CN111523253B - A Method of Determining the Pull-in Time Threshold of Railway Relay Based on Load Fusion Data - Google Patents

Abstract

The invention provides a method for determining a suction time threshold of a railway relay based on load fusion data. The method for determining the suction time threshold of the railway relay based on the load fusion data comprises the following steps: and 6 kinds of relevant parameter data of all samples of the railway relay in the accelerated life test are arranged, the dimension reduction processing is carried out on the initial data after the initial data are arranged, the fusion processing is carried out on the parameter threshold values, the mathematical model of the life prediction of the railway relay is determined, the fusion result is input into the mathematical model, the time when the result reaches the failure threshold value is the life of the railway relay, the suction time is input into the life prediction mathematical model for training, and the corresponding time when the result reaches the life of the railway relay is the suction time threshold value. The accuracy of the life prediction of the railway relay is improved by utilizing the load fusion data, and a good basis is provided for determining the suction time threshold value of the railway relay.

Description

A method for determining the closing time threshold of railway relays based on load fusion data

Technical Field

The invention relates to the technical field of reliability, and in particular to a method for determining a railway relay closing time threshold based on load fusion data.

Background Art

Railway relay is a very important basic component in the railway signal equipment system and an irreplaceable part. The reliability of its operation also affects the safe operation of the entire railway system. Railway relays monitor various parameters during operation. Among them, there is no accurate fixed value for the threshold of the railway relay's pull-in time. The determination of the pull-in time threshold is a very important basis for judging whether the railway relay has failed.

Traditional methods for predicting the life of railway relays are mostly to achieve the purpose of predicting the life of railway relays through the analysis and prediction of a single parameter. Studies have shown that the single parameter life prediction evaluation of relays is not accurate enough. In order to overcome this shortcoming, the present invention has developed a method of load fusion data to predict the life of railway relays, which is more accurate than traditional methods.

In the literature (Miao Jianwei, Wang Wenjun, Li Bin. Intelligent prediction analysis of low-voltage relay life [J]. Electrical Appliances and Energy Efficiency Management Technology, 2018 (04): 61-65.), a method for testing relay life was designed with the pull-in time as the main characteristic parameter. The BP neural network algorithm and the gray theory algorithm were selected to predict and analyze the relay life. The prediction performance of the two intelligent algorithms was evaluated through the prediction results. The prediction ideas are shown in Figure 2. The performance of the relay itself is complex. Obviously, the method of predicting and analyzing the relay life using only one parameter is not perfect, and the prediction results may have large errors.

A single parameter is not comprehensive enough for predicting the life of a railway relay. In order to determine the threshold of the relay's closing time, the present invention provides a method for determining the threshold of the railway relay's closing time based on load fusion data.

Summary of the invention

During the accelerated life test, the parameters monitored after every 200,000 experiments include the make contact pressure, break contact pressure, absolute gap, pull-in voltage, release voltage and contact resistance parameters. The changes in each parameter reflect the performance parameter degradation process of the relay.

When analyzing the degradation performance of railway relays, we should not only focus on the change of one parameter, but also comprehensively consider all the parameters that affect the life of railway relays. There are many parameters that characterize the performance of relays, and only using a certain parameter to predict the life of the relay is not accurate enough and cannot comprehensively evaluate the life of the relay. The closing time threshold of the railway relay is difficult to determine in actual use, and the closing time is an effective basis for detecting whether the railway relay has failed.

In order to solve the above problems and achieve the prediction goal, the present invention provides a method for determining the railway relay pick-up time threshold based on load fusion data. The method for determining the railway relay pick-up time threshold based on load fusion data is as follows:

Firstly, the data of six relevant parameters of railway relays in accelerated life test are sorted out.

The parameters are dimensionally reduced and fused, and the thresholds of the parameters are fused. First, the original data is dimensionless to obtain standardized data:

For each sequence X:

Among them, m is 6 parameters, and n is the amount of data selected for each parameter;

Standardization formula (for the first column), add 1 to i from 1 to n, and get the sequence y _i1 :

in:

Similarly, y _i2 , y _i3 , ...y _im are calculated;

Finally, the transformed sequence Y is obtained:

The covariance matrix of the obtained new matrix is obtained, and the covariance cov(x, y) is defined as follows:

cov(x, y)=E[(x-μ _x )(y-μ _y )]

Among them, E(x)=μ _x , E(y)=μ _y ;

得协方差矩阵为：Therefore, for the matrix

The covariance matrix is:

According to |Y-λE|=0, the eigenvalues λ _i (i=1, 2, ..., m) and eigenvectors e _i (i=1, 2, ..., m) of the covariance matrix are obtained, and the eigenvectors are arranged in order of eigenvalue size.

The proportion of each element and the cumulative proportion are obtained through the characteristic value:

Cumulative percentage:

The next step of calculation is performed by selecting elements whose cumulative proportion is greater than 90%.

The benefit of performing dimensionality reduction processing and re-integration on the original data is that the above process can effectively reduce the dimensionality of the original data and lose as little information of the original data as possible.

According to the above calculation results, calculate the element load:

Among them, e _ij is the value in the eigenvector.

Its load matrix is:

The load matrix represents the load of each parameter in the corresponding element. The original data is fused by the obtained load matrix and the proportion of each element. The number of elements obtained after dimensionality reduction is the same as the number of original parameter categories, that is, m elements. Assuming that the cumulative proportion of the first p (p < m) elements is greater than or equal to 90%, the fusion process is as follows.

The load of each original parameter in the corresponding element is multiplied by the original data, and then added together to get a new matrix as follows:

The proportion of each element is different, and the above fused matrix can be fused again. The proportion of each element's corresponding proportion λ _i (i＜p) to the total proportion of the selected elements lgx _p is multiplied by the corresponding fusion result rh _i , that is:

The selected parameters have fixed thresholds, which are processed as described above to be fused into rhyz ₁ , rhyz ₂ , …, rhyz _p , and finally fused into zrhyz.

The process of establishing the life prediction model of railway relay is as follows:

未知；S1: Generate ^x1 = { ^x0 (1), ^x0 (2), ... ^x0 (n)} by accumulating the original data sequence ^x0 = { ^x0 (1), ^x0 (2), ... ^x0 (n)}. Assume that the generated sequence ^x1 satisfies the first-order ordinary differential equation

unknown;

S2: Create a matrix B, Y _n , Y _n = BU, where:

S3: Find the inverse matrix (B ^T B) ^-1 ;

求估计值

和

S4: According to

Find the estimated value

and

S5: Substitute the obtained result into the equation:

是拟合值；When k = 1, 2, ..., n-1, the

is the fitted value;

为预报值。When k ≥ n,

is the predicted value.

S6: Accuracy test and prediction, input the above fusion results into the prediction model for prediction, and when the result reaches the set threshold, it is judged that the railway relay has reached its service life. When the error is between 0.8 and 0.95 and the ratio is between 0.35 and 0.45, the prediction result is qualified.

The life of the railway relay obtained by the data fusion prediction is input into the prediction model of the existing railway relay's pull-in time. When the life of the railway relay reaches failure, the corresponding data is the threshold of the pull-in time.

As described above, the advantages of the present invention are:

(1) The present invention breaks the traditional method of using a single parameter to predict the life of a relay and proposes a method for predicting the life of a relay based on load fusion data.

(2) The method used in the present invention of first reducing the dimension of the original data and then fusing it can effectively reduce the dimension of the original data while retaining most of the original information of the original data.

(3) The present invention considers the load of each parameter in a certain element during the fusion process, and considers the weight of a certain element in all selected elements, thereby improving the credibility.

(4) The present invention utilizes the parameters that affect the life of the railway relay for data fusion and then predicts its life, comprehensively considering the influence of all parameters, and after obtaining the predicted life of the railway relay, the life prediction model is used again to determine the threshold value of the railway relay's pull-in time.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of a structure for determining a railway relay closure time threshold in the prior art.

Figure 2 shows the prediction process of relay life based on the pull-in time.

FIG3 is a schematic diagram of the structure of the life prediction model of the present invention.

Fig. 4 Railway relay life prediction curve.

Figure 5 Railway relay closing time prediction curve.

DETAILED DESCRIPTION

The specific implementation process of the present invention is given below.

During the test of the railway relay from 0 to 2 million times under constant temperature, the parameters were measured every 200,000 times, for a total of 11 data. The six parameters of one railway relay, namely, the contact pressure of the moving contact, the contact pressure of the breaking contact, the absolute gap, the pull-in voltage, the release voltage and the contact resistance, were sorted out. The original data was standardized to obtain the standardized data:

Table 1 Data table after standardization of each parameter

-0.6444 1.6370 1.2103 -1.7566 -0.4567 -0.7910 1.5366 0.4753 0.4624 -1.6503 0.0864 -0.4691 0.9914 -0.1056 2.4070 0.3707 0.6294 -0.7584 1.5366 1.0561 -0.4352 -0.8171 0.901 -0.8307 -1.1896 1.0561 -0.7343 -0.3384 0.7652 -0.5064 0.4461 0.4753 -1.0335 0.4770 0.5616 -0.4084 -0.6444 -1.2674 -0.2856 0.9557 1.444 -0.8983 -0.0991 -0.6865 -0.4352 1.0266 -0.1173 0.2074 -0.6444 -0.6865 -0.4352 0.6720 -0.5924 1.4718 -0.0991 -0.6865 -0.2856 0.3529 -1.4749 1.4858 -1.1896 -1.2674 -0.4352 0.7075 -1.7464 1.4974

The covariance matrix of the standardized data matrix is shown in Table 2:

Table 2 Covariance matrix of standardized data

1 0.3052 0.3114 -0.3704 0.3462 -0.4138 0.3052 1 0.1927 -0.8209 0.3137 0.5976 0.3114 0.1927 1 -0.3307 0.0546 -0.3394 -0.3704 -0.8209 -0.3307 1 -0.0729 0.4465 0.3462 0.3137 0.0546 -0.0729 1 -0.8522 -0.4138 0.5976 -0.3394 0.4465 -0.8522 1

According to |Y-λE|=0, the eigenvalues and eigenvectors of the covariance matrix are obtained, and the vectors are arranged in the order of the eigenvalues.

The eigenvalues λ _i (i=1, 2, ..., 6) are λ ₁ =3.0102, λ ₂ =1.2307, λ ₃ =0.9312, λ ₄ =0.6516, λ ₅ =0.1292, λ ₆ =0.0470 from large to small, and the corresponding eigenvectors are:

Table 3 The corresponding eigenvectors of each eigenvalue arranged from large to small are the elements obtained:

Element One Element Two Element Three Element Four Element Five Element Six -1.8355 -1.7725 -0.5530 -1.2793 -0.0189 -0.1613 -1.8744 -0.8838 0.3435 0.9675 -0.6295 0.0829 -1.4148 0.2063 2.2951 -0.6403 0.4122 0.0668 -2.0381 0.3210 -0.4245 1.1789 0.0882 0.0235 -0.5557 0.4134 -1.6621 -0.8455 0.1703 0.2005 -0.3204 0.9431 -0.7117 0.6798 0.4373 -0.2253 0.3070 2.1976 0.2272 -0.7345 -0.6274 -0.0015 1.0592 0.7021 0.2153 0.1794 0.1627 -0.2474 1.9275 -0.2400 -0.0450 0.1102 0.1567 0.4828 1.8919 -1.0372 0.2613 0.5847 0.0685 -0.0037 2.8534 -0.8500 0.0539 -0.2008 -0.2202 -0.2173

依次gx₁＝50.1703％，gx₂＝20.5123％，gx₃＝15.5194％，gx₄＝10.8597，gx₅＝2.1541％，gx₅＝0.7841％。Calculate the proportion of each element through the obtained eigenvalue

In order, gx ₁ =50.1703%, gx ₂ =20.5123%, gx ₃ =15.5194%, gx ₄ =10.8597, gx ₅ =2.1541%, and gx ₅ =0.7841%.

依次为lgx₁＝50.1703％，lgx₂＝70.8626％，lgx₃＝86.202％，lgx₄＝97.0618％，lgx₅＝99.216％，lgx₆＝100％。又lgx₄＞90％，因而在后续的计算中，所选择的元素为元素一到元素四。这种选择方法可以有效的降低原始数据维度，同时保留了原始数据的大部分信息。After obtaining the proportion of each element, the cumulative proportion of the elements is obtained by adding them up one by one.

They are lgx ₁ = 50.1703%, lgx ₂ = 70.8626%, lgx ₃ = 86.202%, lgx ₄ = 97.0618%, lgx ₅ = 99.216%, and lgx ₆ = 100%. And lgx ₄ > 90%, so in the subsequent calculations, the selected elements are elements 1 to 4. This selection method can effectively reduce the dimension of the original data while retaining most of the information of the original data.

得载荷矩阵：After obtaining these data, the formula

The load matrix is obtained:

Table 4 Loading matrix

Element One Element Two Element Three Element Four Element Five Element Six Close contact pressure -0.3601 0.0157 0.4229 0.8230 0.0958 -0.0686 Break contact pressure -0.4648 -0.2789 -0.4488 -0.0470 0.7081 0.0343 Absolute clearance -0.2670 -0.2956 0.7429 -0.4962 0.1387 0.1552 Pull-in voltage 0.4271 0.5191 0.2538 -0.0490 0.6537 -0.2324 Release voltage -0.3738 0.6700 -0.0512 -0.0916 -0.0506 0.6307 Contact resistance 0.5107 -0.3410 0.0284 0.2520 0.2008 0.7199

According to the above data, the original data is fused and processed. The original data is fused in the following way. The fusion results are shown in Table 5:

rh ₁ ＝I ₁₁ *X ₁ +I ₂₁ *X ₂ +…+I ₆₁ *X ₆

rh ₂ ＝I ₁₂ *X ₁ +I ₂₂ *X ₂ +…+I ₆₂ *X ₆

rh ₃ ＝I ₁₃ *X ₁ +I ₂₃ *X ₂ +…+I ₆₃ *X ₆

rh ₄ ＝I ₁₄ *X ₁ +I ₂₄ *X ₂ +…+I ₆₄ *X ₆

Table 5 Fusion results

rh _{1 rh2 rh3} rh ₄ -227.2213 -57.1155 31.1934 260.4732 -229.0615 -54.3835 44.1525 277.7655 -225.1479 -52.0378 44.6286 273.4453 -232.0051 -54.9074 41.9331 277.1356 -222.1646 -55.6412 31.4548 256.9094 -224.8146 -53.9298 40.1576 269.5854 -215.2607 -48.9747 42.7000 261.4813 -216.8588 -52.0375 42.7196 266.5798 -212.3495 -54.1149 40.7119 263.8469 -214.1503 -54.2404 42.7965 267.9928 -208.0968 -52.9399 40.8583 260.0092

The proportion of each element is known, and then the proportion of each element's corresponding proportion λ _i (i＜p) to the total proportion of the selected elements lg× ₄ is multiplied by the corresponding fusion result. The final fusion result matrix is shown in Table 6:

Table 6 The final fusion matrix

z -95.3947 -91.7615 -89.6501 -93.8193 -92.8262 -91.0245 -85.5386 -86.4386 -85.1736 -85.3338 -83.1329

According to the obtained element proportion and load matrix, the thresholds of various parameters are fused, and the results after threshold fusion are shown in Table 7:

Table 7 Results after threshold fusion

rhyz ₁ rhyz ₂ rhyz ₃ rhyz ₄ -128.4512 -44.0991 44.2942 209.9117

The above results are combined with the proportion of each element and then merged to obtain zrhyz=-45.1467.

According to the fusion results of the original data obtained above, the fusion results are input into the mathematical model for prediction, and the number of predictions is modified. When the prediction results reach the threshold fusion result, the railway relay is judged to be failed. The prediction curve is shown in Figure 2. Since the parameters used are every 200,000 test data, the corresponding parameter when the horizontal axis is 62 is the predicted life of the railway relay. The number of times corresponding to the predicted result multiplied by 200,000 is the predicted life of the railway relay. The predicted life obtained from the actual data is 12.2 million times.

The test error is 0.8＜0.8182＜0.95, the ratio is 0.35＜0.4231＜0.45, and the test result is qualified.

According to the prediction results, the pull-in time measured during the test is input into the railway relay life prediction model, and the prediction times are input. The value corresponding to the prediction times is the threshold of the pull-in time. The error is 0.8＜0.9052＜0.95, the ratio is 0.35＜0.4138＜0.45, and the test result is qualified. The prediction curve is shown in Figure 3, and the threshold of the pull-in time obtained by the mathematical model is shown in Table 8:

Table 8 Railway relay life and closing time threshold obtained by mathematical model

Predicted life expectancy Pull-on time threshold 12.2 million times 227.71ms

The above results show that the present invention can effectively predict the closing time threshold of railway relays.

Claims (5)

1. A method for determining a railway relay pick-up time threshold based on load fusion data, characterized in that the method for determining a railway relay pick-up time threshold based on load fusion data comprises the following steps: S1: Organize the data of 6 relevant parameters of a specific type of relay in the accelerated life test; S2: Reduce the dimension of the parameters and fuse them, and fuse the thresholds of the parameters at the same time; first, perform dimensionless processing on the original data to obtain standardized data: For each sequence X:

Among them, m is 6 parameters, and n is the amount of data selected for each parameter; Finally, the transformed sequence y is obtained:

S3: Calculate the covariance matrix of the obtained new matrix, and simultaneously calculate the eigenvalue λ _i and eigenvector e _i of the covariance matrix, where i = 1, 2, …, m, and arrange the eigenvectors according to the size of the eigenvalue; Calculate the percentage and cumulative percentage of each element, and select the elements with cumulative percentage greater than 90% for the next step of calculation; use

Calculate the element loading matrix:

The loading matrix represents the loading amount of each parameter on its corresponding element; Perform the added fusion according to the obtained results; S4: The number of elements obtained after dimensionality reduction is the same as the number of original parameter categories, that is, m elements. Assuming that the cumulative proportion of the first p elements is greater than or equal to 90%, where p < m, the fusion process is: the load of each original parameter in the corresponding element is multiplied by the original data, and then added to obtain a new matrix; S5: The proportion of each element is different, and the fused matrix can be fused again. The proportion of the proportion λ _i corresponding to each element to the total proportion lgx _p of the selected elements is multiplied by the corresponding fusion result rhi, and then added to obtain zrh, where i＜p; The selected parameters have fixed thresholds, which are fused into rhyz1, rhyz2, ..., rhyzp, and finally into zrhyz; S6: Determine the mathematical model for predicting the life of railway relays. The modeling process is as follows: The original data sequence X ⁰ = {X ⁰ (1), X ⁰ (2), ... X ⁰ (n)} is accumulated to generate X ¹ = {X ⁰ (1), X ⁰ (2), ... X ⁰ (n)}. Assume that the generated sequence X ⁽¹⁾ satisfies the first-order ordinary differential equation

a, u are unknown; Create a matrix B, Y _n , Y _n = BU, where:

Find the inverse matrix (B ^T B) ^-1 ; according to

Find the estimated value

and

Substituting the result into the equation

When k = 1, 2, ..., n-1, the

is the fitted value; When k ≥ n,

is the forecast value; Accuracy test and prediction: input the fusion results into the prediction model for prediction. When the prediction results reach the set threshold, it is judged that the railway relay has reached its service life. When the error is between 0.8 and 0.95 and the ratio is between 0.35 and 0.45, the prediction result is qualified. S7: Input the existing test data of the pull-in time into the prediction model. When it reaches the service life of the railway relay, the corresponding time is the threshold of the pull-in time. 2. A method for determining a railway relay pick-up time threshold based on load fusion data according to claim 1, characterized in that the standard transformation formula in step S2 is:

in:

3. A method for determining a railway relay pick-up time threshold based on load fusion data according to claim 1, characterized in that the calculation formula for obtaining the proportion of each element and the cumulative proportion by the characteristic value in step S3 is:

Cumulative percentage:

4. A method for determining a railway relay pick-up time threshold based on load fusion data according to claim 1, characterized in that the fusion formula in step S4 is: rh ₁ ＝I ₁₁ *X ₁ +I ₂₁ *X ₂ +…+I _m1 *X _m rh ₂ ＝I ₁₂ *X ₁ +I ₂₂ *X ₂ +…+I _m2 *X _m

rh _p =I _1p *X ₁ +I _2p *X ₂ +…+I _mp *X _m . 5. The method for determining the railway relay closing time threshold based on load fusion data according to claim 1, characterized in that the re-fusion formula in step S5 is:

Images