Method for predicting service life of transformer in high-proportion new energy access zone area based on multi-physics field modeling

Abstract

The invention discloses a method for predicting the service life of a transformer in a high-proportion new energy access zone based on multi-physics field modeling, and the method comprises the steps: building an electromagnetic field-temperature field coupling simulation model based on a COMSOL simulation platform, and setting the electrical parameters, oil physical parameters and material physical attributes of the transformer; secondly, simulating a photovoltaic access working condition by utilizing a multi-power supply superposition mode, obtaining key data such as three-phase voltage, three-phase current, iron core magnetic flux density, oil temperature and winding temperature, calculating a life label based on a working state, and performing quantitative evaluation on the state of the transformer by adopting a health state scoring method; and finally, constructing a Perii-midFormer life prediction model to realize prediction of the residual life of the transformer. Compared with a traditional life evaluation method, the method fully considers the multi-physical field dynamic coupling effect caused by photovoltaic access, improves the adaptability and reliability of life prediction, and provides an important reference for the operation and maintenance of a transformer in a power distribution network.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical engineering, and particularly relates to a method for predicting the life of a distribution transformer under high proportion of new energy access based on multi-physical field modeling. Background Technique

[0002] With the large-scale grid connection of renewable energy, the penetration rate of photovoltaic power generation systems in the power system has been increasing year by year. However, the output of photovoltaic power sources has significant intermittency and volatility, and the harmonic currents and frequency disturbances generated by them will significantly change the operating conditions of the power grid, thereby affecting the long-term reliability of key power equipment. Traditional transformer life assessment methods are mainly based on thermal aging models or empirical formulas under steady-state conditions, and do not fully consider the dynamic coupling effect of multi-physical fields caused by photovoltaic access. Therefore, studying the transformer life prediction method under photovoltaic access conditions has important theoretical value and engineering application significance for ensuring the safe and stable operation of the distribution network.

[0003] There are mainly two types of methods for predicting the remaining life of transformers: model-driven transformer remaining life prediction methods and data-driven transformer remaining life prediction methods. The former starts from the physical and chemical properties of insulating materials and establishes their failure aging models; this method establishes a specific physical model of the relationship between transformer aging characteristic quantities and transformer life through experiments, and then predicts the remaining life of the transformer through real-time monitoring data. Common model-driven prediction methods include prediction methods based on degradation mechanism models and prediction methods based on empirical degradation models. However, in real life, due to the complexity of the internal structure of transformer insulating materials, it is difficult to establish effective physical models and chemical models, and the required monitoring data is not easily obtained.

[0004] Data-driven transformer remaining life prediction methods mainly include the following three categories: health index models, failure rate models, and deep learning models. However, most of these prediction methods are used to model the operating state of transformers under traditional conditions, and do not fully consider the influence of harmonic interference, load fluctuations, and power quality problems brought by photovoltaic access on the transformer aging process. This results in the limited applicability of traditional data-driven methods in the photovoltaic access environment, and cannot accurately characterize the degradation characteristics of transformers under complex operating conditions, thereby affecting the accuracy and reliability of life prediction. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for predicting the life of a distribution transformer under high proportion of new energy access based on multi-physical field modeling, which can realize the prediction of the life of the transformer under high proportion of new energy access to ensure the reliability and safety of equipment operation.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling, comprising the following steps:

[0008] Step 1: Construct a simulation model: Based on the size data of the transformer, establish an electromagnetic - temperature - field coupling simulation model using the COMSOL simulation platform; the electromagnetic - temperature - field coupling simulation model includes the iron core, oil tank, high - voltage winding, low - voltage winding and radiator of the transformer;

[0009] Step 2: Set simulation parameters: Set the electrical parameters, oil physical properties and material physical properties of the transformer in the electromagnetic - temperature - field coupling simulation model;

[0010] Step 3: Connect new energy: Use the multi - power - superposition method to simulate the harmonics and frequency fluctuations brought by photovoltaic access;

[0011] Step 4: Collect original data: Use the electromagnetic - temperature - field coupling simulation model to simulate the transformer to obtain an original data set including three - phase voltage, three - phase current, iron - core magnetic - flux density, oil temperature and winding temperature;

[0012] Step 5: Data pre - processing: Calculate the root - mean - square values of the operating voltage and phase current using the three - phase voltage and three - phase current; construct a feature data set with the root - mean - square values of the operating voltage and phase current, iron - core magnetic - flux density, oil temperature and winding temperature, and normalize the data in the feature data set;

[0013] Step 6: Construct a training data set: Use the health - status scoring method to score the life - health status of the transformer, construct thermal - aging indicators, electrical - stress indicators and magnetic - flux degradation indicators, and introduce an operating - parameter deviation penalty mechanism; construct a training data set with the feature data set and the life - health status score;

[0014] Step 7: Model training: Construct a Peri - midFormer life - prediction model, and use the training data set to train the Peri - midFormer life - prediction model;

[0015] Step 8: Predict the life: Collect the original data of the transformer including three - phase voltage, three - phase current, iron - core magnetic - flux density, oil temperature and winding temperature, convert the original data into feature data, and input the feature data into the Peri - midFormer life - prediction model to predict the remaining service life of the transformer.

[0016] Further, in the third step, in the electromagnetic - temperature - field coupling simulation model, multiple voltage sources with different characteristics are superimposed to simulate the non - ideal voltage waveform that appears during the process of photovoltaic power generation grid - connection. The method is as follows:

[0017] Assume that the phase A, phase B, and phase C are all sinusoidal voltage sources, and their expressions are as follows:

[0018] V basc (t) = V src sin(2πft + θ)

[0019] Where: V basc (t) is the voltage of phase A, phase B, or phase C; V src is the actual voltage; f is the frequency; θ is the phase angle;

[0020] To simulate the low-order harmonics generated by a photovoltaic inverter, multiple harmonic voltage sources are added, and their expressions are as follows:

[0021]

[0022] Where: V harmonic (t) is the harmonic voltage; V h is the voltage amplitude; θ h is the phase angle; h is the harmonic order.

[0023] Furthermore, in step 4, the root mean square value of the operating voltage is:

[0024]

[0025] Where: V op is the root mean square value of the operating voltage; V a , V b and V c are the voltages of phase A, phase B, and phase C respectively;

[0026] The root mean square value of the phase current is:

[0027]

[0028] Where: I rms is the root mean square value of the phase current; I a , I b and I c are the currents of phase A, phase B, and phase C respectively.

[0029] Furthermore, in step 4, the method for normalizing the data in the feature dataset is:

[0030]

[0031] Where: x′ is the feature data after normalization; x is the original feature data; x min is the minimum value in the feature data; x max is the maximum value in the feature data.

[0032] Furthermore, in step 6, the total scoring formula for scoring the life health status of the transformer using the health status scoring method is:

[0033] S health =100×[α·S T +β·S E +γ·S B ]×e ―k·D

[0034] Where: S health represents the transformer life health status score; α, β and γ are weight coefficients, representing the weights of the influence of thermal, electric and magnetic fields on the health status respectively; S T is the thermal aging score; S E is the electrical stress score; S B is the flux degradation score; k is the deviation penalty factor, which controls the sensitivity of the parameters to deviation from the rated value; D is the combined deviation, which quantifies the degree to which the voltage, current and temperature parameters deviate from the rated value; and:

[0035] The thermal aging scoring formula is:

[0036]

[0037] Where: T max is the normalized hot spot temperature of the winding; T base is the reference threshold, indicating the temperature safety boundary;

[0038] The electrical stress rating formula is:

[0039]

[0040] Where: V op is the normalized operating voltage RMS value; V nom is the normalized value of the rated voltage; I rms is the normalized phase current RMS value; I nom is the normalized value of the rated current;

[0041] The flux degradation scoring formula is:

[0042]

[0043] Where: max(B i ) is the maximum value of the normalized magnetic flux density in each region of the core; μ B is the mean value of the core magnetic flux density; B sat is the normalized threshold value of the material saturation flux density;

[0044] Transformer life exponential decay model:

[0045]

[0046] Where: T r is the remaining life days, and λ is the decay rate coefficient.

[0047] Furthermore, the Peri-midFormer life prediction model captures complex periodic features through frequency domain decomposition and multi-level attention mechanisms, including:

[0048] Periodic pyramid construction for extracting multi-scale periodic components;

[0049] Periodic pyramid attention mechanism for modeling cross-level dependencies;

[0050] Periodic feature flow aggregation for integrating multi-scale information.

[0051] Furthermore, the method for constructing the periodic pyramid is as follows:

[0052] The input time series X ∈ R L×C , is decomposed into a trend term X trend and a seasonal term X s ; The seasonal term X s is subjected to a fast Fourier transform to calculate the frequency domain amplitude:

[0053] A = Avg(|FFT(X s )|)

[0054] Where: A ∈ R L represents the amplitude of each frequency; L is the length; C is the channel;

[0055] Select the top k frequencies {f1, …, f k} with the largest amplitudes, and the corresponding period lengths are used to construct the pyramid levels; Each level l contains f l periodic components

[0056]

[0057] The pyramid structure is achieved by stacking the components of each level:

[0058] P = Stack(C1, C2, …, C k )

[0059] Where: C i is the i-th level component, i = 1, 2, …, k; Stack(·) is the stack array function.

[0060] Furthermore, the principle of the periodic pyramid attention mechanism is: The inclusion relationship between components at different levels is determined by the overlap of position indices:

[0061]

[0062] Among them: Indicates the determination flag for the inclusion relationship between levels; Indicates the nth l―1 periodic component in the (l - 1)th level; Indicates an empty set, that is, there is no overlap in any position index;

[0063] For each component the attention range includes: the parent node, sibling nodes, and child nodes. Then the attention weight formula is:

[0064]

[0065] Among them: is the set of relevant components; q, k, and v are the Query, Key, and Value vectors respectively; d K is the dimension representing the key vector Key.

[0066] Furthermore, the method for aggregating the periodic feature stream is: a single path from the top to the bottom of the pyramid constitutes a periodic feature stream, expressed as:

[0067]

[0068] Among them: represents the nth i periodic component processed by the attention mechanism in the ith level, including multi-scale time features, where i = 1, 2,..., k;

[0069] Perform linear mapping and average pooling on multiple feature streams:

[0070]

[0071] Among them: Y s is the predicted output, that is, the superposition of the seasonal term X s and the trend term X trend .

[0072] Furthermore, it also includes Step Nine, which is to determine whether the predicted remaining service life of the transformer is less than the preset life threshold: if so, issue a warning; if not, execute Step Eight.

[0073] The beneficial effects of the present invention are as follows:

[0074] The method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling. First, based on the COMSOL simulation platform, combined with the structural parameters of the transformer, an electromagnetic - temperature - field coupled simulation model is constructed, and the electrical parameters, oil physical properties, and material physical properties of the transformer are set. Then, the multi - power superposition method is used to simulate the photovoltaic access condition, and key data such as three - phase voltage, three - phase current, core magnetic - flux density, oil temperature, and winding temperature are obtained. Based on the working state, a life label is calculated, and a health - state scoring method is used to quantitatively evaluate the state of the transformer. Finally, a Peri - midFormer life - prediction model is constructed, and the model is trained using the data with life labels to realize the prediction of the remaining life of the transformer. Compared with traditional life - assessment methods, the present invention fully considers the multi - physical - field dynamic coupling effect caused by photovoltaic access, improves the adaptability and reliability of life prediction, provides theoretical support and technical guarantee for intelligent operation and maintenance and life management, ensures the reliability and safety of equipment operation, and provides an important reference for the operation and maintenance of transformers in the distribution network. Description of the Drawings

[0075] In order to make the objectives, technical solutions, and beneficial effects of the present invention clearer, the following drawings are provided for the description of the present invention:

[0076] Figure 1 It is a flowchart of the method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling of the present invention;

[0077] Figure 2 It is the three - view drawing of the electromagnetic - temperature - field coupled simulation model constructed based on the transformer in the COMSOL platform;

[0078] Figure 3 It is the architecture diagram of the Peri - midFormer life - prediction model;

[0079] Figure 4 It is the result diagram of the transformer life prediction. Detailed Embodiment

[0080] The following further explains the present invention in combination with the drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the specific embodiments cited do not limit the present invention.

[0081] As Figure 1 shown, the method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling in this embodiment includes the following steps.

[0082] Step 1: Build a simulation model: Based on the size data of the transformer, use the COMSOL simulation platform to establish an electromagnetic field - temperature field coupled simulation model; the electromagnetic field - temperature field coupled simulation model includes the iron core, oil tank, high - voltage winding, low - voltage winding, and radiator of the transformer.

[0083] As shown in Table 3, based on the actual size data of the S22 - M - 1250 / 10 / 0.4 type oil - immersed transformer, use the COMSOL Multiphysics simulation platform to establish an electromagnetic field - temperature field coupled simulation model. The geometric structure of the model is as Figure 2 shown in the three - view drawings, including structures such as the iron core, oil tank, high - voltage winding, low - voltage winding, and radiator, fully restoring the three - dimensional characteristics of the actual equipment.

[0084] Table 1 Transformer structure dimensions

[0085]

[0086] Step 2: Set simulation parameters: Set the electrical parameters, oil physical properties, and material physical properties of the transformer in the electromagnetic field - temperature field coupled simulation model. Specifically, the oil physical properties of the transformer and the physical properties of each material are shown in Table 2 and Table 3.

[0087] Table 2 Transformer oil physical properties

[0088]

[0089] Table 3 Physical properties of each material

[0090]

[0091] Step 3: Connect new energy: Use the multi - power superposition method to simulate the harmonics and frequency fluctuations brought by photovoltaic access.

[0092] Specifically, in the electromagnetic field - temperature field coupled simulation model, superimpose multiple voltage sources with different characteristics to simulate the non - ideal voltage waveform that appears during the grid - connection process of photovoltaic power generation. The method is as follows:

[0093] Assume that the A - phase, B - phase, and C - phase are all sinusoidal voltage sources, and their expressions are:

[0094] V basc (t) = V src sin(2πft + θ)

[0095] where: V basc (t) is the voltage of the A - phase, B - phase, or C - phase; V src is the actual voltage, which takes the value of 1000V in this embodiment; f is the frequency, which takes the value of 50Hz in this embodiment; θ is the phase angle. In this embodiment, the phase angle of the A - phase power source The phase angle of the B-phase power supply θ = 0; the phase angle of the C-phase power supply

[0096] To simulate the low-order harmonics generated by a photovoltaic inverter, multiple harmonic voltage sources are added, and their expressions are:

[0097]

[0098] where: V harmonic (t) is the harmonic voltage; V h is the voltage amplitude; θ h is the phase angle; h is the harmonic order, as shown in Table 4.

[0099] Table 6 Harmonic parameters

[0100]

[0101] Step 4: Collect the original data: Use the electromagnetic field-temperature field coupling simulation model to simulate the transformer to obtain the original data set including three-phase voltages, three-phase currents, core magnetic flux density, oil temperature, and winding temperature.

[0102] Step 5: Data preprocessing: Calculate the root mean square values of the operating voltage and phase current using the three-phase voltages and three-phase currents; construct a feature data set with the root mean square values of the operating voltage and phase current, core magnetic flux density, oil temperature, and winding temperature, and normalize the data in the feature data set.

[0103] The root mean square value of the operating voltage is:

[0104]

[0105] where: V op is the root mean square value of the operating voltage; V a 、V b and V c are the voltages of phase A, phase B, and phase C respectively.

[0106] The root mean square value of the phase current is:

[0107]

[0108] where: I rms is the root mean square value of the phase current; I a 、I b and I c are the currents of phase A, phase B, and phase C respectively.

[0109] The method for normalizing the data in the feature data set is:

[0110]

[0111] Where: x' is the feature data after normalization, with a range of [0, 1]; x is the original feature data; x min is the minimum value in the feature data; x max is the maximum value in the feature data.

[0112] Step Six: Construct a training dataset: Use a health status scoring method to score the life health status of the transformer, construct thermal aging indicators, electrical stress indicators, and flux degradation indicators, and introduce an operating parameter deviation penalty mechanism; construct a training dataset from the feature dataset and the life health status score.

[0113] In this embodiment, the total score formula for using the health status scoring method to score the life health status of the transformer is:

[0114] S health = 100×[α·S T +β·S E +γ·S B ×e ―k·D

[0115] Where: S health represents the life health status score of the transformer; α, β, and γ are weight coefficients, respectively representing the influence weights of heat, electricity, and magnetic fields on the health status. In this embodiment, α, β, and γ take values of 0.3, 0.4, and 0.3 respectively; S T is the thermal aging score; S E is the electrical stress score; S B is the flux degradation score; k is the deviation penalty coefficient, controlling the sensitivity of the parameter deviation from the rated value; D is the comprehensive deviation, quantifying the degree of deviation of voltage, current, and temperature parameters from the rated value.

[0116] The thermal aging score formula is:

[0117]

[0118] Where: T max is the normalized winding hot spot temperature; T base is the reference threshold, which takes a value of 0.8 in this embodiment, representing the temperature safety boundary.

[0119] The electrical stress score formula is:

[0120]

[0121] Where: V op is the normalized root mean square value of the operating voltage; V nom is the normalized value of the rated voltage, which takes a value of 1 in this embodiment; I rms is the normalized root mean square value of the phase current; Inom is the normalized value of the rated current, and the value in this embodiment is 1.

[0122] The magnetic flux degradation scoring formula is:

[0123]

[0124] Where: max(B i ) is the maximum value of the normalized magnetic flux density in each region of the iron core; μ B is the average value of the magnetic flux density of the iron core; B sat is the normalized threshold of the material saturation magnetic flux density, and the value in this embodiment is 0.9.

[0125] Transformer life index decay model:

[0126]

[0127] Where: T r is the remaining life in days; λ is the decay rate coefficient, and the value in this embodiment is 0.5.

[0128] S health is obtained by weighted fusion of the thermal aging score, the electrical stress score, and the magnetic flux degradation score through dynamic weight coefficients, and an exponential deviation penalty factor is superimposed, and finally mapped to the 0-100 score interval.

[0129] Step 7: Model training: Construct a Peri-midFormer life prediction model, as Figure 3 shown. Use the training data set to train the Peri-midFormer life prediction model.

[0130] Peri-midFormer is a time series prediction model based on a periodic pyramid structure. The Peri-midFormer life prediction model captures complex periodic features through frequency domain decomposition and multi-level attention mechanisms, including:

[0131] Periodic pyramid construction, used to extract multi-scale periodic components.

[0132] Periodic pyramid attention mechanism, used to model cross-level dependencies.

[0133] Periodic feature flow aggregation, used to integrate multi-scale information.

[0134] (1) The method for constructing the periodic pyramid is:

[0135] The input time series X ∈ R L×C , is decomposed into a trend term X trend and a seasonal term X s ; for the seasonal term X sPerform a fast Fourier transform to calculate the frequency-domain amplitude:

[0136] A = Avg(|FFT(X s )|)

[0137] where: A ∈ R L represents the amplitude of each frequency; L is the length; C is the channel.

[0138] Select the top k frequencies {f1, …, f k} with the largest amplitudes, corresponding to the period lengths , and construct a pyramid hierarchy; each level l contains f l periodic components

[0139]

[0140] The pyramid structure is achieved by stacking the components of each level:

[0141] P = Stack(C1, C2, …, C k )

[0142] where: C i is the i-th level component, i = 1, 2, …, k; Stack(·) is the stack array function.

[0143] (2) The principle of the periodic pyramid attention mechanism is as follows: The inclusion relationship between components at different levels is determined by the overlap of position indices:

[0144]

[0145] where: represents the inclusion relationship determination flag between levels; represents the n l―1 -th periodic component in the (l - 1)-th level; represents the empty set, that is, there is no overlap of any position indices.

[0146] The attention range of each component includes: the parent node (the upper level), the sibling nodes (the same level), and the child nodes (the lower level). Then the attention weight formula:

[0147]

[0148] where: is the set of relevant components; q, k, and v are the Query, Key, and Value vectors respectively; d K is the dimension representing the key vector Key.

[0149] (3) The method for aggregating the periodic feature stream is as follows: A single path from the top to the bottom of the pyramid constitutes a periodic feature stream, which is expressed as:

[0150]

[0151] Where: represents the nth i periodic component after being processed by the attention mechanism in the i-th layer, which contains multi-scale time features, and i = 1, 2,..., k.

[0152] Perform linear mapping and average pooling on multiple feature streams:

[0153]

[0154] Where: Y s is the predicted output, that is, the superposition of the seasonal term X s and the trend term X trend .

[0155] Step Eight: Predict the lifespan: Collect the original data of the transformer including three-phase voltage, three-phase current, core magnetic flux density, oil temperature, and winding temperature, convert the original data into feature data, and input the feature data into the Peri-midFormer lifespan prediction model to predict the remaining service life of the transformer, as Figure 4 shown.

[0156] Step Nine: Determine whether the predicted remaining service life of the transformer is less than the preset lifespan threshold: If so, issue a warning and recommend replacing the transformer or taking maintenance measures; if not, execute Step Eight and regularly evaluate the health status.

[0157] The lifespan threshold in this embodiment is set to 8 years. Specifically, as Figure 4 shown, the predicted lifespan in this embodiment is 10150 days, which is higher than 8 years (2920 days), so continue to run and regularly evaluate the health status.

[0158] The above-described embodiments are only the preferred embodiments cited to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims

1. A method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling, characterized in that: It includes the following steps: Step 1: Construct a simulation model: Based on the size data of the transformer, establish an electromagnetic field-temperature field coupling simulation model using the COMSOL simulation platform; the electromagnetic field-temperature field coupling simulation model includes the iron core, oil tank, high-voltage winding, low-voltage winding, and radiator of the transformer; Step 2: Set simulation parameters: Set the electrical parameters, oil physical properties, and material physical properties of the transformer in the electromagnetic field-temperature field coupling simulation model; Step 3: Connect new energy: Use the multi-power superposition method to simulate the harmonics and frequency fluctuations brought by photovoltaic access; Step 4: Collect original data: Use the electromagnetic field-temperature field coupling simulation model to simulate the transformer to obtain an original data set including three-phase voltage, three-phase current, iron core magnetic flux density, oil temperature, and winding temperature; Step 5: Data preprocessing: Calculate the root mean square values of the operating voltage and phase current using the three-phase voltage and three-phase current; construct a feature data set with the root mean square values of the operating voltage and phase current, iron core magnetic flux density, oil temperature, and winding temperature, and normalize the data in the feature data set; Step 6: Construct a training data set: Use the health state scoring method to score the life health state of the transformer, construct thermal aging indicators, electrical stress indicators, and magnetic flux degradation indicators, and introduce an operating parameter deviation penalty mechanism; construct a training data set with the feature data set and life health state scoring; Step 7: Model training: Construct a Peri-midFormer life prediction model, and use the training data set to train the Peri-midFormer life prediction model; Step 8: Predict the life: Collect the original data of the transformer including three-phase voltage, three-phase current, iron core magnetic flux density, oil temperature, and winding temperature, convert the original data into feature data, input the feature data into the Peri-midFormer life prediction model, and predict the remaining service life of the transformer.

2. The method for predicting the service life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling according to claim 1, wherein: In step 3, in the electromagnetic field-temperature field coupling simulation model, multiple voltage sources with different characteristics are superimposed to simulate the non-ideal voltage waveform that appears during the grid connection process of photovoltaic power generation. The method is as follows: Suppose that phase A, phase B, and phase C are all sinusoidal voltage sources, and their expressions are: V basc (t) = V src sin(2πft + θ) Where: V basc (t) is the voltage of phase A, phase B or phase C; V src is the actual voltage; f is the frequency; θ is the phase angle; To simulate the low-order harmonics generated by the photovoltaic inverter, multiple harmonic voltage sources are added, and their expressions are: Where: V harmonic (t) is the harmonic voltage; V h is the voltage amplitude; θ h is the phase angle; h is the harmonic order.

3. The method for predicting the service life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to claim 1, characterized in that: In step 4, the root mean square value of the operating voltage is: Where: V op is the root mean square value of the operating voltage; V a , V b and V c are the voltages of phase A, phase B and phase C respectively; The root mean square value of the phase current is: Where: I rms is the root mean square value of the phase current; I a , I b and I c are the currents of phase A, phase B, and phase C, respectively.

4. The method for predicting the service life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to claim 1, characterized in that: In step 4, the method for normalizing the data in the feature data set is: where: x′ is the feature data after normalization; x is the original feature data; x min is the minimum value in the feature data; x max is the maximum value in the feature data.

5. The method for predicting the service life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to claim 1, wherein: In step 6, the total scoring formula for using the health state scoring method to score the life health state of the transformer is: S health = 100 × [α·S T + β·S E + γ·S B × e ―k·D Where: S health represents the transformer life health status score; α, β, and γ are weight coefficients, representing the influence weights of heat, electricity, and magnetic fields on the health status respectively; S T is the thermal aging score; S E is the electrical stress score; S B is the magnetic flux deterioration score; k is the deviation penalty coefficient, controlling the sensitivity of the parameter deviation from the rated value; D is the comprehensive deviation, quantifying the degree of deviation of voltage, current, and temperature parameters from the rated value; and: The thermal aging scoring formula is: Where: T max is the hot-spot temperature of the winding after normalization; T base is the reference threshold, representing the temperature safety boundary; The electrical stress scoring formula is: Where: V op is the root mean square value of the operating voltage after normalization; V nom is the normalized value of the rated voltage; I rms is the root mean square value of the phase current after normalization; I nom is the normalized value of the rated current; The magnetic flux degradation scoring formula is: where: max(B i ) is the maximum value of the normalized magnetic flux density in each region of the iron core; μ B is the average value of the magnetic flux density of the iron core; B sat is the normalization threshold of the material saturation magnetic flux density; Transformer life exponential decay model: Where: T r is the remaining life in days, and λ is the decay rate coefficient.

6. The method for predicting the service life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling according to claim 1, wherein: The Peri-midFormer life prediction model captures complex periodic features through frequency domain decomposition and multi-level attention mechanisms, including: Periodic pyramid construction, used to extract multi-scale periodic components; Periodic pyramid attention mechanism, used to model cross-level dependencies; Periodic feature flow aggregation, used to integrate multi-scale information.

7. The method for predicting the life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to claim 6, wherein: The method for constructing a periodic pyramid is: Input time series \(X\in\mathbb{R}\) L×C , decomposed into a trend term \(X\) trend and a seasonal term \(X\) s ; perform a fast Fourier transform on the seasonal term \(X\) s to calculate the frequency domain amplitude: A = Avg(|FFT(X s )|) where: A ∈ R L represents the amplitude of each frequency; L is the length; C is the channel; Select the top k frequencies {f1, …, f k} with the largest amplitudes, corresponding to the period lengths Construct a pyramid hierarchy; each level l contains f l periodic components The pyramid structure is achieved by stacking components at each level: P = Stack(C1, C2, …, C k ) Where: C i is the i-th hierarchical component, i = 1, 2, …, k; Stack(·) is a stacking array function.

8. The method for predicting the life of a distribution transformer under high - proportion new - energy access based on multi - physical - field modeling according to claim 7, wherein: The principle of the periodic pyramid attention mechanism is that the inclusion relationship of components between levels is determined by the overlap of position indices: Wherein: Represents the determination flag for the inclusion relationship between levels; Represents the th periodic component in the Represents an empty set, that is, there is no overlap in any position index; Each component The attention range includes: the parent node, sibling nodes, and child nodes. Then the attention weight formula is: Wherein: is a set of related components; q, k, and v are Query, Key, and Value vectors respectively; d K is the dimension representing the key vector Key.

9. The method for predicting the service life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to claim 8, wherein: The method for aggregating the periodic feature stream is that a single path from the top to the bottom of the pyramid forms a periodic feature stream, denoted as: Wherein: represents the nth i periodic component processed by the attention mechanism in the ith layer, including multi-scale time features, where i = 1, 2, …, k; Perform linear mapping and average pooling on multiple feature streams: Where: Y s is the predicted output, i.e., the sum of the seasonal term X s and the trend term X trend .

10. The method for predicting the service life of a distribution transformer under high-proportion new energy access based on multi-physical field modeling according to any one of claims 1-9, characterized in that: It also includes Step Nine, which determines whether the predicted remaining service life of the transformer is less than the preset life threshold. If so, a warning is issued; if not, Step Eight is executed.

Images