Transformer direct current magnetic bias degree evaluation method based on multi-characteristic quantity fusion

Abstract

The invention discloses a transformer direct current magnetic bias degree evaluation method based on multi-characteristic quantity fusion. The method specifically comprises the following steps: firstly, synchronously acquiring a transformer grounding neutral point current signal, a box body surface vibration signal and a surrounding noise signal; secondly, extracting characteristic quantities including the average value and the low-frequency energy ratio of a current signal, the kurtosis and the harmonic energy ratio of a vibration / noise signal and the like for representing a direct-current magnetic bias state from a time domain and a frequency domain; thirdly, combining a subjective weight determined by an analytic hierarchy process and an objective weight determined by a variable coefficient method, calculating the fuzzy density of each characteristic quantity through weighted average, and solving a lambda value and fuzzy measurement; and establishing a membership function and a fuzzy evaluation matrix of the characteristic quantity to the direct-current magnetic bias degree of the transformer. And finally, fusing the multi-characteristic quantity information by adopting a fuzzy integral algorithm, and evaluating the direct current magnetic bias degree grade of the transformer according to a maximum membership degree principle.

Description

Technical Field

[0001] This invention relates to the technical field of evaluating the DC bias of transformers, and specifically to a method for evaluating the DC bias of transformers based on the fusion of multiple characteristic quantities. Background Technology

[0002] my country's power grid has a complex structure and close electrical connections, resulting in numerous points of intrusion for DC bias currents. This leads to increasingly frequent intrusions of bias currents into the power grid, including stray currents from urban rail transit, ground currents from high-voltage DC transmission systems, and geomagnetic induced currents. These intruding bias currents accelerate corrosion of metal structures such as grounding grids and buried oil / gas pipelines. Furthermore, they induce DC bias in urban power grid transformers, causing severe consequences such as increased transformer non-stationary vibration and relay protection failure, seriously jeopardizing the safety of urban infrastructure.

[0003] Accurately assessing the DC bias degree of a transformer is fundamental to obtaining its health status and implementing corresponding bias suppression measures. However, traditional DC bias degree assessment methods based on single characteristics cannot effectively address the complex operating conditions of transformers, are susceptible to load fluctuations and environmental conditions, and suffer from low reliability; once the sensor fails, the assessment of the transformer's DC bias degree cannot be carried out. Transformer DC bias degree assessment methods based on multi-characteristic fusion, by comprehensively considering multi-dimensional data information and cross-validating multi-dimensional features, possess advantages such as strong anti-interference capabilities and high assessment accuracy. However, they also face challenges such as difficulty in accurately characterizing the DC bias degree through feature selection and difficulty in fusing assessments among multiple characteristics. Therefore, it is necessary to research transformer DC bias degree assessment methods based on multi-characteristic fusion to guide the implementation of transformer DC bias response and suppression efforts.

[0004] In summary, in the relevant technologies, when evaluating the DC bias of transformers, there are technical problems such as relying on a single characteristic quantity for evaluation, which is easily affected by operating conditions, and the simple weighted fusion method cannot effectively characterize the interaction relationship between multiple characteristic quantities, thus leading to insufficient reliability and accuracy of the evaluation results. Summary of the Invention

[0005] The technical problem this invention aims to solve is that, in related technologies, the evaluation of transformer DC bias relies on a single characteristic quantity, making it susceptible to operating condition interference. Furthermore, simple weighted fusion methods cannot effectively characterize the interaction between multiple characteristic quantities, leading to insufficient reliability and accuracy of the evaluation results. The purpose is to provide a transformer DC bias evaluation method based on multi-characteristic fusion, thereby resolving the technical problem of insufficient reliability and accuracy of the evaluation results.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a method for evaluating the DC bias degree of a transformer based on the fusion of multiple characteristic quantities, including:

[0008] Acquire multi-source signals during transformer operation;

[0009] Extract multiple feature quantities from multi-source signals to characterize the degree of DC bias.

[0010] Based on the aforementioned multiple feature quantities, the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method are integrated to obtain the comprehensive weights of each feature quantity as the fuzzy density.

[0011] Based on the fuzzy density, an interaction parameter is determined to characterize the interaction relationship between features, and a fuzzy measure is calculated for any non-empty subset among multiple feature quantities based on the interaction parameter; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them.

[0012] Establish membership functions for each characteristic quantity with respect to multiple preset DC bias levels to form a fuzzy membership evaluation matrix;

[0013] For each level of biased magnetization, a fusion calculation is performed based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix to obtain the comprehensive evaluation value of each level.

[0014] The final DC bias degree of the transformer is determined based on the comprehensive evaluation values ​​of each level.

[0015] Furthermore, the step of acquiring multi-source signals during transformer operation includes:

[0016] The grounding neutral point current signal is obtained by a current sensor placed at the grounding neutral point of the transformer;

[0017] Vibration signals of the transformer enclosure are acquired by vibration sensors arranged on the surface of the enclosure.

[0018] Ambient noise signals are acquired by noise sensors placed at predetermined locations around the transformer;

[0019] The grounding neutral point current signal, enclosure vibration signal, and environmental noise signal are synchronized and timed.

[0020] Further, the multi-source signals include: grounding neutral point current signal, enclosure vibration signal, and environmental noise signal; wherein, the step of extracting multiple feature quantities for characterizing the degree of DC bias from the multi-source signals includes:

[0021] The grounding neutral point current signal, the enclosure vibration signal, and the environmental noise signal are preprocessed respectively;

[0022] From the preprocessed grounding neutral point current signal, extract its time-domain statistical features and frequency-domain energy distribution features based on power spectral density;

[0023] Temporal kurtosis features and harmonic energy ratio features based on power spectral density are extracted from the preprocessed enclosure vibration signal and the preprocessed environmental noise signal, respectively. The extracted features together constitute multiple features used to characterize the degree of DC bias.

[0024] Further, the step of fusing the subjective weights determined by the analytic hierarchy process (AHP) and the objective weights determined by the coefficient of variation method based on the multiple feature quantities to obtain the comprehensive weights of each feature quantity as the fuzzy density includes:

[0025] A judgment matrix is ​​constructed based on the relative importance of each characteristic quantity to the degree of DC bias. The subjective weights are obtained by solving the largest eigenvalue and its corresponding eigenvector of the judgment matrix and then normalizing the results.

[0026] Calculate the coefficient of variation of each feature quantity, and normalize the coefficient of variation of all feature quantities to obtain the objective weight;

[0027] By using a weighted average method, the subjective weights and the objective weights are integrated to obtain the comprehensive weights of each feature quantity, and the comprehensive weights are determined as the fuzzy density of each feature quantity; wherein, the coefficient of the weighted average is a preset preference factor, which is used to adjust the degree of preference for subjective weights or objective weights.

[0028] Further, the step of determining the interaction parameter used to characterize the interaction relationship between features based on the fuzzy density, and calculating the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameter, includes:

[0029] Based on the fuzzy density of each feature quantity, the interaction parameter is solved; wherein, the value of the interaction parameter is obtained by solving a polynomial equation with each fuzzy density as a variable; wherein, the polynomial equation is configured to be the interaction parameter plus one, which is equal to the product of the corresponding features; wherein, the product is a product of the interaction parameter multiplied by its fuzzy density.

[0030] For any non-empty subset of the plurality of features, the value of its fuzzy measure is calculated as follows: First, calculate the product of the interaction parameter and the fuzzy density corresponding to each feature in the subset. Then, divide the result of the product minus one by the interaction parameter. The quotient is the fuzzy measure value of the feature subset.

[0031] Furthermore, the step of establishing membership functions for each feature quantity with respect to multiple preset DC bias levels to form a fuzzy membership evaluation matrix includes:

[0032] For each feature quantity, the deviation under the current operating state is calculated based on its actual measured value and the preset benchmark value;

[0033] For each preset DC bias level, a trapezoidal membership function with the deviation degree as the independent variable is defined for each feature quantity, wherein the shape of the trapezoidal membership function is determined by a threshold parameter preset for the DC bias level; wherein the multiple preset DC bias levels include four levels, namely normal state, slight bias, moderate bias and heavy bias.

[0034] Based on the trapezoidal membership function, the membership degree of each characteristic quantity's deviation to each DC bias level is calculated;

[0035] Based on all the calculated membership degrees, a fuzzy membership degree evaluation matrix is ​​constructed; where the rows of the fuzzy membership degree evaluation matrix correspond to the feature quantities, the columns correspond to the degree of magnetic bias, and the matrix elements are the membership degrees of the corresponding feature quantities for that degree.

[0036] Further, the step of performing a fusion calculation for each level of biased magnetization, based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix, to obtain the comprehensive evaluation value for each level, includes:

[0037] For each level of magnetic bias, extract the membership vector corresponding to the level from the fuzzy membership evaluation matrix, and sort the membership vectors in descending order of membership value.

[0038] Based on the sorted membership vector and the fuzzy measure, perform Choquet integration to obtain a comprehensive evaluation value for the degree of magnetic bias.

[0039] The step of determining the final DC bias degree of the transformer based on the comprehensive evaluation values ​​of each level includes:

[0040] The comprehensive evaluation values ​​of all bias levels are compared, and the bias level with the highest comprehensive evaluation value is determined as the final DC bias level of the transformer.

[0041] Secondly, the present invention provides a transformer DC bias degree evaluation device based on multi-feature quantity fusion, comprising:

[0042] The acquisition module is used to acquire multi-source signals during transformer operation.

[0043] An extraction module is used to extract multiple feature quantities from multi-source signals to characterize the degree of DC bias.

[0044] The fuzzy density fusion module is used to fuse the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method based on the multiple feature quantities to obtain the comprehensive weight of each feature quantity as the fuzzy density.

[0045] The fuzzy measure determination module is used to determine the interaction parameters used to characterize the interaction relationship between features based on the fuzzy density, and to calculate the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameters; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them.

[0046] A module is established to establish the membership function of each feature quantity for multiple preset DC bias levels, so as to form a fuzzy membership evaluation matrix.

[0047] The comprehensive evaluation value calculation module is used to perform fusion calculation for each level of magnetic bias, based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix, to obtain the comprehensive evaluation value for each level.

[0048] The determination module is used to determine the final DC bias of the transformer based on the comprehensive evaluation values ​​of each level.

[0049] Thirdly, the present invention provides an electronic device, comprising: a memory, and one or more processors communicatively connected to the memory; the memory stores instructions executable by the one or more processors, the instructions being executed by the one or more processors to cause the one or more processors to implement the method described above.

[0050] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] This invention first acquires multi-source signals from transformer operation and extracts multiple feature quantities, effectively utilizing multi-dimensional information such as current, vibration, and noise, thus enhancing the information foundation and anti-interference capability of the assessment. Based on this, it integrates the subjective weights of the analytic hierarchy process (AHP) and the objective weights of the coefficient of variation method to determine the fuzzy density of each feature quantity, taking into account both expert experience and statistical data patterns, providing a reasonable initial importance measure for subsequent modeling. Furthermore, based on this fuzzy density, interaction parameters are determined and fuzzy measures of arbitrary feature subsets are calculated, thereby quantifying the possible nonlinear interaction relationships such as complementarity or redundancy between features. Subsequently, a fuzzy evaluation matrix is ​​formed by establishing a trapezoidal membership function based on deviation, achieving fair comparison of different feature quantities under a unified standard. Finally, for each bias level, the sorted membership vector is nonlinearly fused using the Choquet integral based on the fuzzy measure, effectively utilizing the interaction between features in the decision-making process and overcoming the inherent defects of the traditional linear weighted model. Therefore, this method significantly improves the reliability and accuracy of transformer DC bias degree assessment by scientifically modeling and nonlinearly fusing multiple feature quantities. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0054] Figure 1 A flowchart illustrating a method for evaluating the DC bias of a transformer based on the fusion of multiple features, as provided in the embodiments of this specification.

[0055] Figure 2 This is an architecture diagram of a transformer DC bias degree evaluation method based on multi-feature fusion provided in the embodiments of this specification;

[0056] Figure 3 This is a block diagram of a transformer DC bias evaluation device based on multi-feature fusion, provided in the embodiments of this specification.

[0057] Figure 4 This is a block diagram of an electronic device provided in the embodiments of this specification. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0059] In related technologies, with the widespread application of high-voltage direct current (HVDC) transmission and urban rail transit systems, the intrusion of direct current into the AC power grid can cause DC bias in transformers. DC bias can lead to increased transformer vibration and noise, increased losses, and may affect its safe operation. Therefore, effectively assessing the degree of DC bias in transformers is a necessary step in implementing suppression measures and maintaining stable equipment operation.

[0060] In related technologies, one possible solution is to monitor the current at the transformer's grounding neutral point and assess the degree of magnetic bias by analyzing the DC component or low-frequency harmonics in this current signal. However, this method relies on a single characteristic quantity and is easily affected by interference factors such as load changes and power grid harmonics in actual operation, leading to unstable assessment results. If the current signal is distorted or the sensor fails, the assessment cannot be performed.

[0061] To improve the robustness of the assessment, one possible technical solution is to simultaneously acquire multiple features of current, vibration, and noise signals. During feature fusion, weights are first assigned to each feature, then linearly weighted to obtain a comprehensive evaluation value, which is then used to determine the magnetic bias level. However, this method assumes that the features are independent and their contributions can be directly added. However, the current, vibration, and noise characteristics generated by a transformer under magnetic bias are inherently correlated, potentially exhibiting complex interactions such as complementarity or redundancy. The linear weighting model cannot characterize these non-independent interactions, resulting in insufficient utilization of the overall information of the feature set and thus affecting the accuracy of the assessment results.

[0062] To address the aforementioned problems, this invention proposes a novel evaluation method. Its core concept lies in: quantifying the interaction relationships between features using fuzzy metrics, and employing a nonlinear integral method based on these metrics for feature fusion, thereby more scientifically characterizing the DC bias state of the transformer under the combined effect of multiple features and improving the reliability of the evaluation.

[0063] This embodiment provides a method for evaluating the DC bias of a transformer based on multi-feature fusion. The method can be implemented by an embedded monitoring device, which can be configured with multiple analog / digital signal interfaces for direct connection to current sensors, vibration acceleration sensors, and noise sensors, and has a built-in microprocessor unit or digital signal processor. Alternatively, the method can be implemented by an edge computing gateway, which can be an edge computing device deployed in a substation or distribution room. It can aggregate data from one or more transformer sensors via wired or wireless means. Finally, the method can be implemented by a server or server cluster located in a control center.

[0064] like Figure 1 and Figure 2 As shown, the method may include:

[0065] Step S10: Acquire multi-source signals during transformer operation.

[0066] In this embodiment, multi-source signals can be represented as various physical quantity signals that can reflect the DC bias magnetic state of the transformer from different dimensions. Specifically, these multi-source signals may include: grounding neutral point current signal, tank vibration signal, and environmental noise signal. More specifically, the grounding neutral point current signal can be acquired by a Hall current sensor nested within the transformer's grounding neutral point. The tank vibration signal can be acquired by an accelerometer fixed to the surface of the transformer tank. It should be noted that the sensor placement should avoid the tank's reinforcing ribs, and the area with the most direct vibration transmission (e.g., the middle of the tank sidewall) can be selected to reduce structural vibration interference and ensure that the acquired vibration signal accurately reflects the transformer core saturation state. Environmental noise signals can be acquired by noise sensors placed around the transformer.

[0067] In this embodiment, a BeiDou time synchronization module can be used to timestamp the signals collected by each sensor, with timestamp accuracy reaching the millisecond level. For the collected grounding neutral point current signal, enclosure vibration signal, and environmental noise signal, alignment processing can be performed based on the timestamps to eliminate time differences caused by acquisition delays from different sensors, ensuring the correspondence of multi-dimensional signals at the same time node and avoiding the impact of time asynchrony on the accuracy of subsequent feature extraction. Furthermore, preliminary format conversion can be performed on the collected raw signals, converting the analog signals output by the sensors into digital signals to facilitate subsequent data processing. Simultaneously, obvious outliers in the signals (e.g., sudden changes caused by momentary sensor malfunctions) are removed to ensure the validity of the raw data.

[0068] Step S12: Extract multiple feature quantities from the multi-source signal to characterize the degree of DC bias.

[0069] In this embodiment, the characteristic quantity can be represented as a parameter extracted from multi-source signals that can quantify the degree of DC bias of the transformer.

[0070] In one possible and specific implementation, the synchronized grounding neutral point current signal, enclosure vibration signal, and environmental noise signal can be preprocessed separately. Preprocessing operations can include filtering and signal smoothing. Filtering can employ a finite impulse response filter to remove power frequency interference and high-frequency noise from the signal. Signal smoothing can use a moving average method to reduce random fluctuations in the signal, retaining the trend components related to DC bias, and ensuring the stability of subsequent feature extraction.

[0071] In one possible and specific implementation, feature extraction for the grounded neutral point current signal can be performed as follows: The average value of the preprocessed current signal can be calculated, directly reflecting the average magnitude of the DC bias current intruding into the transformer; the more severe the bias, the larger the average current value. The power spectral density of the current signal can be calculated using a segmented periodogram averaging method. Specifically, the long current signal can be segmented into preset lengths, with overlapping portions between adjacent segments. A Hanning window is applied to each segment to reduce spectral leakage, and the periodograms of each segment are then averaged to obtain the final power spectral density. Based on this power spectral density, the ratio of the integral value in the low-frequency range to the integral value across the entire frequency range is calculated, i.e., the low-frequency current energy ratio. This ratio highlights the low-frequency characteristics of the DC bias current and is unaffected by power frequency and high-frequency interference.

[0072] In one possible and specific implementation, feature extraction for the transformer enclosure vibration signal can be performed as follows: The kurtosis of the preprocessed vibration signal can be calculated. Kurtosis is a statistical measure characterizing the steepness of a signal waveform. When a transformer experiences bias saturation, the nonlinearity of the core vibration increases, and the kurtosis value significantly increases, effectively distinguishing between normal and biased states. The power spectral density of the vibration signal can be calculated using the same piecewise periodogram averaging method as for current signals. The dominant vibration frequency of a normal transformer is concentrated at 100Hz. The ratio of the integral value of the non-100Hz frequency band to the integral value of the 100Hz dominant frequency band can be calculated, i.e., the vibration harmonic energy ratio. The more severe the bias, the higher the harmonic energy ratio.

[0073] In one possible and specific implementation, feature extraction for environmental noise signals can be performed as follows: The kurtosis of the preprocessed noise signal can be calculated, its physical meaning being consistent with the kurtosis of the vibration signal. The increased vibration of the iron core caused by bias magnetization will drive the vibration of the surrounding air, increasing the kurtosis value of the noise signal. The power spectral density of the noise signal can be calculated using the piecewise periodogram averaging method, again using 100Hz as the dominant frequency. The integral ratio of the non-100Hz frequency band to the 100Hz frequency band, i.e., the noise harmonic energy ratio, is calculated. This feature complements the vibration harmonic energy ratio, improving the comprehensiveness of the features.

[0074] Step S14: Based on the multiple feature quantities, the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method are fused to obtain the comprehensive weights of each feature quantity as the fuzzy density.

[0075] In this embodiment, the subjective weight can be represented as a weight value assigned based on domain experience and professional knowledge, determining the relative importance of each feature quantity to the DC bias assessment. The objective weight can be represented as a weight value obtained through quantitative calculation based on the actual data variation of each feature quantity, reflecting the distinguishing ability of the feature quantity itself. The fuzzy density is a parameter used to characterize the importance of a single feature quantity in the multi-feature fusion assessment.

[0076] In one possible and specific implementation, the subjective weights can be determined using the analytic hierarchy process (AHP), which may specifically include:

[0077] A judgment matrix can be constructed based on the relative importance of each characteristic quantity to the degree of DC bias. The scale of the judgment matrix can be determined based on the experience of those skilled in the art, and the scale values ​​include: 1, 2, 3, ..., 9; where 1 indicates that the two characteristics are equally important, 3 indicates that one characteristic quantity is slightly more important than the other, 5 indicates that it is relatively important, 7 indicates that it is very important, 9 indicates that it is absolutely important, and 2, 4, 6, and 8 are intermediate states between the above adjacent scales. By calculating the largest eigenvalue of the judgment matrix and its corresponding normalized eigenvector, the subjective weight vector of each characteristic quantity is obtained. To ensure the logical consistency of the judgment, a consistency ratio can be calculated for verification; if it fails, the judgment matrix can be adjusted.

[0078] In one possible and specific implementation, the objective weights can be determined using the coefficient of variation method. Specifically, this can include calculating the coefficient of variation (i.e., the ratio of standard deviation to mean) for each feature based on sample data collected over a certain period. A larger coefficient of variation indicates a more significant difference in the feature among different samples, and potentially a stronger ability to distinguish between different bias states. Normalizing the coefficients of variation of all features yields the objective weight vector for each feature.

[0079] In one possible and specific implementation, for each feature, its subjective weight and objective weight can be linearly weighted and fused to obtain the comprehensive weight of that feature. During the weighted fusion, a preset preference factor can be used to adjust the ratio of subjective to objective weights; for example, a preference factor of 0.5 indicates that both are treated equally. Finally, the comprehensive weight of each feature is directly defined as the fuzzy density of that feature.

[0080] Step S16: Based on the fuzzy density, determine the interaction parameter used to characterize the interaction relationship between features, and calculate the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameter; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them.

[0081] In this embodiment, the interaction parameter can be a parameter characterizing the degree of correlation and mutual influence among multiple feature quantities. The feature subset can be a set composed of one or more feature quantities selected from all feature quantities, which can reflect the importance of different feature combinations. The fuzzy measure can be a quantitative index characterizing the importance of the feature subset in the fusion evaluation and the interaction relationship between feature quantities within the subset. It differs from traditional weighted summation and can reflect the synergistic or inhibitory effects between features.

[0082] In one possible and specific implementation, the interaction parameter can be a numerical parameter, the specific value of which can be obtained by solving a polynomial equation with the fuzzy density of each feature quantity as the variable. The core logic of this polynomial equation can be: the interaction parameter plus one equals the product of all the corresponding features, which is the product of "one and the interaction parameter multiplied by the corresponding feature quantity's fuzzy density". Specifically, this equation can be solved by a numerical iterative method. The obtained interaction parameter has a unique solution and satisfies the constraint that 0 < interaction parameter < 1. This constraint ensures that there are reasonable interaction relationships between the features, neither being independent nor overly coupled.

[0083] In one possible and specific implementation, for any non-empty subset of multiple features, the calculation logic for the fuzzy measure can be as follows: First, calculate the product of "one and the interaction parameter multiplied by the fuzzy density of the feature" for each feature in the subset. Then, subtract 1 from this product, divide the difference by the interaction parameter, and the quotient is the fuzzy measure value of the feature subset. It should be noted that if the calculated interaction parameter is equal to 0, it indicates that the features are independent and there is no interaction relationship. In this case, the calculation of the fuzzy measure can be simplified to: the fuzzy measure value of the feature subset is equal to the sum of the fuzzy densities of all features within the subset.

[0084] In this embodiment, when calculating the fuzzy measure, all possible non-empty feature subsets can be covered, including: subsets composed of a single feature, subsets composed of two features, ..., and the entire set composed of all features, to ensure that subsequent fusion calculations can fully consider the impact of different feature combinations.

[0085] Step S18: Establish the membership function of each feature quantity for multiple preset DC bias levels to form a fuzzy membership evaluation matrix.

[0086] In this embodiment, the membership function characterizes the mapping relationship between the deviation of a certain feature quantity and a certain preset DC bias level. Its output value (membership degree) ranges from 0 to 1, reflecting the degree of membership of the feature quantity to that level. The preset DC bias level can be represented as different levels of DC bias state of the transformer, based on industry standards and actual operating experience, used to clarify the quantitative standard of the evaluation results. The fuzzy membership evaluation matrix can be a two-dimensional matrix with feature quantities as rows, bias levels as columns, and matrix elements representing the membership values ​​of the corresponding feature quantity to the corresponding level.

[0087] In this embodiment, the DC bias degree of the preset transformer can be divided into multiple levels, for example, four levels: normal state, slight bias, moderate bias and heavy bias.

[0088] In this embodiment, to eliminate differences in transformer models or operating benchmarks, the actual measured values ​​of each characteristic quantity can be standardized. Specifically, the deviation in the current operating state can be calculated based on the actual measured value of the characteristic quantity and a preset benchmark value (e.g., a typical value of the transformer under normal operating conditions). The deviation reflects the magnitude of change of the characteristic quantity relative to its normal benchmark. Specifically, for each characteristic quantity, its deviation in the current operating state can be calculated. The calculation logic for the deviation is as follows: using the value of the characteristic quantity under normal transformer operating conditions as the benchmark value, and using the maximum allowable limit value of the characteristic quantity under severe magnetization as the limit value, the characteristic quantity value is standardized to a value between 0 and 1 by comparing the currently measured characteristic quantity value with the benchmark value and the limit value, i.e., the deviation. A deviation of 0 indicates that the characteristic quantity is in the normal benchmark state, and a deviation of 1 indicates that the characteristic quantity has reached the limit state of severe magnetization. This calculation method can eliminate the differences in characteristic quantity benchmarks between transformers of different voltage levels, ensuring the fairness of the evaluation.

[0089] In this embodiment, for each preset level of magnetic bias, a trapezoidal membership function with deviation as the independent variable can be defined for each feature. The shape of this trapezoidal function can be determined by four preset threshold parameters for that magnetic bias level, which respectively define the interval between function values ​​of 0 and 1, as well as the transition interval. The deviation of each feature can be input into the membership function of its corresponding level to calculate the degree value of the feature belonging to the four levels of normal, mild, moderate, and severe, i.e., the membership degree. The membership degrees of all features to all levels of magnetic bias are organized into a matrix. The rows of this matrix correspond to different features, the columns correspond to different levels of magnetic bias, and each element in the matrix is ​​the membership degree of a specific feature to a specific level of magnetic bias.

[0090] Step S110: For each level of biased magnetization, perform a fusion calculation based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix to obtain the comprehensive evaluation value of each level.

[0091] In this embodiment, the membership vector can be represented as a column vector in the fuzzy membership evaluation matrix corresponding to a certain degree of magnetic bias, which includes the membership values ​​of all feature quantities for that degree. The fusion calculation can combine the membership vector and the fuzzy measure, and through a preset integral operation, fuse the membership information of multiple feature quantities into a single comprehensive index.

[0092] In one possible and specific implementation, for each preset level of magnetic bias (e.g., moderate magnetic bias), the following operations are performed: First, extract the column vector corresponding to the level from the fuzzy membership evaluation matrix, i.e., the vector composed of the membership degrees of all feature quantities to the level. Then, sort the elements in this membership vector in descending order of value. Finally, based on the sorted membership vector and the fuzzy measures of each feature subset calculated in step S16, perform a nonlinear fusion operation based on Choquet integral. The specific process of this operation can be: successively subtract the adjacent membership values ​​after sorting, multiply each difference by its corresponding fuzzy measure value of the feature subset composed of the current and subsequent feature quantities, and finally add all the products together. The sum obtained is the comprehensive evaluation value of the level of magnetic bias.

[0093] Step S112: Determine the final DC bias degree of the transformer based on the comprehensive evaluation values ​​of each level.

[0094] In this embodiment, after obtaining the comprehensive evaluation values ​​for each of the preset bias levels (e.g., normal, mild, moderate, severe), these comprehensive evaluation values ​​are compared. Based on the principle of maximum membership, the bias level corresponding to the largest comprehensive evaluation value is determined as the final DC bias level evaluation result for the transformer. For example, if the comprehensive evaluation value for the moderate bias level is the highest, then the transformer is determined to be in a moderate DC bias state.

[0095] In some embodiments, the step of acquiring multi-source signals during transformer operation includes:

[0096] Step S102: Obtain the grounding neutral point current signal by using a current sensor placed at the grounding neutral point of the transformer.

[0097] In this embodiment, a current sensor capable of measuring the DC component can be used, such as a Hall current sensor or a zero-flux current sensor. This sensor can be directly mounted or connected in series with the transformer grounding neutral point lead to ensure it can accurately sense and output a voltage or current signal proportional to the neutral point current. The acquired grounding neutral point current signal can be a continuously varying electrical signal over time, which may include power frequency AC components, harmonic components, and the DC bias component of interest.

[0098] Step S104: Obtain the vibration signal of the transformer enclosure by means of vibration sensors arranged on the surface of the transformer enclosure.

[0099] In this embodiment, a piezoelectric accelerometer can be selected. This type of sensor has high sensitivity and can accurately detect weak vibrations.

[0100] In this embodiment, the vibration sensor can be fixed at a preset position on the surface of the transformer tank. The middle area of ​​the tank sidewall can be selected, as vibration is directly transmitted at this location, accurately reflecting the core vibration state. During installation, structures such as tank reinforcing ribs, welds, and valves should be avoided to prevent interference from structural vibrations. Specifically, bolt fixing or strong magnetic fixation can be used to ensure a tight fit between the sensor and the tank surface without any loose gaps. After fixing, a vibration transmission test must be performed to verify the effectiveness of signal transmission.

[0101] Step S106: Acquire ambient noise signals by using noise sensors placed at preset locations around the transformer.

[0102] In this embodiment, a sound pressure sensor (microphone) can be used. This sensor can be installed at a predetermined location around the transformer, which can be within 1 to 3 meters of the transformer housing and facing the transformer body. The noise sensor can be installed on a separate bracket.

[0103] In one possible and specific implementation, the noise sensor can be positioned at a predetermined location around the transformer, at a straight-line distance of 2 meters from the transformer housing, with the sensor probe facing towards the transformer housing to ensure that the collected noise signal is primarily transformer-radiated noise. The sensor is positioned at the same height as the center of the transformer housing to avoid the influence of ground reflections or high-altitude airflow on the signal. Furthermore, there should be no obstructions (e.g., fences, equipment cabinets) between the sensor and the transformer to reduce attenuation along the sound wave propagation path. During installation, a bracket is used for fixation, and the bottom of the bracket is reinforced with shock-absorbing material to prevent environmental vibrations from being transmitted to the sensor and affecting measurement accuracy.

[0104] Step S108: Perform synchronous time synchronization processing on the grounding neutral point current signal, the enclosure vibration signal, and the environmental noise signal.

[0105] In this embodiment, the synchronization time processing can be represented as marking each sampling point of the three types of signals with a consistent timestamp using a unified time reference, eliminating the time difference caused by the acquisition delay of different sensors, and ensuring the correspondence of the three types of signals at the same time node.

[0106] In this embodiment, a BeiDou time synchronization module can be used as a unified time reference. This module has all-weather operation capability, can provide a stable standard time signal to the data processing unit, and can output UTC time information in real time. After receiving the signal, the execution subject can synchronously update its local clock to ensure that the clock error is controlled within the allowable range. While receiving the output signals from the current sensor, vibration sensor, and noise sensor, the execution subject can use the current standard time information as a timestamp and bind it to the signal data of the corresponding sampling point for storage. The timestamp format for each sampling point is "year-month-day-hour-minute-second-millisecond" to ensure the uniqueness and readability of the time information. For the three types of signals collected, alignment processing can be performed based on the bound timestamps. Specifically, using the timestamp as an index, the sampled values ​​of the current signal, vibration signal, and noise signal at the same millisecond are filtered to form a set of synchronization data. If a certain type of signal has no sampled value at that millisecond, linear interpolation is used to supplement the missing data to ensure that each set of synchronization data includes complete information of the three types of signals. The synchronized data can be stored in a structured format for easy retrieval in subsequent steps.

[0107] In some embodiments, the multi-source signals include: a ground neutral point current signal, a housing vibration signal, and an ambient noise signal; wherein, the step of extracting multiple characteristic quantities from the multi-source signals to characterize the degree of DC bias includes:

[0108] Step S122: Preprocess the grounding neutral point current signal, the enclosure vibration signal, and the environmental noise signal respectively.

[0109] In this embodiment, before extracting quantitative features, a series of standardized preprocessing operations can be performed on the original acquired signal to eliminate or reduce non-target interference components, improve signal quality, and thus ensure that the subsequently extracted features can more stably and accurately reflect the essential state of DC bias.

[0110] In a possible and specific implementation, the grounding neutral point current signal, enclosure vibration signal, and environmental noise signal can all be preprocessed according to a unified process of "filtering → signal smoothing → outlier removal" to ensure consistent processing standards for the three types of signals and avoid characteristic quantity deviations caused by differences in processing methods. Specifically, a finite impulse response (FIR) filter can be used to filter the three types of signals separately. The filtered signals can be smoothed using a moving average method to reduce random fluctuations in the signal. The window length of the moving average can be set to 5 sampling points, that is, the smoothed value of each sampling point is the average of that point and the two sampling points before and after it. For the beginning and end of the signal, a one-sided moving average method can be used to supplement the calculation to avoid data loss and ensure signal integrity. Finally, the 3σ criterion can be used to detect and remove outliers from the smoothed signal. The mean and standard deviation of each signal are calculated, and sampling points exceeding the range of "mean ± 3 times standard deviation" are identified as outliers. Outliers can be replaced by the mean of two adjacent valid sampling points using linear interpolation to ensure signal continuity.

[0111] Step S124: Extract the time-domain statistical features and frequency-domain energy distribution features based on power spectral density from the preprocessed grounding neutral point current signal.

[0112] In this embodiment, the time-domain statistical characteristics can be statistical parameters extracted from the time-domain waveform of the current signal. The core is the average current value, which can directly quantify the average magnitude of the DC bias current and is a fundamental characteristic characterizing the degree of bias intrusion. The frequency-domain energy distribution characteristics based on the power spectral density can be obtained by calculating the power spectral density of the current signal and analyzing the energy ratio of different frequency ranges. The core is the low-frequency current energy ratio, which can highlight the low-frequency characteristics of the DC bias current and is not affected by power frequency and high-frequency interference.

[0113] In this embodiment, the frequency domain energy distribution features can be extracted in the following way:

[0114] First, the power spectral density of the preprocessed current signal is estimated using a segmented periodogram averaging method. This method divides the long-time-history signal into several segments (allowing partial overlap between segments), and after windowing each segment, the square of its discrete Fourier transform amplitude is calculated to obtain the periodogram of that segment. Finally, the periodograms of all segments are averaged to obtain a smoothed power spectral density estimate. Then, based on the estimated power spectral density, the ratio of its energy integral value in the extremely low-frequency sub-band (e.g., the 0Hz to 0.1Hz band) to the total energy integral value in the entire analysis band is calculated. This ratio can effectively quantify the abnormal extremely low-frequency energy concentration phenomenon that may be caused by the core operating point shift due to DC bias.

[0115] Step S126: Extract time-domain kurtosis features and harmonic energy ratio features based on power spectral density from the preprocessed box vibration signal and the preprocessed environmental noise signal, respectively; wherein, the extracted features together constitute multiple features used to characterize the degree of DC bias.

[0116] In this embodiment, the time-domain kurtosis feature can be represented as a statistical parameter characterizing the steepness of the waveform of the vibration signal and the noise signal. It can reflect the strength of the impact component in the signal. When the transformer is magnetically saturated, the core vibration will intensify, which will lead to a significant increase in the signal kurtosis value. It is a key feature to distinguish between the normal state and the magnetically saturated state.

[0117] In this embodiment, the harmonic energy ratio characteristic based on power spectral density can be expressed as the energy ratio between the non-dominant frequency range and the dominant frequency range by calculating the power spectral density of the vibration signal and the noise signal. The core is the harmonic energy ratio, which can characterize the degree of harmonic component enhancement caused by bias magnetic saturation.

[0118] In this embodiment, the time-domain kurtosis features of the preprocessed enclosure vibration signal and environmental noise signal can be extracted separately as follows: First, a stable data segment of several consecutive seconds is selected to ensure the temporal correlation of the feature quantities. Then, the average value of the signal within the data segment is calculated, followed by the variance of the signal (reflecting the dispersion of the signal). Finally, based on the average value, variance, and expected value of the signal, the kurtosis value is calculated. After the calculation is completed, kurtosis outliers caused by sudden impacts are removed (using the 3σ criterion) to ensure that the kurtosis features can truly reflect the signal characteristics under biased magnetic conditions.

[0119] In one possible and specific implementation, the power spectral density can be calculated as follows:

[0120] The preprocessed signals (current signal, enclosure vibration signal, and environmental noise signal) are segmented according to a preset length. The segment length (number of points per segment) can be set to 1024 points, with a 50% overlap between adjacent segments. This means the next segment starts from the 512th sampling point of the first segment, ensuring signal continuity and computational accuracy. Then, a Hanning window is applied to each segment to reduce spectral leakage. The Hanning window smooths the signal at both ends within a segment, avoiding high-frequency interference caused by signal truncation and ensuring the accuracy of spectral analysis. Next, a Fast Fourier Transform (FFT) is performed on each windowed segment to convert the time-domain signal to the frequency-domain signal. The square of the frequency-domain amplitude of each segment is then calculated to obtain the periodogram of each segment. Next, the periodograms of all segments are arithmetically averaged to obtain the final power spectral density of the current signal, which reflects the energy distribution of the current signal at different frequencies. Finally, a window normalization factor is introduced to correct the power spectral density to compensate for energy loss caused by the window function.

[0121] In one possible and specific implementation, the dominant vibration and noise frequencies of a transformer under normal conditions are often concentrated around 100Hz (determined by the fundamental frequency of the iron core's magnetostriction). Therefore, 95Hz-105Hz can be defined as the dominant frequency range to ensure complete coverage of the dominant frequency's energy distribution. The non-dominant frequency range can be any frequency range within the 0-1kHz range other than the dominant frequency range (95Hz-105Hz), which includes the harmonic component energy generated during bias saturation. The power spectral density integral values ​​of the dominant frequency range and the non-dominant frequency range can be calculated separately. Then, the energy integral value of the non-dominant frequency range can be divided by the energy integral value of the dominant frequency range, and the resulting ratio is the harmonic energy ratio.

[0122] In some implementations, the step of fusing the subjective weights determined by the analytic hierarchy process (AHP) and the objective weights determined by the coefficient of variation method based on the plurality of feature quantities to obtain the comprehensive weights of each feature quantity as the fuzzy density includes:

[0123] Step S142: Construct a judgment matrix based on the relative importance of each characteristic quantity to the degree of DC bias, and obtain the subjective weight by solving the maximum eigenvalue of the judgment matrix and its corresponding eigenvector and normalizing it.

[0124] In this embodiment, the judgment matrix can be a square matrix constructed based on the professional experience of industry experts, quantifying the relative importance of each characteristic quantity to the evaluation of DC bias. Its order is consistent with the number of characteristic quantities, and its matrix elements reflect the importance comparison relationship between pairs of characteristic quantities.

[0125] In one possible and specific implementation, the judgment matrix can be constructed in the following way:

[0126] First, the set of features involved in the weighting calculation is determined, namely, the average current value and low-frequency current energy ratio of the grounding neutral point current signal; the vibration kurtosis and vibration harmonic energy ratio of the enclosure vibration signal; and the noise kurtosis and noise harmonic energy ratio of the environmental noise signal, totaling six features. The order of the judgment matrix is ​​6. Then, the 1-9 scaling method is used to quantify the relative importance of each pair of features. Finally, using the six features as rows and columns, the importance of each pair of features is compared sequentially, filling the matrix elements. If the importance scale of feature A to feature B is 'a', then the scale of feature B to feature A is 1 / a, ensuring that the judgment matrix is ​​a positive reciprocal matrix, which meets the mathematical requirements of the analytic hierarchy process (AHP).

[0127] In this implementation, linear algebraic operations can be used to solve for the eigenvalues ​​and corresponding eigenvectors of the matrix. An eigenvalue is a value that satisfies the condition: matrix × eigenvector = eigenvalue × eigenvector; an eigenvector is a non-zero vector corresponding to each eigenvalue. Then, the eigenvalue with the largest value is selected from all eigenvalues, and the eigenvector corresponding to this largest eigenvalue is extracted. Finally, the eigenvector corresponding to the extracted largest eigenvalue can be normalized to obtain the subjective weights. The normalization logic can be: dividing the value of each element in the eigenvector by the sum of all elements in the eigenvector, ensuring that after normalization, the values ​​of all elements are between 0 and 1, and the sum is 1.

[0128] Step S144: Calculate the coefficient of variation of each feature quantity, and normalize the coefficient of variation of all feature quantities to obtain the objective weight.

[0129] In this embodiment, multiple sets of sample data can be collected in advance during the normal operation and possible biased magnetic state of the transformer over a period of time. For each characteristic quantity, the ratio of the absolute value of its statistical standard deviation to the absolute value of its arithmetic mean across all sample data can be calculated. This ratio is the coefficient of variation for that characteristic quantity. The coefficient of variation can be a dimensionless statistic, which eliminates the influence of the dimension and absolute value of the characteristic quantity itself, and purely reflects the magnitude of the relative fluctuation of the characteristic quantity among samples.

[0130] In this embodiment, after calculating the coefficients of variation for all feature quantities, all coefficients of variation can be normalized. Specifically, the coefficient of variation for each feature quantity can be divided by the sum of the coefficients of variation for all feature quantities. After this processing, each feature quantity can obtain a value between 0 and 1, and the sum of these values ​​for all feature quantities is 1. This value is the objective weight of the feature quantity. It can be understood that a feature quantity with a larger coefficient of variation indicates that its data difference is more obvious under different operating states (e.g., normal and biased magnetization), and therefore it is assigned a higher objective weight in order to play a greater distinguishing role in the evaluation.

[0131] Step S146: By using a weighted average method, the subjective weight and the objective weight are integrated to obtain the comprehensive weight of each feature quantity, and the comprehensive weight is determined as the fuzzy density of each feature quantity; wherein, the coefficient of the weighted average is a preset preference factor, which is used to adjust the degree of preference for the subjective weight or the objective weight.

[0132] In this embodiment, the purpose of this step is to organically combine subjective weights based on expert experience with objective weights based on data statistical characteristics to form a comprehensive weight that takes into account both prior knowledge and data-driven information. This comprehensive weight is then directly defined as the fuzzy density in the subsequent fuzzy measurement theory to characterize the fundamental importance of a single feature quantity.

[0133] In this embodiment, for each feature, its subjective weight and objective weight can be linearly weighted and averaged to calculate its comprehensive weight. The linear weighted average can be calculated as follows: Comprehensive Weight = Preference Factor × Subjective Weight + (1 - Preference Factor) × Objective Weight. The preference factor can be a pre-set constant between 0 and 1. Specifically, the preference factor is used to adjust the proportion of subjective experience and objective data in the final comprehensive weight. If the preference factor is set to 0.5, it indicates that subjective and objective factors are equally important; if the preference factor is greater than 0.5, it indicates a greater tendency to rely on expert experience; if the preference factor is less than 0.5, it indicates a greater tendency to rely on the statistical characteristics of the data itself.

[0134] In this embodiment, the comprehensive weights of each feature quantity calculated above can be directly determined as the fuzzy density of the corresponding feature quantity. The value range of the fuzzy density can be consistent with the comprehensive weight, both being between 0 and 1. The larger the fuzzy density of a single feature quantity, the higher its fundamental importance in the subsequent fusion evaluation.

[0135] In some implementations, the step of determining an interaction parameter to characterize the interaction relationship between features based on the fuzzy density, and calculating a fuzzy measure of any non-empty subset of multiple feature quantities based on the interaction parameter, includes:

[0136] Step S162: Solve for the interaction parameters based on the fuzzy density of each feature quantity; wherein the value of the interaction parameters is obtained by solving a polynomial equation with each fuzzy density as a variable; wherein the polynomial equation is configured to be the interaction parameter plus one, which is equal to the product of the corresponding features; wherein the product is a product of the interaction parameter multiplied by its fuzzy density.

[0137] In this embodiment, the polynomial equation can be expressed as a mathematical equation with the fuzzy density of each feature quantity as the variable and the interaction parameter as the solution objective.

[0138] In a possible and specific implementation scheme, a polynomial equation can be constructed according to a preset logic. The core relationship of the equation is: the value of the interaction parameter plus one is equal to the product of the corresponding features. The specific construction method of the product is as follows: for each feature, calculate the product of 1 and the interaction parameter multiplied by the fuzzy density of that feature, and then multiply the calculated results for all features sequentially to obtain the final product. For example, for 6 features, the product is (1+λ×m1)×(1+λ×m2)×(1+λ×m3)×(1+λ×m4)×(1+λ×m5)×(1+λ×m6); where m1 to m6 are the fuzzy densities of the 6 features, λ is the interaction parameter, and the overall form of the equation can be:

[0139] 1+λ=(1+λ×m1)×(1+λ×m2)×…×(1+λ×m6).

[0140] In one possible and specific implementation, the above polynomial equations can be solved using a numerical iteration method, specifically:

[0141] First, the range of interaction parameters can be set between -1 and positive infinity, the initial value of the iteration can be set to 0.5, and the iteration precision (convergence threshold) can be set to 10⁻⁶ to ensure the accuracy of the solution results.

[0142] Then, substitute the initial values ​​into the polynomial equation and calculate the values ​​on the left side (1+λ) and the right side (the product of all characteristic quantities) respectively, and calculate the absolute value of the difference between the two. If the absolute value of the difference is greater than the convergence threshold, adjust the value of the interaction parameter according to the sign of the difference (e.g., decrease λ if the left side is greater than the right side, and increase λ otherwise), and repeat the substitution calculation. If the absolute value of the difference is less than or equal to the convergence threshold, stop the iteration, and the current value of λ is the solution to the equation. The interaction parameter needs to satisfy the constraint that there is a unique solution and 0 < λ < 1.

[0143] After obtaining the results through iteration, the uniqueness of the solution is first verified (by selecting multiple different initial points near the initial value and repeating the iteration to confirm that the final solution is consistent), and then the value is verified to be within the range of 0-1. If the constraints are not met, the validity of the fuzzy density data is checked (whether there are outliers, whether the sum is 1), and the fuzzy density calculation in step S146 is re-executed before the interactive parameter solution is performed.

[0144] After iterative solving and constraint verification, a unique interaction parameter value satisfying 0 < λ < 1 is obtained. This value can be used for fuzzy measure calculation of all subsequent feature subsets, and its magnitude characterizes the strength of the interaction relationship between features. That is, the closer the λ value is to 1, the stronger the cooperative or inhibitory effect between features. The closer the λ value is to 0, the weaker the interaction between features, and the closer it is to an independent state.

[0145] Step S164: For any non-empty subset of the plurality of features, the value of its fuzzy measure is calculated as follows: First, calculate the product of the interaction parameter and the fuzzy density corresponding to each feature in the subset. Then, divide the result of the product minus one by the interaction parameter. The quotient is the fuzzy measure value of the feature subset.

[0146] In one possible and specific implementation, the feature subset can be a set composed of one or more features arbitrarily selected from all six features. Specifically, a full enumeration of the six features can be performed to obtain a single feature subset (6 in total), two feature subsets (15 in total), three to five feature subsets (20, 15, and 6 respectively), or a complete set of six features (1).

[0147] In one possible and specific implementation, the fuzzy measure value can be calculated as follows: First, extract the fuzzy density values ​​corresponding to all features within the current subset from the fuzzy density vector. Next, calculate the product of 1 and the interaction parameter multiplied by its fuzzy density for each feature in the subset, and then multiply these products sequentially to obtain a chain product. Then, subtract 1 from the above chain product, divide the resulting difference by the interaction parameter λ, and the final quotient is the fuzzy measure value of the feature subset.

[0148] In some implementations, the step of establishing membership functions for each feature quantity with respect to multiple preset DC bias levels to form a fuzzy membership evaluation matrix includes:

[0149] Step S182: For each feature quantity, calculate the deviation in the current operating state based on its actual measured value and the preset benchmark value.

[0150] In this embodiment, the reference value can be the standard value of the characteristic quantity under normal transformer operation, and is a reference benchmark for judging whether the characteristic quantity deviates from the normal state.

[0151] The benchmark value for the average current can be the average value of the grounding neutral point current during normal transformer operation. The benchmark value for the low-frequency current energy ratio can be the energy ratio of the interval under normal conditions. The benchmark values ​​for vibration kurtosis and noise kurtosis can be 3 (the kurtosis value of a normally distributed signal, corresponding to the signal characteristics during normal operation). The benchmark values ​​for vibration harmonic energy ratio and noise harmonic energy ratio can be the energy ratio of non-dominant frequency to dominant frequency under normal conditions.

[0152] In this embodiment, the limit value can be the maximum allowable value of the characteristic quantity under a heavily biased magnetic state, which is the value of the characteristic quantity when the degree of biased magnetic state reaches the danger threshold.

[0153] In this embodiment, the deviation can be calculated using the following formula: Deviation = (Actual measured value of characteristic quantity - Reference value) / (Limit value - Reference value).

[0154] Step S184: For each preset DC bias level, define a trapezoidal membership function with the deviation as the independent variable for each feature quantity, wherein the shape of the trapezoidal membership function is determined by a threshold parameter preset for the DC bias level; wherein the multiple preset DC bias levels include four levels, namely normal state, slight bias, moderate bias and heavy bias.

[0155] In this embodiment, the trapezoidal membership function can be represented as a piecewise function that outputs membership degrees with deviation as the independent variable. Its function curve can be trapezoidal, specifically defined by four threshold parameters (a, b, c, d), which enables a smooth transition between different bias levels. Specifically, b and c define the core interval of the bias level, while a and d define the transition interval with adjacent levels.

[0156] In this embodiment, the normal state can be a state with no bias or negligible bias, the mild bias can be a state with a slight bias that does not affect the normal operation of the transformer, the moderate bias can be a state with obvious bias, and the severe bias can be a state with severe bias that may endanger the safety of the equipment.

[0157] In one possible and specific implementation scheme, all threshold parameters can be determined based on industry standards, actual operating experience, and the deviation range (0-1) of characteristic quantities. The threshold parameters for the four levels are set uniformly to ensure consistent evaluation standards.

[0158] In one possible and specific implementation, for each characteristic quantity, a corresponding trapezoidal membership function can be defined for each of the four bias levels. The function form for all characteristic quantities is uniform, with the membership output differing only due to variations in the degree of deviation. The segmented definition logic of the function can be:

[0159] When the deviation x ≤ a, the membership degree = 0; when a < x ≤ b, the membership degree = (xa) / (ba) (linearly increasing).

[0160] When b < x ≤ c, the membership degree = 1 (core interval); when c < x ≤ d, the membership degree = (dx) / (dc) (linearly decreasing); when x > d, the membership degree = 0.

[0161] Step S186: Based on the trapezoidal membership function, calculate the membership degree of each characteristic quantity's deviation degree to each DC bias level.

[0162] In this embodiment, the specific deviation value of each characteristic quantity calculated in step S182 can be input into the trapezoidal membership function defined for the four bias levels of the characteristic quantity, and the four corresponding membership values ​​can be obtained by function calculation.

[0163] Step S188: Construct a fuzzy membership evaluation matrix based on all calculated membership degrees; wherein, the rows of the fuzzy membership evaluation matrix correspond to the feature quantities, the columns correspond to the degree of magnetic bias, and the matrix elements are the membership degrees of the corresponding feature quantities at the corresponding degree.

[0164] In this embodiment, the membership evaluation results of all feature quantities can be systematically organized into a matrix data structure. This matrix completely represents the membership relationship of the transformer's current state to each bias level when observed from all feature dimensions. Specifically, assuming there are m feature quantities and n bias levels are preset (n=4 in this embodiment), an m-row, n-column matrix R can be constructed. The i-th row of matrix R corresponds to the i-th feature quantity, and the j-th column corresponds to the j-th bias level (for example, the first column corresponds to the normal state, the second column corresponds to slight bias, and so on). The membership vector of each feature quantity calculated in step S186 can be sequentially filled into the corresponding row of matrix R. That is, the element rij in the i-th row and j-th column of matrix R has the value of the membership degree of the i-th feature quantity to the j-th bias level.

[0165] In some implementations, the step of performing a fusion calculation for each bias level, based on the membership vector corresponding to that level in the fuzzy membership evaluation matrix and the fuzzy measure, to obtain a comprehensive evaluation value for each level, includes:

[0166] Step S1102: For each level of magnetic bias, extract the membership vector corresponding to the level from the fuzzy membership evaluation matrix, and sort the membership vectors in descending order of membership value.

[0167] In this embodiment, the membership vector is a vector in the fuzzy membership evaluation matrix that corresponds to a certain degree of bias. The dimension is the same as the number of features (6 dimensions in this embodiment). Each element in the vector corresponds to the membership value of a feature for that degree.

[0168] Step S1104: Based on the sorted membership vector and the fuzzy measure, perform Choquet integration to obtain the comprehensive evaluation value of the degree of magnetic bias.

[0169] In this embodiment, the Choquet integral operation can be represented as a multi-feature fusion algorithm adapted to fuzzy measures. It can integrate the gradient change of the membership degree after sorting and the interaction relationship between features through the cumulative operation of the membership degree difference × the corresponding feature subset fuzzy measure, thereby realizing the deep fusion of multi-dimensional information.

[0170] Specifically, we can first calculate the difference between adjacent membership degrees. It's understandable that, for the sorted membership sequence, the elements in the sequence are arranged in descending order. We can calculate the difference between each pair of adjacent elements in the sequence, that is, the difference between the first and second elements, the difference between the second and third elements, and so on, until we calculate the difference between the last element and zero.

[0171] For each calculated difference, the executing entity can dynamically assign a corresponding weight. This weight may not be a fixed value but is determined by a fuzzy measure. Specifically, for the i-th difference (corresponding to all features starting from the i-th position after sorting), its weight can be the fuzzy measure value of the corresponding feature subset. This subset can include all features from the i-th position onwards in the sorted sequence. For example, the weight of the first difference (corresponding to the highest membership level) can be the fuzzy measure value of the set consisting of all features. The weight of the last difference (corresponding to the difference between the lowest membership level and zero) is the fuzzy measure value of the single-point set including only the last feature in the sorting. Then, the executing entity can multiply each difference by its assigned feature subset fuzzy measure value to obtain multiple weighted products. Finally, all weighted products are summed, and the resulting total is the comprehensive evaluation value for that degree of magnetic bias.

[0172] The step of determining the final DC bias degree of the transformer based on the comprehensive evaluation values ​​of each level includes:

[0173] Step S1122: Compare the comprehensive evaluation values ​​of all bias levels, and determine the bias level with the highest comprehensive evaluation value as the final DC bias level of the transformer.

[0174] In one specific implementation plan, a method for evaluating the DC bias of a transformer based on the fusion of multiple characteristic quantities is provided, as follows:

[0175] Step 1: Collect the transformer grounding neutral point current signal, the tank surface vibration signal, and the surrounding noise signal, and synchronize the multi-dimensional data in time.

[0176] S21. The transformer neutral point current signal is obtained from a Hall current sensor that is nested in the transformer grounding neutral point; the vibration signal is obtained from an acceleration sensor fixed to the surface of the transformer tank (avoiding the reinforcing ribs); and the noise signal is obtained from a noise sensor located 2m away from the transformer and facing the transformer tank.

[0177] S22. Use the BeiDou time synchronization module to mark data timestamps and synchronize multi-dimensional data.

[0178] Step 2: Preprocess the multidimensional data to extract characteristic quantities that characterize the DC bias of the transformer;

[0179] S31. For current signals, in the time domain, calculate the average value of the neutral point current signal; in the frequency domain, use Welch to segment the long signal, calculate the periodogram, and average the results to obtain the power spectral density of the current signal.

[0180] (1);

[0181] In formula (1), The value is rounded down to the nearest integer, representing the number of segments in the signal x(n); D is the number of overlaps between two adjacent segments, typically taken as... L represents the segment length and the number of points in each segment; the i-th segment of the signal... ,in and Window function .

[0182] Specifically, This is the power spectral density of the current signal at frequency k obtained by the Welch method, used to quantify the energy distribution intensity of the current signal at that frequency. This is the periodogram of the i-th segment of the current signal at frequency point k. It is an intermediate result for calculating the final power spectral density. Each segment of the signal corresponds to a periodogram. The sampling frequency of the current signal; The segment length of each signal segment; For window normalization factor; Let i be the sampling sequence of the i-th segment of the current signal; denoted as the window function applied to each segment of the signal.

[0183] In addition, a window normalization factor W is introduced to correct the power loss caused by the window function:

[0184] (2);

[0185] In the formula, m is the index of the sampling point within the segment.

[0186] Calculate the ratio of the integral in the 0~0.1Hz frequency range to the integral over the entire frequency range, i.e., the low-frequency current energy ratio; characterize the magnitude of the DC bias current intruding into the transformer using the above characteristic quantities.

[0187] S32. For vibration signals, calculate the kurtosis of vibration acceleration in the time domain:

[0188] (3);

[0189] In equation (3), This represents a single sample value in the vibration acceleration signal. The average value of the signal. Let E be the signal variance and E be the expected value of the data.

[0190] In the frequency domain, the Welch method is used to obtain the power spectral density of the vibration signal. Since the dominant vibration frequency of a normal transformer is concentrated at 100Hz, the ratio of the integral in the non-dominant frequency range to the integral in the dominant frequency range is calculated, which is the vibration harmonic energy ratio. The above characteristic quantities are used to characterize the vibration magnitude when the transformer experiences biased magnetic saturation.

[0191] S33. For noise signals, calculate the kurtosis of the noise in the time domain; in the frequency domain, use the Welch method to obtain the power spectral density of the noise signal, and calculate the ratio of the integral in the non-dominant frequency range (100Hz) to the integral in the dominant frequency range, i.e., the noise harmonic energy ratio; characterize the noise magnitude when the transformer experiences bias saturation through the above characteristic quantities.

[0192] Step 3: Calculate the subjective and objective weights of the characteristic quantities for the DC bias of the transformer using the analytic hierarchy process (AHP) and the coefficient of variation method, respectively. Then, determine the fuzzy density of each characteristic quantity using the weighted average method, and calculate the fuzzy density-based... Values ​​and fuzzy measures;

[0193] S41. Calculate the subjective weight of each feature based on the Analytic Hierarchy Process (AHP); Define... For characteristic quantity right The relative importance is shown in Table 1. (where n is the number of features) and construct the judgment matrix. ; Calculate the judgment matrix according to equation (4) eigenvalues :

[0194] (4);

[0195] In formula (4), This refers to the determinant operation of a matrix; This is a judgment matrix constructed based on the relative importance of feature quantities;

[0196] To determine the eigenvalues ​​of the matrix; I is the identity matrix.

[0197] Solving for eigenvalues Find the largest eigenvalue Substitute into formula (5) to solve. The corresponding solution space S:

[0198] (5);

[0199] Normalize S to obtain subjective weights ,in for:

[0200] (6);

[0201] Calculate the consistency ratio (CR):

[0202] (7);

[0203] In formula (7), To determine the consistency ratio of the matrix; The average random consistency index; The total number of features involved in the weight calculation; RI is obtained by looking up Table 2. If the consistency ratio CR < 0.1, the judgment matrix is ​​considered... Consistency is acceptable; otherwise, the judgment matrix needs to be adjusted. ;

[0204] S42. Calculate the objective weight of each feature quantity based on the coefficient of variation method; define... For characteristic quantity The coefficient of variation can be obtained from equation (8):

[0205] (8);

[0206] In formula (8), , These are the standard deviation and mean of the i-th characteristic quantity, respectively. The greater the variation of the characteristic quantity, the stronger its ability to distinguish the DC bias of the transformer, and it is assigned a higher objective weight according to formula (9). ,in for:

[0207] (9);

[0208] In formula (9), The objective weight vector for all features is an n-dimensional vector (where n is the total number of features). Each element in the vector corresponds to the objective weight of a feature, which is used to integrate the objective importance of all features. The objective weight of the i-th feature is the final weight value after normalization. Let be the coefficient of variation of the i-th feature.

[0209] S43, Subjective Weighting and objective weight Perform linear weighting to calculate the comprehensive weight, i.e., the fuzzy density. :

[0210] (10);

[0211] In equation (10), , These are the subjective and objective weights of the first feature, respectively, where n is the number of features. This is the preference factor, representing the degree of preference for subjective weights, and is generally taken as 0.5.

[0212] S44. Define fuzzy density It is equal to the overall weight. The parameters representing the fuzzy measure are solved by equation (11). :

[0213] (11);

[0214] In formula (11), The vector is composed of the combined weights of each feature quantity, and is an n-dimensional vector (where n is the total number of features). The elements in the vector... It is the fuzzy density of the nth feature. The weight vector of each feature obtained by fusing subjective and objective weights is also an n-dimensional vector, and the elements in the vector are... The comprehensive weight of the nth feature; These are parameters that characterize the interaction relationships between feature quantities (i.e., interaction parameters). For all n feature quantities corresponding to "AND" The product of the features and their fuzzy densities is multiplied together.

[0215] parameter There exists a unique solution and and For those belonging to Any nonempty subset F can be determined according to Calculate the fuzzy measure of each feature:

[0216] (12);

[0217] In formula (12), Let F be the fuzzy measure corresponding to any non-empty subset F of the feature set X. Let X be any non-empty combination of features selected from the set X consisting of all features.

[0218] like This indicates that the features are independent of each other, and the fuzzy measure can be expressed as a simple summation:

[0219] (13);

[0220] Step 4: Calculate the membership function of each characteristic quantity with respect to the DC bias of the transformer, and establish a fuzzy membership evaluation matrix;

[0221] S51. Determine the DC bias degree classification of the transformer: normal state, slight bias, moderate bias, and severe bias, respectively corresponding to... , , , ;

[0222] S52. Calculate the deviation of each characteristic quantity to ensure fairness in the evaluation of transformer DC bias across different voltage levels:

[0223] (14);

[0224] According to equation (14), the four membership functions corresponding to the DC bias of the 4-stage transformer for each characteristic quantity are calculated as follows: , , , :

[0225] (15);

[0226] In formula (15), parameters b and c determine the core interval of the membership function. The parameters b and c corresponding to the four bias levels correspond to 0% and 22.5%, 27.5% and 47.5%, 52.5% and 72.5%, and 77.5% and 100% of the maximum allowable limit in industry standards or actual operation, respectively. The parameters a and d are extended outward by 5% based on the parameters b and c, and are used for smooth transition between different bias levels. The calculation of the above parameters is converted into deviation degree.

[0227] S53. Establish the fuzzy membership evaluation matrix R as shown in equation (16):

[0228] (16);

[0229] Step 5: Calculate the fuzzy integral by combining the fuzzy measure and fuzzy membership evaluation matrix of each feature quantity to determine the DC bias of the transformer;

[0230] S61, Extract the j-th ( The membership vector of the DC bias level is the j-th column of the fuzzy membership evaluation matrix R, and different feature quantities are arranged according to the membership degree.

[0231] (17);

[0232] The Choquet integral is calculated according to equation (18), and the overall effect of all characteristic quantities on the j-th DC bias level is obtained. :

[0233] (18);

[0234] In formula (18), This represents the overall degree of integration of all characteristic quantities with respect to the j-th DC bias level (i.e., the overall evaluation value of that bias level). The i-th membership value after sorting in the membership vector corresponding to the j-th bias level; For the set of characteristic quantities The corresponding fuzzy measure value; The difference between two adjacent membership values ​​after sorting reflects the contribution of the characteristic quantity corresponding to the i-th position to the change in the membership degree of the bias level.

[0235] Let be the sorted membership degree, and , This represents the set of the i-th to n-th features sorted by membership degree, corresponding to the fuzzy measure values. .

[0236] S62. Calculate the Choquet integral for all bias levels in sequence to obtain the fuzzy integral vector. The comprehensive degree of all characteristic quantities belonging to each DC bias level of the transformer is obtained after fuzzy integration calculation. According to the principle of maximum membership, the value with the largest value is the current DC bias level of the transformer.

[0237] Table 1: Scale of the Matrix

[0238]

[0239] Table 2: Average Random Consistency Index

[0240]

[0241] According to an embodiment of the present invention, a transformer DC bias degree evaluation device based on multi-feature fusion is provided. Please refer to [link to relevant documentation]. Figure 3 ,include:

[0242] The acquisition module is used to acquire multi-source signals during transformer operation.

[0243] An extraction module is used to extract multiple feature quantities from multi-source signals to characterize the degree of DC bias.

[0244] The fuzzy density fusion module is used to fuse the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method based on the multiple feature quantities to obtain the comprehensive weight of each feature quantity as the fuzzy density.

[0245] The fuzzy measure determination module is used to determine the interaction parameters used to characterize the interaction relationship between features based on the fuzzy density, and to calculate the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameters; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them.

[0246] A module is established to establish the membership function of each feature quantity for multiple preset DC bias levels, so as to form a fuzzy membership evaluation matrix.

[0247] The comprehensive evaluation value calculation module is used to perform fusion calculation for each level of magnetic bias, based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix, to obtain the comprehensive evaluation value for each level.

[0248] The determination module is used to determine the final DC bias of the transformer based on the comprehensive evaluation values ​​of each level.

[0249] According to an embodiment of the present invention, an electronic device is provided; please refer to... Figure 4 The electronic device in this embodiment may include one or more of the following components: a processor, a network interface, memory, non-volatile memory, and one or more application programs, wherein the one or more application programs may be stored in non-volatile memory and configured to be executed by one or more processors, and the one or more programs are configured to perform the methods as described in the foregoing method embodiments.

[0250] According to embodiments of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a computer, causes the computer to perform the method described in any of the above embodiments.

[0251] According to embodiments of the present invention, a computer program product comprising instructions is also provided, which, when executed by a computer, cause the computer to perform a method in any of the above embodiments.

[0252] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the DC bias degree of a transformer based on the fusion of multiple characteristic quantities, characterized in that, include: Acquire multi-source signals during transformer operation; Extract multiple feature quantities from multi-source signals to characterize the degree of DC bias. Based on the aforementioned multiple feature quantities, the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method are integrated to obtain the comprehensive weights of each feature quantity as the fuzzy density. Based on the fuzzy density, an interaction parameter is determined to characterize the interaction relationship between features, and a fuzzy measure is calculated for any non-empty subset among multiple feature quantities based on the interaction parameter; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them. Establish membership functions for each characteristic quantity with respect to multiple preset DC bias levels to form a fuzzy membership evaluation matrix; For each level of biased magnetization, a fusion calculation is performed based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix to obtain the comprehensive evaluation value of each level. The final DC bias degree of the transformer is determined based on the comprehensive evaluation values ​​of each level.

2. The method according to claim 1, characterized in that, The step of acquiring multi-source signals during transformer operation includes: The grounding neutral point current signal is obtained by a current sensor placed at the grounding neutral point of the transformer; Vibration signals of the transformer enclosure are acquired by vibration sensors arranged on the surface of the enclosure. Ambient noise signals are acquired by noise sensors placed at predetermined locations around the transformer; The grounding neutral point current signal, enclosure vibration signal, and environmental noise signal are synchronized and timed.

3. The method according to claim 1, characterized in that, The multi-source signals include: grounding neutral point current signal, enclosure vibration signal, and environmental noise signal; wherein, the step of extracting multiple characteristic quantities from the multi-source signals to characterize the degree of DC bias includes: The grounding neutral point current signal, the enclosure vibration signal, and the environmental noise signal are preprocessed respectively; From the preprocessed grounding neutral point current signal, extract its time-domain statistical features and frequency-domain energy distribution features based on power spectral density; Temporal kurtosis features and harmonic energy ratio features based on power spectral density are extracted from the preprocessed enclosure vibration signal and the preprocessed environmental noise signal, respectively. The extracted features together constitute multiple features used to characterize the degree of DC bias.

4. The method according to claim 3, characterized in that, The step of fusing the subjective weights determined by the analytic hierarchy process (AHP) and the objective weights determined by the coefficient of variation method to obtain the comprehensive weights of each feature quantity as the fuzzy density, based on the multiple feature quantities, includes: A judgment matrix is ​​constructed based on the relative importance of each characteristic quantity to the degree of DC bias. The subjective weights are obtained by solving the largest eigenvalue and its corresponding eigenvector of the judgment matrix and then normalizing the results. Calculate the coefficient of variation of each feature quantity, and normalize the coefficient of variation of all feature quantities to obtain the objective weight; By using a weighted average method, the subjective weights and the objective weights are integrated to obtain the comprehensive weights of each feature quantity, and the comprehensive weights are determined as the fuzzy density of each feature quantity; wherein, the coefficient of the weighted average is a preset preference factor, which is used to adjust the degree of preference for subjective weights or objective weights.

5. The method according to claim 4, characterized in that, The step of determining the interaction parameter used to characterize the interaction relationship between features based on the fuzzy density, and calculating the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameter, includes: Based on the fuzzy density of each feature quantity, the interaction parameter is solved; wherein, the value of the interaction parameter is obtained by solving a polynomial equation with each fuzzy density as a variable; wherein, the polynomial equation is configured to be the interaction parameter plus one, which is equal to the product of the corresponding features; wherein, the product is a product of the interaction parameter multiplied by its fuzzy density. For any non-empty subset of the plurality of features, the value of its fuzzy measure is calculated as follows: First, calculate the product of the interaction parameter and the fuzzy density corresponding to each feature in the subset. Then, divide the result of the product minus one by the interaction parameter. The quotient is the fuzzy measure value of the feature subset.

6. The method according to claim 4, characterized in that, The step of establishing membership functions for each feature quantity with respect to multiple preset DC bias levels to form a fuzzy membership evaluation matrix includes: For each feature quantity, the deviation under the current operating state is calculated based on its actual measured value and the preset benchmark value; For each preset DC bias level, a trapezoidal membership function with the deviation degree as the independent variable is defined for each feature quantity, wherein the shape of the trapezoidal membership function is determined by a threshold parameter preset for the DC bias level; wherein the multiple preset DC bias levels include four levels, namely normal state, slight bias, moderate bias and heavy bias. Based on the trapezoidal membership function, the membership degree of each characteristic quantity's deviation to each DC bias level is calculated; Based on all the calculated membership degrees, a fuzzy membership degree evaluation matrix is ​​constructed; where the rows of the fuzzy membership degree evaluation matrix correspond to the feature quantities, the columns correspond to the degree of magnetic bias, and the matrix elements are the membership degrees of the corresponding feature quantities for that degree.

7. The method according to claim 6, characterized in that, The step of performing a fusion calculation for each level of magnetic bias, based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix, to obtain the comprehensive evaluation value for each level, includes: For each level of magnetic bias, extract the membership vector corresponding to the level from the fuzzy membership evaluation matrix, and sort the membership vectors in descending order of membership value. Based on the sorted membership vector and the fuzzy measure, perform Choquet integration to obtain a comprehensive evaluation value for the degree of magnetic bias. The step of determining the final DC bias degree of the transformer based on the comprehensive evaluation values ​​of each level includes: The comprehensive evaluation values ​​of all bias levels are compared, and the bias level with the highest comprehensive evaluation value is determined as the final DC bias level of the transformer.

8. A transformer DC bias degree evaluation device based on multi-feature quantity fusion, characterized in that, include: The acquisition module is used to acquire multi-source signals during transformer operation. An extraction module is used to extract multiple feature quantities from multi-source signals to characterize the degree of DC bias. The fuzzy density fusion module is used to fuse the subjective weights determined by the analytic hierarchy process and the objective weights determined by the coefficient of variation method based on the multiple feature quantities to obtain the comprehensive weight of each feature quantity as the fuzzy density. The fuzzy measure determination module is used to determine the interaction parameters used to characterize the interaction relationship between features based on the fuzzy density, and to calculate the fuzzy measure of any non-empty subset among multiple feature quantities based on the interaction parameters; wherein, the fuzzy measure is used to characterize the importance of feature subsets and the interaction relationship between them. A module is established to establish the membership function of each feature quantity for multiple preset DC bias levels, so as to form a fuzzy membership evaluation matrix. The comprehensive evaluation value calculation module is used to perform fusion calculation for each level of magnetic bias, based on the membership vector and fuzzy measure corresponding to that level in the fuzzy membership evaluation matrix, to obtain the comprehensive evaluation value for each level. The determination module is used to determine the final DC bias of the transformer based on the comprehensive evaluation values ​​of each level.

9. An electronic device, characterized in that, include: A memory, and one or more processors communicatively connected to the memory; The memory stores instructions that can be executed by the one or more processors to cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images