CN108872777A - Winding in Power Transformer state evaluating method based on improved system delay Order- reduction - Google Patents

Abstract

The invention discloses a kind of Winding in Power Transformer state evaluating methods based on improved system delay Order- reduction, this method is based on basket vibration mechanism of production, preferably reflect the mechanical structure state of winding, winding failure is directly associated with characteristic quantity, provides foundation for the validity and science of fault diagnosis.Before the present invention is compared to improvement, extracted characteristic quantity and the practical order of system are more closely, can more accurately reflect system performance and winding machinery configuration state；In addition, the present invention implements and transformer is without any electrical connection, and without powering off to transformer, to the influence on system operation very little of entire electric system.

Description

Winding State Evaluation of Power Transformer Based on Improved System Delay Order Estimation method

technical field

The invention belongs to the technical field of power transformer safety fault detection, and in particular relates to a power transformer winding state evaluation method based on an improved system delay order estimation.

Background technique

As a very important part of the power system, large power transformers play an important role in ensuring the safe and reliable power supply of the power grid. In power transformer faults, the proportion of winding faults is as high as 46.4%, which is the most important part of transformer faults, and the serious faults caused by mechanical deformation of windings under the action of electrodynamic force account for 70% of the total faults of windings. Even a slight deformation of the winding will cause deterioration of the mechanical properties of the winding, reduction of insulation strength and short-circuit resistance, and other problems, which will bring great safety hazards. Therefore, it is very necessary and important to realize the state monitoring and evaluation of the main faulty component winding in the transformer live monitoring.

The vibration analysis method analyzes the collected vibration signals of the oil tank wall to obtain the status information of the internal components of the transformer. When the internal mechanical structure of the transformer changes, it will inevitably lead to changes in the system response of the vibration system. Therefore, the relationship model between the input electromagnetic force and the vibration response of the output winding can be constructed, and the mechanical structure state of the transformer winding can be reflected by extracting the system characteristics of the model.

The Chinese patent with the application number 201310047702.0 proposes a power transformer fault diagnosis method based on the electric-vibration model, which establishes the vibration frequency components by collecting the voltage signal, current signal, oil temperature signal and multiple vibration measuring points of the transformer. and the regression model between current, voltage and oil temperature, and compare the measured data of vibration with the predicted data obtained through the model to diagnose the fault of the transformer; although this method considers the nonlinearity in the vibration of the transformer to a certain extent, However, there are still some problems that some frequency components introduced by strong nonlinearity such as hysteresis effect cannot be modeled or the model error is large, so this method still has certain limitations in accuracy and effectiveness.

The Chinese patent with the application number 201710867458.0 proposes and constructs a Hammerstein nonlinear model between current and winding vibration based on the principle of transformer winding vibration, and extracts the system characteristics of the model - the system delay order, reflecting the transformer winding The state of the mechanical structure; this method takes into account the physical mechanism of the influence of mechanical faults on the characteristics of the model system, and only extracts part of the fault-related information in the model as a basis for diagnosis, which can effectively avoid misjudgments directly caused by model errors. When the delay order extraction algorithm has a large difference between the input and output orders of the system, the estimation error will be large or even wrong, which will affect the accuracy of the winding state evaluation results.

Contents of the invention

In view of the above, the present invention provides a power transformer winding state evaluation method based on the improved system delay order estimation, which can detect the mechanical structure state of the power transformer winding online, and improves the order estimation when the input and output order differences are large. Inaccurate problems, so that the extracted parameters can more accurately reflect the actual state of the winding.

A method for evaluating the state of a power transformer winding based on an improved system delay order estimation, comprising the following steps:

(1) Dispersively arrange multiple vibration sensors at the positions corresponding to the windings on the surface of the power transformer oil tank, record the vibration signals of each vibration sensor of the power transformer under low-voltage load operating conditions, and collect the primary side current of the power transformer synchronously;

(2) Preprocessing the vibration signal and the current signal, including demagnetization of the current signal;

(3) For any vibration sensor, the relationship model between current and winding vibration is constructed, and the delay order n of the model is estimated as the characteristic quantity of the system characteristic according to the preprocessed vibration signal and current signal of the vibration sensor, n= Na + N _b , _Na and N _b are the actual linear delay orders of the system output and input respectively, and both _Na and N _b are natural numbers; and _then judge based on the vibration sensor data based on n, _Na and N _b The state of mechanical structure of power transformer windings;

(4) Traverse all vibration sensors according to step (3). When the mechanical structure state of the winding is determined to be normal based on a certain proportion of the vibration sensors, it is finally determined that the power transformer winding is normal, otherwise it is determined that the power transformer winding is abnormal.

Further, the specific process of preprocessing the vibration signal and the current signal in the step (2) is: first, normalize the vibration signal and the current signal; then, according to the following formula, normalize the current The signal is transformed nonlinearly:

Among them: i(t) is the current signal after normalization, t is the time, _ip (t) is the current signal after nonlinear transformation.

Further, the specific process of estimating the model delay order n in the step (3) is as follows:

3.1 Calculate the Lipschitz coefficients of all sampling point combinations in the relational model according to the following formula:

Among them: l _ij ⁽ⁿ⁾ represents the Lipschitz coefficient of the i-th sampling point and the j-th sampling point combination in the relationship model when the delay order is n, i and j are the serial numbers of the sampling points; when n is an even number, γ ₁ ( i)～γ _n (i) corresponds to y(i-1),x(i-1),y(i-2),x(i-2),...,y(iN _a ),x( iN _b ), γ ₁ (j)～γ _n (j) corresponds to y(j-1),x(j-1),y(j-2),x(j-2),...,y (jN _a ),x(jN _b ); when n is an odd number, γ ₁ (i)～γ _n (i) corresponds to y(i-1),x(i-1),y(i-2) ,x(i-2),...,y(iN _b ),x(iN _b ),y(iN _a ), γ ₁ (j)～γ _n (j) correspond to y(j-1), x(j-1),y(j-2),x(j-2),...,y(jN _b ),x(jN _b ),y(jN _a ); y(i) and y( j) are the signal values corresponding to the i-th sampling point and the j-th sampling point in the vibration signal preprocessed by the vibration sensor respectively, and y(i-1) and y(j-1) are the vibration signal preprocessed by the vibration sensor y(i-2) and y(j-2) are the signal values corresponding to the i-1th sampling point and the j-1th sampling point in the vibration sensor, respectively, corresponding to the i-2th sampling point in the vibration signal preprocessed by the vibration sensor and the signal value of the j-2th sampling point, y(iN _a ) and y(jN _a ) are the signal values corresponding to the iN _a -th sampling point and the jN _a -th sampling point in the vibration signal preprocessed by the vibration sensor respectively, y (iN _b ) and y(jN _b ) are the signal values corresponding to the iN _b sampling point and the jN _b sampling point in the vibration signal preprocessed by the vibration sensor, x(i-1) and x(j-1) are the signal values corresponding to the i-1th sampling point and the j-1th sampling point in the processed current signal respectively, and x(i-2) and x(j-2) are the signal values corresponding to the i-th sampling point in the processed current signal respectively. -2 sampling point and the signal value of the j-2th sampling point, x(iN _b ) and x(jN _b ) are respectively the signal values corresponding to the iN _b sampling point and the jN _b sampling point in the processed current signal;

3.2 Sort all the Lipschitz coefficients calculated in step 3.1 from large to small and intercept the first m Lipschitz coefficients, and then calculate the average Lipschitz coefficient of the relational model when the delay order is n according to the following formula:

Among them: l ⁽ⁿ⁾ represents the Lipschitz average coefficient of the relationship model when the delay order is n, l ⁽ⁿ⁾ (z) is the zth Lipschitz coefficient sorted from large to small, m usually takes 0.01N _set , N _set is the total number of sampling points;

3.3 Initialize delay order n=2 and N _a =N _b =1;

3.4 Add 1 to Na, and then calculate the corresponding _Lipschitz average coefficient _l (Na ₊ ^Nb) and _l ⁽ _Na- ^1+Nb) and make the following judgments:

If l ^(Na+Nb) -l ^(Na-1+Nb) ≤ε×l (Na-1 ₊ ^Nb) and N _b is not determined, determine that Na is the value before this accumulation and perform step 3.5;

If l ^(Na+Nb) -l ^(Na-1+Nb )≤ε×l (Na-1 ₊ ^Nb) and N _b has been determined, then determine Na as the value before this accumulation and make the determined _Na The addition of N _b is the delay order n;

If l ^(Na+Nb) -l ^(Na-1+Nb )>ε×l ^(Na-1+Nb) and N _b is not determined, then perform step 3.5;

If l ^(Na+Nb) -l ^(Na-1+Nb) >ε× _l (Na-1 ₊ ^Nb) and N _b has been determined, repeat step 3.4 until Na is determined and the determined Na and N The addition of _b is the delay order n;

3.5 Add 1 to N _b , and then calculate the corresponding Lipschitz average coefficient _{l (} Na ₊ ^Nb) and l ₍ ^{Na+ Nb-1)} and make the following judgments:

If l ^(Na+Nb) -l ^(Na+Nb-1) ≤ε×l ^(Na+Nb-1) and N _a is not determined, then determine N _b as the value before this accumulation and return to step 3.4;

If l ^(Na+Nb) -l ^(Na+Nb-1) ≤ε×l ^(Na+Nb-1) and N _a has been determined, then determine N _b as the value before this accumulation and make the determined N _a The addition of N _b is the delay order n;

If l ^(Na+Nb) -l ^(Na+Nb-1) >ε× _l ^(Na+Nb-1) and Na is not determined, return to step 3.4;

If l ^(Na+Nb) -l ^(Na+Nb-1) >ε× _l ^(Na+Nb-1) and Na has been determined, then repeat step 3.5 until N _b is determined and the determined _Na and N The addition of _b is the delay order n;

Where: ε is the convergence coefficient.

Further, when neither N _a nor N _b is determined, the convergence coefficient ε=ε ₁ ; when one of N _a and N _b is determined, the convergence coefficient ε=ε ₂ ; where ε ₁ and ε ₂ are set respectively two different constants.

Further, the specific process of judging the state of the mechanical structure of the power transformer winding in the step (3) is as follows: first, the delay order n of the model is calculated according to the steps (1) to (3) under the condition that the power transformer winding is normal And _{Na and N b ;} Then, make n, _Na and N _{b obtained under the actual measurement and calculation based on the current vibration sensor data situation in step (3) and n, Na and N b under normal} conditions of the power transformer winding Yes, if the two data match, it is determined that the mechanical structure state of the power transformer winding based on the current vibration sensor data is normal; if the two data do not match, then it is determined that the mechanical structure state of the power transformer winding is based on the current vibration sensor data is abnormal.

Based on the above technical scheme, the beneficial technical effects of the present invention are as follows:

1. The present invention does not have any electrical connection with the transformer, and does not need to cut off the power of the transformer, which has little impact on the operation of the entire power system.

2. The method of the present invention is based on the winding vibration generation mechanism, better reflects the mechanical structure state of the winding, directly associates the winding fault with the characteristic quantity, and provides a basis for the validity and scientificity of the fault diagnosis.

3. Compared with the method before the improvement, the extracted feature quantity is closer to the actual order of the system, and can more accurately reflect the characteristics of the system and the state of the mechanical structure of the winding.

Description of drawings

Fig. 1 is a schematic flow chart of the steps of the method of the present invention.

Figure 2 is a layout diagram of measuring points on the surface of the power transformer oil tank.

Figure 3 shows the input current i(t) and the vibration component relationship graph.

Figure 4 shows the input current i _p (t) and the vibration component after demagnetization transformation relationship graph.

FIG. 5 is a schematic flow chart of the improved delay order estimation algorithm of the present invention.

Figure 6(a) is a schematic diagram of the order estimation process and results of the original algorithm.

Fig. 6(b) is a schematic diagram of the order estimation process and results of the improved algorithm of the present invention.

Detailed ways

In order to describe the present invention more specifically, the technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

The specific experimental object of this embodiment is a 110kV three-phase oil-immersed power transformer. In order to verify the effectiveness of the method of the present invention, the vibration characteristics of the same winding under normal conditions and fault conditions were deliberately compared. Faulty windings are caused by man-made damage to normal windings through short-circuit impact. Transformer short-circuit test is a special test method for windings. By short-circuiting the low-voltage end and applying voltage on the high-voltage end, the current of the transformer winding can reach the rated value.

In diagnosis, in order to obtain vibration signals of different amplitudes without distortion, a vibration sensor with high sensitivity must be selected; in order to ensure the vibration response of the sensor within the sampling filter frequency band, the vibration sensor should be fixed on the side wall of the fuel tank. It adopts the method of magnetic seat adsorption or glue bonding. The vibration sampling device includes main modules such as preamplification, anti-aliasing filter, and AD sampling. The number of AD sampling bits is at least 12, and the cut-off frequency of the anti-aliasing filter is 2000 Hz; when sampling vibration signals, the sampling frequency is at least 4000 Hz. In this embodiment, the sampling frequency for collecting vibration signals is set to 10000 Hz, the number of sampling bits of the AD module is 16 bits, and the continuous sampling mode is used to record the whole process of the experiment.

As shown in Figure 1, the present invention is based on the power transformer winding state evaluation method of improved system delay order estimation, comprises the following steps:

(1) Arrange vibration measuring points.

Five measuring points are arranged on the surface of the oil tank of the power transformer, as shown in Figure 2, because the A-phase winding will be artificially damaged in the experiment, so the five measuring points are arranged on the oil tank wall corresponding to the A-phase winding.

(2) Collect vibration signals and current signals under normal and abnormal conditions.

First conduct a short-circuit test on a new transformer, gradually increase the voltage on the high-voltage side to increase the current on the winding, increase the current by 10% each time, and gradually reach the rated value; after the current increases by 10%, keep it stable for 30 seconds Change, use the continuous sampling mode to record the vibration of all measuring points at all times.

Then use the short-circuit impact current to artificially destroy the A-phase winding, confirm that the A-phase winding has a large degree of deformation in the middle, and then conduct the same short-circuit experiment on the transformer, and also record the vibration of all measuring points during the whole process.

(3) Preprocess the signal.

When the input current i(t) is taken as the independent variable, the winding vibration is at the measuring point Contributed vibration component When plotting the relationship curve between the two for the dependent variable (measured by point For example, as shown in Figure 3), it can be seen that the output-input relationship curve presents obvious hysteresis characteristics, which is due to the strong nonlinear characteristics introduced by the magnetic field where the winding is located. After the following nonlinear transformation is performed on the input signal:

Among them: i'(t) is the differential form of the input signal i(t), and i _p (t) is the transformed input signal. This nonlinear change maps the original input signal from its low-latitude space to a high-dimensional space, and removes the hysteresis characteristic of the original nonlinear module in the system. Figure 4 shows the input current after the above transformation i _p (t) and vibration components relationship curve.

(4) Use the improved order estimation algorithm to extract the delay order of the relational model to judge the state of the winding mechanical structure.

It can be seen from the vibration mechanism of the transformer that the unit electrodynamic distribution force of the winding is mainly generated by the current flowing through the winding coil, and it is related to the leakage magnetic field near the winding. The height of the coil is related; thus the unit electromotive force distribution can be Treated as a nonlinear function F ₁ (i(t)) that depends only on the current i(t) and a nonlinear function that depends only on the coil position The combination. Therefore, the relationship formula between winding vibration and input current can be written as:

in: Indicates the winding vibration excitation force to the tank wall response point The overall equivalent unit impulse response at .

Observing the above formula, we can see that the current-winding vibration relationship model consists of a static nonlinear module F ₁ () and a dynamic linear module The composed nonlinear system is consistent with the structure of the classical nonlinear Hammerstein model, that is, it consists of a nonlinear static module at the input end and a linear dynamic module connected in series.

From the above model, it can be seen that the mechanical structure characteristics of the winding will directly affect the nonlinear function related to the position The characteristics and transfer characteristics will directly affect the characteristics of the linear module in the model, so the system characteristics of the linear module can be extracted as the characteristic quantity of the mechanical structure state of the winding.

For a linear system, the change of its delay order will directly affect its system response. Therefore, if the delay order of the system can be directly estimated according to the characteristics of the input and output signals, the change of the delay order can be used to analyze the mechanical properties of the transformer winding. Structural features are evaluated.

For a nonlinear system, the output y(t) can be described as a function related to the delayed input and output, namely:

y(t)=g(y(t-1),...,y(tN _a ),x(t-1),...,x(tN _b ))

＝g(γ ₁ ,γ ₂ ,…,γ _n )

Where: x(t) and y(t) are the input and output of the nonlinear system respectively, and N _a and N _b are the actual linear delay orders of the system output and input respectively, g() is a nonlinear function, and Assume that the function has continuous characteristics and satisfies the Lipschitz continuity condition. Here, γ _i , i=1,...,n are used as the labels of the independent variables of the nonlinear function, and n=N _a +N _b is the number of independent variables.

Define the Lipschitz coefficient l _ij ⁽ⁿ⁾ to characterize the continuity of the nonlinear function g(γ ₁ ,γ ₂ ,…,γ _n ):

When the necessary variable γ _h is missing in the independent variable of the function, the Lipschitz coefficient l _ij ^(h-1) value when the variable is absent will be much larger than the coefficient value l _ij ^(h) when the necessary variable γ _h exists, that is, when l _ij ^(h-1) ＞＞l _ij ^(h) ; in contrast, if the variable γ _h+1 is a redundant or unnecessary independent variable, then its corresponding Lipschitz coefficient l _ij ^(h+1) and When this variable is missing, the values of the coefficient l _ij ^(h) are approximately equal, that is, l _ij ^(h+1) ≈ l _ij ^(h) .

In order to avoid the influence of the noise introduced during the measurement on the above results, the Lipschitz average coefficient l ^{(n) is} used to replace the coefficient l _ij ⁽ⁿ⁾ :

Among them: l ⁽ⁿ⁾ (z) is the zth coefficient value of l _ij ⁽ⁿ⁾ reordered according to the decreasing rule, m usually uses 0.01N _set , and N _set is the number of input-output pairs used for order estimation .

The main idea of improving the pre-order extraction algorithm is to find the minimum number of independent variables n(n=N _a +N _b ), and make its corresponding Lipschitz average coefficient meet the following conditions:

Among them: ε is an empirical threshold value, generally ε=0.1. However, it is not difficult to see that the order obtained by the above method, the order N _a and N _b are alternately stepped up in the process of finding the optimal order. Therefore, when the optimal order n is finally determined, when the order n is an even number, When the order n is an even number, Na _a =N _b +1. However, for an actual system, its actual order _{Na and N b are} often not equal. Taking the case of N _a << N _b as an example, in the order estimation process, the order _{Na and N b are} alternately increased in steps until N _a approaches the actual value, the order estimation process meets the termination condition and the process is terminated early , and the order N _b is far smaller than the actual value at this time, which is not accurate.

Therefore, in order to accurately estimate the delay order, thereby more accurately evaluating the system state, the method of the present invention adopts an improved order estimation algorithm, and its steps are shown in Figure 5:

① Firstly, the delay order n is initialized to 2, that is, _{Na = 1, N b =} 1, and the current Lipschitz coefficient l ⁽²⁾ is recorded and calculated.

②When the order N _a is not determined, increment the undetermined order, set N _a =N _a +1, record and calculate the current Lipschitz average coefficient If the optimal order N _a has been determined, go to step ④.

③ comparison and If the difference between the two satisfies the condition Then the estimation of the optimal order N _a ends, and its value N _a =N _a -1; if not satisfied, execute the next step.

④ When the order N _b is not determined, increment the undetermined order, set N _b = N _b +1, record and calculate the current Lipschitz average coefficient If the optimal order N _b has been determined, the algorithm stops and the order estimation ends.

⑤ Contrast and If the difference between the two satisfies the condition Then the estimation of the optimal order N _b ends, and its value N _b =N _b -1 is set; if the condition is not met, step ② is performed again.

In the above algorithm, ε is the set convergence threshold, and Where ε ₁ =0.1, ε ₂ =0.04.

In order to verify the accuracy and effectiveness of the above algorithm, and to illustrate its advancement compared with the improved original algorithm, firstly, the delay order of a nonlinear system is estimated by mathematical simulation, and the actual order is compared with the estimated order. The nonlinear system is composed of a nonlinear module and a linear module connected in series, where the relational expression of the nonlinear module is expressed as:

v(t)=sin(0.6πx(t))+0.1 cos(1.5πx(t))

And the difference equation characterizing the linear module is:

y(t)=-0.28y(t-1)+0.47y(t-6)+0.11v(t-1)+0.08v(t-3)

It can be seen from the above formula that the actual delay order of the system is N ^a =6, N ^b =3, and the system input x(t) is defined as a random sequence distributed in (-1,1).

Use 500 data pairs for order estimation, and follow γ ₁ =y(t-1), γ ₂ =x(t-1), γ ₃ =y(t-2), γ ₄ =x(t-2 ), ... define potential input variables, and add variable units sequentially according to the steps in the algorithm and calculate the corresponding Lipschitz coefficients. Figure 6(a) and Figure 6(b) show the difference ratio of Lipschitz coefficients obtained by using the original order estimation algorithm and the improved algorithm respectively It is not difficult to find from the figure that Δl ⁽³⁺³⁾ ≤ 0.1l ⁽³⁺³⁾ in the original algorithm, that is, the algorithm termination condition is satisfied at the order (3,3), the algorithm is terminated early, and the algorithm is terminated with (3,3 ) as the final order estimate, which obviously does not match the actual order. Analyzing the order estimation process shows that at the order (3,3), since the actual order N _b = 3 is equal to the actual value, the termination condition is satisfied, while another order N _a = 3 is still far away. less than the actual order of 6. However, when using the improved algorithm for order estimation, the process first satisfies the termination condition at the order (3,3) for the first time, and selects Na ₌ 3 as the final estimated value, and then the algorithm continues, and Gradually increase the N _b order value until the stop condition is met again at the order (6,3), and end the algorithm, and finally select N _a =3 and N _b =3 as the optimal estimated value of the order, and the actual order value match, the algorithm result is correct. The above simulation results show that the improved algorithm can still accurately and effectively estimate the delay order of the nonlinear system when _{Na and N b are} quite different from the actual order. Therefore, the order estimation algorithm can be used to evaluate the system characteristic change of the linear module of the winding vibration system, so it can be further applied to the evaluation and diagnosis of the mechanical structure state of the transformer winding.

In the winding fault detection experiment, the input and output data with a time length of 0.5s are selected to estimate the delay order, and the delay order of the vibration-current relationship model of all measuring points under normal winding and abnormal winding state is estimated, as shown in Table 1. Show the result:

Table 1. Order estimation results of all measuring points in different winding states (100% load)

It can be seen from Table 1 that when the mechanical structure of the winding is abnormal (deformed), the delay order of the current-vibration relationship model corresponding to most of the measuring points will be found to change significantly. Condition monitoring of the structural state.

The above description of the embodiments is for those of ordinary skill in the art to understand and apply the present invention. It is obvious that those skilled in the art can easily make various modifications to the above-mentioned embodiments, and apply the general principles described here to other embodiments without creative efforts. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art according to the disclosure of the present invention should fall within the protection scope of the present invention.

Claims (5)

1. A method for evaluating the winding state of a power transformer based on an improved system delay order estimation, comprising the steps of: (1) Dispersively arrange multiple vibration sensors at the positions corresponding to the windings on the surface of the power transformer oil tank, record the vibration signals of each vibration sensor of the power transformer under low-voltage load operating conditions, and collect the primary side current of the power transformer synchronously; (2) Preprocessing the vibration signal and the current signal, including demagnetization of the current signal; (3) For any vibration sensor, the relationship model between current and winding vibration is constructed, and the delay order n of the model is estimated as the characteristic quantity of the system characteristic according to the preprocessed vibration signal and current signal of the vibration sensor, n= Na + N _b , _Na and N _b are the actual linear delay orders of the system output and input respectively, and both _Na and N _b are natural numbers; and _then judge based on the vibration sensor data based on n, _Na and N _b The state of mechanical structure of power transformer windings; (4) Traverse all vibration sensors according to step (3). When the mechanical structure state of the winding is determined to be normal based on a certain proportion of the vibration sensors, it is finally determined that the power transformer winding is normal, otherwise it is determined that the power transformer winding is abnormal. 2. The power transformer winding state evaluation method according to claim 1, characterized in that: the specific process of preprocessing the vibration signal and the current signal in the step (2) is: first, the vibration signal and the current signal are Normalization processing; then, according to the following formula, the normalized current signal is nonlinearly transformed: Among them: i(t) is the current signal after normalization, t is the time, _ip (t) is the current signal after nonlinear transformation. 3. power transformer winding state evaluation method according to claim 1, is characterized in that: the specific process of estimating model delay order n in described step (3) is as follows: 3.1 Calculate the Lipschitz coefficients of all sampling point combinations in the relational model according to the following formula: Among them: l _ij ⁽ⁿ⁾ represents the Lipschitz coefficient of the i-th sampling point and the j-th sampling point combination in the relationship model when the delay order is n, i and j are the serial numbers of the sampling points; when n is an even number, γ ₁ ( i)～γ _n (i) corresponds to y(i-1),x(i-1),y(i-2),x(i-2),...,y(iN _a ),x( iN _b ), γ ₁ (j)～γ _n (j) corresponds to y(j-1),x(j-1),y(j-2),x(j-2),...,y (jN _a ),x(jN _b ); when n is an odd number, γ ₁ (i)～γ _n (i) corresponds to y(i-1),x(i-1),y(i-2) ,x(i-2),...,y(iN _b ),x(iN _b ),y(iN _a ), γ ₁ (j)～γ _n (j) correspond to y(j-1), x(j-1),y(j-2),x(j-2),...,y(jN _b ),x(jN _b ),y(jN _a ); y(i) and y( j) are the signal values corresponding to the i-th sampling point and the j-th sampling point in the vibration signal preprocessed by the vibration sensor respectively, and y(i-1) and y(j-1) are the vibration signal preprocessed by the vibration sensor y(i-2) and y(j-2) are the signal values corresponding to the i-1th sampling point and the j-1th sampling point in the vibration sensor, respectively, corresponding to the i-2th sampling point in the vibration signal preprocessed by the vibration sensor and the signal value of the j-2th sampling point, y(iN _a ) and y(jN _a ) are the signal values corresponding to the iN _a -th sampling point and the jN _a -th sampling point in the vibration signal preprocessed by the vibration sensor respectively, y (iN _b ) and y(jN _b ) are the signal values corresponding to the iN _b sampling point and the jN _b sampling point in the vibration signal preprocessed by the vibration sensor, x(i-1) and x(j-1) are the signal values corresponding to the i-1th sampling point and the j-1th sampling point in the processed current signal respectively, and x(i-2) and x(j-2) are the signal values corresponding to the i-th sampling point in the processed current signal respectively. -2 sampling point and the signal value of the j-2th sampling point, x(iN _b ) and x(jN _b ) are respectively the signal values corresponding to the iN _b sampling point and the jN _b sampling point in the processed current signal; 3.2 Sort all the Lipschitz coefficients calculated in step 3.1 from large to small and intercept the first m Lipschitz coefficients, and then calculate the average Lipschitz coefficient of the relational model when the delay order is n according to the following formula: Among them: l ⁽ⁿ⁾ represents the Lipschitz average coefficient of the relationship model when the delay order is n, l ⁽ⁿ⁾ (z) is the zth Lipschitz coefficient sorted from large to small, m usually takes 0.01N _set , N _set is the total number of sampling points; 3.3 Initialize delay order n=2 and N _a =N _b =1; 3.4 Add 1 to Na, and then calculate the corresponding _Lipschitz average coefficient _l (Na ₊ ^Nb) and _l ⁽ _Na- ^1+Nb) and make the following judgments: If l ^(Na+Nb) -l ^(Na-1+Nb) ≤ε×l (Na-1 ₊ ^Nb) and N _b is not determined, determine that Na is the value before this accumulation and perform step 3.5; If l ^(Na+Nb) -l ^(Na-1+Nb) ≤ε×l (Na-1 ₊ ^Nb) and N _b has been determined, then determine Na as the value before this accumulation and make the determined _Na The addition of N _b is the delay order n; If l ^(Na+Nb) -l ^(Na-1+Nb) >ε×l ^(Na-1+Nb) and N _b is not determined, then perform step 3.5; If l ^(Na+Nb) -l ^(Na-1+Nb) >ε× _l (Na-1 ₊ ^Nb) and N _b has been determined, repeat step 3.4 until Na is determined and the determined Na and N The addition of _b is the delay order n; 3.5 Add 1 to N _b , and then calculate the corresponding Lipschitz average coefficient _{l (} Na ₊ ^Nb) and l ₍ ^{Na+ Nb-1)} and make the following judgments: If l ^(Na+Nb) -l ^(Na+Nb-1) ≤ε×l ^(Na+Nb-1) and N _a is not determined, then determine N _b as the value before this accumulation and return to step 3.4; If l ^(Na+Nb) -l ^(Na+Nb-1) ≤ε×l ^(Na+Nb-1) and N _a has been determined, then determine N _b as the value before this accumulation and make the determined N _a The addition of N _b is the delay order n; If l ^(Na+Nb) -l ^(Na+Nb-1) >ε× _l ^(Na+Nb-1) and Na is not determined, return to step 3.4; If l ^(Na+Nb) -l ^(Na+Nb-1) >ε× _l ^(Na+Nb-1) and Na has been determined, then repeat step 3.5 until N _b is determined and the determined _Na and N The addition of _b is the delay order n; Where: ε is the convergence coefficient. 4. The power transformer winding state evaluation method according to claim 3, characterized in that: when _{Na and N b are not determined, the convergence coefficient ε=ε 1; when one of Na and N b is determined , the} convergence Coefficient ε=ε ₂ ; where ε ₁ and ε ₂ are two different constants set respectively. 5. The method for evaluating the state of a power transformer winding state according to claim 1, characterized in that: the specific process of judging the state of the mechanical structure of the power transformer winding in the step (3) is as follows: Steps (1) to (3) process to calculate the delay order n and Na and N _b of the model; then, make n, _Na and N _b actually measured and calculated based on the current vibration sensor data in step (3 ₎ Compare with n, N _a and N _b of the power transformer winding under normal conditions, if the two data are consistent, it is determined that the mechanical structure state of the power transformer winding based on the current vibration sensor data is normal; if the two data do not match , it is determined that the mechanical structure state of the power transformer winding is abnormal based on the current vibration sensor data.