CN108362488A - OLTC mechanical failure diagnostic methods based on MPE and SVM - Google Patents

Abstract

The invention discloses a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM, include the following steps：1）The vibration signal under load ratio bridging switch OLTC normal conditions, the vibration signal under malfunction are acquired, and are pre-processed to vibration signal by acceleration transducer；2）Multiple dimensioned arrangement entropy MPE is carried out to collected vibration signal to calculate, construction feature vector, the input as support vector machines；3）By step 2）Obtained feature vector is input in support vector machines, is trained to support vector machines, and test data is input to trained SVM, to judge that the fault mode of OLTC, the present invention do not need the training that mass data carries out SVM, diagnostic accuracy higher；BP neural network is substantially better than to the diagnosis effect of OLTC.

Description

OLTC Mechanical Fault Diagnosis Method Based on MPE and SVM

technical field

The invention relates to the technical field of on-load tap changer fault diagnosis, in particular to a method for diagnosing mechanical faults of an on-load tap changer OLTC based on MPE and SVM.

Background technique

With the improvement of power quality requirements, automatic voltage control and other systems are widely used in the power grid. The existing on-load tap-changer (OLTC) is adjusted quite frequently, and the failure rate is very high. According to statistics at home and abroad, tap changer failures account for more than 20% of transformer failures, and they are mainly mechanical failures. If they are not discovered and dealt with in time, their failures will seriously damage the inherent structure of OLTC and transformers, and affect the normal operation of power equipment and systems. Operate safely with serious consequences. Therefore, in order to ensure the safe and reliable operation of the tap changer, it is necessary to carry out relevant research on the mechanical fault diagnosis method of the tap changer.

During the operation of the on-load tap-changer, the collision or friction between the mechanism parts will generate vibration signals, which contain rich equipment status information. At present, the analysis based on vibration signal has become an important means of mechanical fault diagnosis of on-load tap-changer. The existing vibration signal analysis methods include wavelet singularity detection, self-organizing map method, EMD (empirical mode decomposition) and wavelet packet, etc. Most of these methods decompose the non-stationary signal into the sum of several simple stationary signals, and then process each component to extract time-frequency features. However, studies have shown that the vibration signal during the OLTC switching process exhibits obvious nonlinear behavior, and the time-frequency analysis method is used to decompose the signal into a smooth signal, which inevitably has certain limitations. Therefore, the present invention uses a multi-scale permutation entropy nonlinear analysis method to diagnose OLTC mechanical faults, and can directly extract fault information from mechanical vibration signals that cannot be extracted by other methods.

Aiming at the nonlinear characteristics of OLTC mechanical vibration signals, the present invention starts from the time series randomness and dynamic mutation characteristics of OLTC vibration signals, and applies multi-scale permutation entropy (MPE) to the extraction of OLTC fault features. Since support vector machine (SVM) analysis has a good diagnostic effect in fault diagnosis of small sample data, on the basis of extracting fault features from MPE, combined with SVM as fault type judgment, a load classification method based on MPE and SVM is proposed. Take the switch mechanical fault diagnosis method and apply it to the analysis of OLTC experimental data. The results show that this method can effectively diagnose OLTC mechanical failure types.

Contents of the invention

In order to solve the deficiencies in the prior art, the present invention provides a method for diagnosing mechanical faults of on-load tap-changers based on MPE and SVM. The diagnosis result is indeed high in accuracy, simple in structure and strong in operability.

In order to achieve the above object, the present invention adopts following technical scheme: a kind of OLTC mechanical fault diagnosis method based on MPE and SVM, it is characterized in that: comprise the following steps:

1) Collect the vibration signals of the on-load tap-changer OLTC under normal conditions and fault conditions through the acceleration sensor, and preprocess the vibration signals;

2) Carry out multi-scale permutation entropy MPE calculation on the collected vibration signal, construct feature vector, as the input of support vector machine SVM;

3) Input the feature vector obtained in step 2) into the support vector machine SVM, train the support vector machine SVM, and input the test data into the trained SVM, so as to judge the failure mode of the OLTC.

The aforementioned method for OLTC mechanical fault diagnosis based on MPE and SVM is characterized in that: the preprocessing specifically includes: performing noise reduction processing on the collected vibration signals.

The aforementioned method for diagnosing mechanical faults of OLTC based on MPE and SVM is characterized in that: the fault state includes loosening of OLTC contacts of on-load tap changer and degradation of spring performance.

The above-mentioned OLTC mechanical fault diagnosis method based on MPE and SVM is characterized in that: the acceleration sensor is installed on the top of the tap changer.

Aforesaid a kind of OLTC mechanical fault diagnosis method based on MPE and SVM is characterized in that: the specific steps of calculating the multi-scale permutation entropy MPE are:

1) Coarse-grain the original time series {x ₁ ,x ₂ ,…,x _N }, and construct a multi-scale time series {y _l ^(s) } according to the following formula:

In the formula, s is the scale factor, N is the length of the original time series, x _i represents the data points in the original time series, the range of i is 1-N, y _l ^(s) represents the lth time series under the s window average value;

2) After the coarse-grained processing of the original time series, the normalized permutation entropy of each coarse-grained time series is calculated, that is, the multi-scale permutation entropy is obtained, which constitutes the feature quantity.

Aforesaid a kind of OLTC mechanical fault diagnosis method based on MPE and SVM is characterized in that: the permutation entropy concrete calculation steps after described normalization are:

Assuming a set of time series {xi | _i =1,2,…,N}, perform phase space reconstruction on them to obtain the reconstructed time series X _i :

X _i ＝[x _i ,x _i+τ ,…,x _i+(m-1)τ ] (1)

Among them, m is the embedding dimension, τ is the delay time, x _i is the i-th data point in the time series _Xi , x _i+τ is the i+τ data point in the time series _Xi , x _{i+(m- 1) τ} is the i+(m-1)τ data point in the time series X _i , and the m data points in the time series X _i are arranged in ascending order, that is r represents the position of the data point in the reconstructed sequence _Xi , It is the second data point after the data points in the reconstructed sequence _Xi are arranged in ascending order, similarly is the mth data point after the data points in the reconstructed sequence _Xi are arranged in ascending order; when there is when the data point Arrange according to the size of r _j and r _k , that is, if r _j < r _k , then consider Represents the r _jth data point in the reconstructed sequence _Xi , r _j and r _k represent the location of the data point in the reconstructed sequence _Xi , Represents the r _kth data point in the reconstructed sequence _Xi ;

Time series X _i have m! Arrangement, for any arrangement ω, T(ω) represents the number of times it appears, then the probability of its occurrence is:

Therefore, the permutation entropy _HPE of time series _Xi can be defined as:

H _PE ＝-∑P(ω)lnP(ω) (3)

After normalization, the normalized permutation entropy PE is obtained:

The aforementioned method for OLTC mechanical fault diagnosis based on MPE and SVM is characterized in that: the vibration data is divided into two groups, each group includes normal vibration signals and fault signals, one group is used for SVM training, and the other group is used for The test of SVM; in step 3), earlier the feature vector is normalized between [0,1], the training feature vector after normalization is input in the support vector machine SVM, the support vector machine is trained, Then the normalized test feature vector is input to the trained SVM to judge the failure mode of OLTC.

The beneficial effect that the present invention reaches:

1. The present invention uses a multi-scale permutation entropy nonlinear analysis method to diagnose OLTC mechanical faults, which can directly extract fault information that cannot be extracted by other methods in mechanical vibration signals, such as the randomness of fault information;

2. The present invention does not require a large amount of data for SVM training, and the diagnosis accuracy is higher;

3. The diagnostic effect of the present invention on OLTC is obviously better than that of BP neural network.

Description of drawings

Figure 1 is a feature extraction process diagram based on multi-scale permutation entropy;

Figure 2 is the multi-scale permutation entropy MPE distribution diagram in three states.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention.

A method for diagnosing mechanical faults of an on-load tap changer OLTC based on MPE and SVM, comprising the following steps:

1) Collect the vibration signal of the on-load tap-changer OLTC under normal state and fault state through the acceleration sensor, and preprocess the vibration signal, that is, perform noise reduction processing on the collected vibration signal;

2) Perform multi-scale permutation entropy MPE analysis on the collected vibration signals, construct feature vectors, and use them as the input of support vector machine SVM;

3) Input the feature vector constructed in step 2) into the support vector machine SVM, train the support vector machine SVM, and input the test data into the trained SVM, so as to judge the failure mode of the OLTC.

In step 1), the acceleration sensor adopts the installation method that the permanent magnet is adsorbed on the surface of the OLTC test point of the on-load tap-changer. This installation method is simple and easy, and is suitable for occasions where the test point is frequently replaced. Considering the propagation medium of the vibration signal and the damping of the propagation process, the present invention installs the vibration sensor on the top of the tap changer. The high-frequency attenuation of the vibration signal picked up at this position is relatively small and the signal is relatively complete; the collected vibration data is divided into Two groups, each group includes normal vibration signals and fault signals, one group is used for SVM training, and the other group is used for SVM testing.

The fault state in step 1) refers to the two situations of loose contact of on-load tap-changer OLTC and decline of spring performance;

In step 1), the vibration signal is preprocessed, specifically, noise reduction processing is performed on the vibration signal.

In step 2), multi-scale permutation entropy analysis is performed on the collected vibration signals to construct feature vectors. Multiscale permutation entropy is to calculate the permutation entropy of time series at multiple scales.

The specific algorithm of permutation entropy is:

Assuming a set of time series {xi | _i =1,2,…,N}, perform phase space reconstruction on them to obtain the reconstructed time series X _i :

X _i ＝[x _i ,x _i+τ ,…,x _i+(m-1)τ ] (1)

Among them, m is the embedding dimension, τ is the delay time, N is the length of the time series, x _i is the i-th data point in the time series X _i , x _i+τ is the i+τ number in the time series X _i , x _i+(m-1)τ is the number i+(m-1)τ in the time series X _i , and the m data points in the time series X _i are arranged in ascending order, that is r represents the position of the data point in the reconstructed sequence _Xi , It is the second number after the data points in the reconstructed sequence _Xi are arranged in ascending order, similarly is the mth data point after the data points in the reconstructed sequence _Xi are arranged in ascending order; when there is when the data point Arrange according to the size of r _j and r _k , that is, if r _j < r _k , then consider Represents the r _jth data point in the reconstructed sequence _Xi , r _j and r _k represent the location of the data point in the reconstructed sequence _Xi , Represents the r _kth data point in the reconstructed sequence _Xi .

Any time series X _i has m! Arrangement, for any arrangement ω, T(ω) represents the number of times it appears, then the probability of its occurrence is:

Therefore, the permutation entropy _HPE of time series _Xi can be defined as:

H _PE ＝-∑P(ω)lnP(ω) (3)

After normalization, the normalized permutation entropy PE is obtained:

The size of the PE value reflects the complexity and randomness of the time series signal. The larger the value, the more complex the time series signal, and vice versa, the more regular it is. Therefore, the transformation of the PE value reflects and amplifies the local subtle transformation of the time series.

As shown in Figure 1, the specific steps of the multiscale permutation entropy (MPE) analysis method are as follows:

1) Coarse-grain the original time series {x ₁ ,x ₂ ,…,x _N }, and construct a multi-scale time series {y _l ^(s) } according to the following formula:

In the formula, s is the scale factor, N is the length of the original time series, _xi represents the data points in the original time series, and y _l ^(s) represents the lth average value of the time series under the s window.

2) After coarse-graining the original time series, calculate the normalized permutation entropy of each coarse-grained time series according to formulas (1)-(4), that is, obtain the multi-scale permutation entropy, which is the feature vector.

In step 3), first use the data processing-normalization function mapminmax that comes with matlab to normalize the feature vector to [0, 1], and input the normalized training feature vector into the SVM. The SVM is trained, and then the normalized test feature vector is input to the trained SVM to judge the failure mode of the OLTC.

Example:

Simulation experiment on CMIII-500-63B-10193W tap changer. This tap changer is a three-phase Y connection with a maximum of 19 tap positions. The vibration sensor adopts the LC0151 piezoelectric acceleration sensor with high resolution and strong anti-interference ability. In the present invention, the vibration sensor is installed on the top of the tap changer, the vibration signal picked up at this position has less high-frequency attenuation and the signal is more complete. The acceleration sensor collects the vibration signal under the normal state of the OLTC and the vibration signal under the fault state (OLTC contact looseness and spring performance degradation), and divides the collected vibration data into two groups, each group includes normal vibration signals and faults Signals, one set is used for SVM training and the other set is used for SVM testing;

MPE analysis is carried out on the collected vibration signal, and the feature vector is constructed, which is used as the input of SVM, and the SVM is trained, and the test data is input into the trained SVM, so as to judge the failure mode of the OLTC.

First use the mapminmax that comes with matlab to normalize the feature quantity between [0, 1], input the normalized training feature quantity into the SVM, train the SVM, and then normalize the test feature quantity The amount is input to the trained SVM to judge the failure mode of OLTC.

The influencing factors of multi-scale permutation entropy are: when using multi-scale permutation entropy for feature extraction, the embedding dimension m, delay time τ and scale factor s will directly affect the calculation results: if m is too small, the state contained in the reconstructed vector If the amount is small, the algorithm loses its meaning. If the value of m is too large, the reconstruction of the phase space will homogenize the time series. At this time, not only the calculation time is long, but also the subtle transformation of the sequence cannot be reflected. Bandt suggested that the embedding dimension m be taken as 3 to 7, this embodiment takes 7; the delay time τ has little influence on the calculation of the sequence, this embodiment selects τ=1; and when the scale factor s>10, the dynamic change law of the system at different scales will be shown , s is 12 in this embodiment.

Figure 2 shows the multi-scale permutation entropy under the three states. It can be seen from figure 2 that the normalized permutation entropy PE of the tap changer in the normal state is obviously higher than that in the fault state, while the contact looseness and the spring performance decrease. The difference in PE between the two types of mechanical faults is small after s is 8, which shows that it is appropriate to take 12 as the maximum value of s, which fully expresses the dynamic law of the signal. In addition, when OLTC has a mechanical failure, the randomness of the vibration signal is smaller and the complexity is lower, and the PE value of the vibration signal is smaller at this time; conversely, when the OLTC is in a normal state, the randomness of the vibration signal is the largest, and the PE value is also smaller. maximum. This shows that the change of PE value can well reflect the degree of mechanical failure of OLTC.

Comparing the diagnostic effect of SVM based on MPE feature quantity on OLTC with BP neural network, the prediction comparison results shown in Table 1 are obtained:

Table 1 The present invention compares with BP neural network prediction result

Note: 1 means that the tap changer is in a normal state, 2 means that the tap changer is in a loose contact fault, and 3 means that the tap changer is in a spring performance degradation fault.

The same 15 sets of data are used to train the BP neural network, and the 15 sets of data are predicted. It can be seen from the table that the OLTC mechanical fault diagnosis method based on MPE and SVM can effectively identify the fault type, thus illustrating the The feasibility of the present invention is also known from Table 1, and the diagnostic effect of the present invention is better than that of the BP neural network.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (7)

1. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM, it is characterised in that：Include the following steps：

1) by acceleration transducer to the vibration signal under load ratio bridging switch OLTC normal conditions, the vibration under malfunction Signal is acquired, and is pre-processed to vibration signal；

2) it carries out multiple dimensioned arrangement entropy MPE to collected vibration signal to calculate, construction feature vector, as support vector machines The input of SVM；

3) feature vector that step 2) obtains is input in support vector machines, support vector machines is trained, it will Test data is input to trained SVM, to judge the fault mode of OLTC.

2. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 1, it is characterized in that：It is described Pretreatment is specially：Noise reduction process is carried out to collected vibration signal.

3. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 1, it is characterized in that：It is described Malfunction includes that load ratio bridging switch OLTC contact slaps and spring performance decline.

4. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 1, it is characterized in that：It is described Acceleration transducer is mounted on the top of tap switch.

5. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 1, it is characterized in that：It is described Multiple dimensioned arrangement entropy MPE calculate the specific steps are：

1) to original time series { x₁,x₂,…,x_NCoarse processing is carried out, multiple dimensioned time series { y is constructed according to the following formula_l ^(s)}：

In formula, s is scale factor, and N is the length of original time series, x_iIndicate the data point in original time series, the model of i It encloses for 1-N, y_l ^(s)Indicate first of average value of time series under s windows；

2) after to the processing of original time series coarse, calculate the arrangement entropy after the normalization of each coarseness time series to get Multiple dimensioned arrangement entropy, constitutive characteristic amount are arrived.

6. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 5, it is characterized in that：It is described Arrangement entropy after normalization specifically calculates step：

Assuming that one group of time series { x_i| i=1,2 ..., N }, phase space reconfiguration is carried out to it, the time series X reconstructed_i：

X_i=[x_i,x_i+τ,…,x_i+(m-1)τ] (1)

Wherein, m is Embedded dimensions, and τ is delay time, x_iFor time series X_iIn i-th of data point, x_i+τFor time series X_iIn I-th+τ data points, x_i+(m-1)τFor time series X_iIn i-th+(m-1) τ data point, by time series X_iIn m data Ascending order arrangement is pressed, i.e.,R indicates reproducing sequence X_iThe position of middle data point,For reconstruct Sequence X_iMiddle data point is by the 2nd data point after ascending order arrangement, similarlyTo reconstruct sequence X_iMiddle data point is arranged by ascending order M-th strong point afterwards；Work as presenceWhen, data pointBy r_j、r_kSize arranged, even r_j<r_k, then Think Indicate reproducing sequence X_iIn r_jA data point, r_j、r_kIndicate reproducing sequence X_iMiddle data point institute is in place It sets,Indicate reproducing sequence X_iIn r_kA data point；

Time series X_iThere is m！Middle arrangement mode, to any arrangement mode ω, T (ω) indicates its number occurred, then it occurs Probability be：

Therefore, time series X_iArrangement entropy H_PEIt may be defined as：

H_PE=-∑ P (ω) ln P (ω) (3)

Arrangement entropy PE after being normalized after normalization：

7. a kind of OLTC mechanical failure diagnostic methods based on MPE and SVM according to claim 1, it is characterized in that：It is described Vibration data is divided into two groups, and every group includes normal vibration signal and fault-signal, and one group of training for being used for SVM, another group is used for The test of SVM；In step 3), first by feature vector normalized between [0,1], by the training feature vector after normalization It is input in support vector machines, support vector machines is trained, be then input to the testing feature vector after normalization Trained SVM, to judge the fault mode of OLTC.