into standard values; The time series analysis method is used to model the sequence values ​​of the standard values ​​and generate the monitoring data sequence; If the sequence value fluctuates, the isolation forest algorithm is used to detect outliers; Based on the test results, the regression algorithm is used to predict the transformer operating status; Generate monitoring data series.

3. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S2: Obtain transformer monitoring data sequence and 3D geometric model, process the data sequence using finite element analysis algorithm, and construct a thermal flow field simulation model; On the three-dimensional geometric model, based on the finite element analysis results, the boundary conditions and initial parameters of the thermal flow field simulation model are initialized to determine the solution domain of the numerical simulation; The oil temperature gradient distribution in the thermal flow field simulation model is calculated through numerical simulation algorithms to obtain the spatial variation trend of the oil temperature gradient. According to the oil temperature gradient distribution results, the flow field analysis algorithm is used to calculate the numerical changes of the flow field distribution and obtain the spatial characteristics of the flow field distribution; A correlation model between oil temperature gradient and flow field distribution is established to obtain the changing trends of oil temperature gradient and flow field distribution.

4. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S3: Obtain the temperature distribution data output by the thermal flow field simulation model and establish the initial boundary conditions for oil flow; Using fluid dynamics equations, calculate the pressure field and velocity field distribution of oil under the action of temperature gradient; According to the pressure field and velocity field distribution, the flow velocity value and flow angle of the oil flow are determined; Analyze the changing trend of oil flow velocity and flow direction to obtain the movement law of oil convection field; If the temperature gradient changes, the pressure field and velocity field distribution of the oil flow are recalculated; Through iterative calculation, the dynamic change characteristics of the oil flow field are obtained; Machine learning algorithms are used to classify and predict the dynamic change characteristics of oil flow.

5. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S4: Obtain the dynamic change data of oil convection field, combine it with the physical mechanism of water migration, and establish the water diffusion equation in oil; The diffusion equation is used to calculate the water migration path under convection and determine the spatial distribution of water in the oil; According to the results of water spatial distribution and combined with the dynamic characteristics of convection field, the law of water concentration changing with time is obtained; Machine learning algorithms are used to classify water migration paths and identify the main migration patterns; Based on the migration pattern and concentration distribution data, the spatiotemporal evolution of water migration is obtained.

6. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S5: Obtain historical monitoring data of transformer micro-water content, process the monitoring data using time series analysis methods, and obtain the time distribution characteristics of moisture content; The monitoring data is processed by spatial interpolation algorithm to determine the spatial distribution characteristics of moisture content, and combined with the temporal distribution characteristics, the spatiotemporal evolution of moisture migration is analyzed; Convolutional neural network is used to extract the features of the spatiotemporal evolution of water migration and obtain the dynamic feature vector of water migration. Compare and analyze the historical monitoring data of micro-water content with the dynamic characteristic vector of water migration to determine the correlation between the two; A convolutional neural network was used to establish a nonlinear mapping relationship between water migration and trace water content, and a water migration-trace water content prediction model was obtained.

7. The method for predicting the water content in transformer oil based on artificial neural network according to claim 6, characterized in that: The gradient descent algorithm was used to optimize the parameters of the water migration-micro-water content prediction model, and the weights and biases of the convolutional neural network were adjusted to improve the prediction accuracy of the model. The optimized prediction model is applied to new monitoring data to output the prediction results of water migration and trace water content.

8. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: In S6: Obtain real-time monitoring data and input the monitoring data into the pre-established water migration-micro-water content prediction model; The gradient boosting algorithm is used to calculate the input data and obtain the predicted value of the micro-water content at the current moment.

9. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1, characterized in that: After S6, the following steps are included for prediction model iteration: Obtain the initial prediction value of the water migration-micro-water content prediction model and compare it with the preset value; If the difference between the initial predicted value and the preset value exceeds the preset threshold, the adaptive learning algorithm is started to adjust the model; Adopt online adjustment method to obtain the current model parameter set and optimize the parameter combination through iteration method; According to the optimized parameter set, the structure of the water migration-micro-water content prediction model is updated; Use the gradient descent algorithm to calculate the model accuracy value and determine whether it meets the preset standard; If the preset standards are not met, the migration value and moisture content parameters are readjusted; An updated moisture migration-micro-water content prediction model is obtained.

10. The method for predicting the water content in transformer oil based on artificial neural network according to claim 1 or 9, characterized in that: It also includes steps to provide decision support for transformer operation and maintenance: Obtain monitoring data during transformer operation and standardize the data; The water migration-micro-water content prediction model is used to input the standardized operating data and calculate the micro-water content prediction value; According to the output results of the prediction model, a trend curve of the change of micro-water content over time is generated; If the predicted value of the trace water content exceeds the preset threshold, the early warning mechanism is triggered and an abnormality report is generated; According to the trend curve of micro-water content, the transformer operation status can be judged and the changing rules can be analyzed; Use machine learning algorithms to combine historical data with real-time forecast values ​​to optimize forecast model parameters; Integrate the predicted values, trend curves and warning information to form an operation and maintenance decision support report.

Images