A method for transformer fault early warning based on DGA

Abstract

The application discloses a kind of transformer fault early warning method based on DGA, comprising: step S1. the historical concentration data group of dissolved gas in transformer oil is collected to form concentration sequence;Step S2. the gas concentration of the last i+1 moment is selected, and i polynomial is used to fit, and fitting curve p is obtained;Step S3. the gas concentration of the last i+2 moment is selected, and i+1 polynomial is used to fit, and fitting curve q is obtained;Step S4. the similarity of fitting curve q and fitting curve p is calculated;Step S5. repeat step S3 and step S4, until the first decline appears in fitting curve similarity;Step S6. with the m-1 fitting curve, the dissolved gas in oil is predicted, and fault early warning is carried out.The application uses multivariate nonlinear regression to carry out the fitting of curve, uses curve similarity as criterion, can determine the highest order of final fitting equation faster and more flexibly, and can guarantee the accuracy of curve prediction.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault early warning, and in particular to a method for transformer fault early warning based on DGA. Background Technology

[0002] Transformers are the second most important equipment in a power system after generators. Their safe and stable operation is crucial to the stable development of the national economy, and their significance is self-evident. Therefore, research on transformers is essential. If we can predict the types, times, and locations of potential transformer failures as early and accurately as possible, we can prevent and avoid major risks and accidents.

[0003] Traditional fault early warning methods mainly rely on the deployment of online monitoring devices or other operational information to obtain multi-dimensional data for fault early warning. However, in reality, data acquisition is difficult, requiring a lot of manpower and resources, and integration is quite troublesome. In addition, most early warning methods use a large amount of data for regression prediction, which has poor adaptability and the accuracy will gradually shift, adversely affecting the early warning.

[0004] Therefore, this patent uses historical data of dissolved gases in oil as much as possible and adaptively selects a limited number of data points, fits curves, and uses the concept of similarity to achieve fault warning at a high confidence level. Summary of the Invention

[0005] Based on the aforementioned technical problems, the purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for transformer fault early warning based on DGA. This invention utilizes the concept of curve similarity to reasonably fit gas sequences, thereby laying the foundation for fault early warning.

[0006] To solve the above-mentioned computational problems, the present invention adopts the following technical solution:

[0007] A method for transformer fault early warning based on DGA includes the following steps:

[0008] Step S1. Collect historical concentration data of dissolved gases in transformer oil based on DGA technology to form a concentration sequence;

[0009] Step S2. Select the gas concentration at the most recent i+1 time, and use an i-th degree polynomial to fit the curve to obtain the fitted curve p;

[0010] Step S3. Select the gas concentrations at the most recent i+2 time points and use an i+1 degree polynomial to fit the curve q.

[0011] Step S4. Calculate the similarity between the fitted curve q and the fitted curve p;

[0012] Step S5. Repeat steps S3 and S4 until the similarity between the m-th fitted curve and the (m-1)-th fitted curve first decreases;

[0013] Step S6. Use the (m-1)th fitted curve to predict dissolved gases in the oil and provide fault warnings.

[0014] Furthermore, the historical concentration data of dissolved gases in step S1 refers to the concentrations of H2, CH4, C2H2, C2H4, and C2H6 gases dissolved in transformer oil, and the gas concentrations are obtained by taking oil samples online using an online oil chromatography monitoring device.

[0015] Further, step S2 includes the following sub-steps:

[0016] Step 2.1: Set the i+1 time points to -i, -i+1, -i+2, ..., -2, -1, 0 respectively;

[0017] Step 2.2: The gas concentration at time i+1 is fitted using an i-th degree polynomial, and the following system of equations is established:

[0018]

[0019] In the formula, X, B, and Y are the independent variable matrix, coefficient matrix, and dependent variable matrix, respectively, and n is the number of polynomial terms; x -i For time -i, β0, β1, β2, ..., β n y represents the coefficients of the constant term, linear term, quadratic term, ..., nth degree term of the polynomial, respectively. -i Let B be the gas concentration at time -i; B = (X T X) -1 X T Y, thus the fitted curve p can be obtained.

[0020] Further, step S3 includes:

[0021] Step 3.1: Let i = i+1, and use a polynomial of degree i+1 to fit the curve q.

[0022] Further, step S4 includes the following sub-steps:

[0023] Step 4.1: Define the curve coefficients of the fitted curves p and q as follows:

[0024]

[0025] Where G = {y p -i-1 y p -i y p-i+1 ...,y p 0} represents the i-th order polynomial fitting of the gas concentration sampling values ​​at times -i, -i+1, ..., 0, and the resulting fitted curve p predicts the gas concentration y at times -i-1, -i, -i+1, ..., 0. p - i-1 y p - i y p - i+1 ... y p 0, H = {y q -i-1 y q -i y q -i+1 ... y0 q} represents the (i+1)th degree polynomial fitting of the gas concentration sampling values ​​at times -i-1, -i, -i+1, ..., 0, and the resulting fitted curve q predicts the gas concentration y at times -i-1, -i, -i+1, ..., 0. q -i-1 y q -i y q -i+1 ... y0 q μ p μ q These are the means of G and H, respectively;

[0026] Step 4.2: Define the similarity between the fitted curve p and the fitted curve q, specifically as follows:

[0027]

[0028] Further, step S5 includes:

[0029] Step 5.1: Initialize S0 = 0, and based on step 2.2, obtain the fitted curve p;

[0030] Step 5.2: Let i = i + 1, and based on step 2.2, obtain the fitted curve q;

[0031] Step 5.3: Combine step S4 to calculate the similarity S between the fitted curve p and the fitted curve q, i.e., between sequence G and sequence H. Determine whether S has decreased compared to S0. If it has decreased, select the fitting equation of the fitted curve p as the equation to be predicted; otherwise, set S0 = S, rename the fitted curve q to the fitted curve p, and repeat steps 5.1 to 5.3.

[0032] Further, step S6 specifically includes:

[0033] The concentration prediction value of dissolved gas in oil is obtained by predicting the concentration of dissolved gas in oil using the equation to be predicted. Based on the concentration prediction value of dissolved gas in oil, the three-ratio method or the four-ratio method is used for fault diagnosis and fault early warning.

[0034] Compared with the prior art, this application has the following beneficial effects:

[0035] 1. Previous methods have focused on fitting and predicting gas concentrations with a large amount of data. However, this ignores the changes in the gas concentration pattern, resulting in a large error in the predicted gas concentration of the fitted equation. The method proposed in this application does not require a large amount of data and can better adaptively consider the changes in the gas concentration pattern, thus having higher accuracy.

[0036] 2. Traditional methods require repeated attempts to obtain a better fit for the polynomial, and once determined, the fitting formula is rarely improved. However, the method proposed in this application can flexibly adjust the highest order of the polynomial fitting function. In addition, since a large number of blind trials are avoided, this method is faster. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the method flow of this invention. Detailed Implementation

[0038] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, embodiments and comparative examples.

[0039] Example

[0040] This embodiment provides a method for transformer fault early warning based on DGA, including the following steps:

[0041] Step S1. Collect historical concentration data of dissolved gases in transformer oil based on DGA technology to form a concentration sequence;

[0042] Step S2. Select the gas concentration at the most recent i+1 time, and use an i-th degree polynomial to fit the curve to obtain the fitted curve p;

[0043] Step S3. Select the gas concentrations at the most recent i+2 time points and use an i+1 degree polynomial to fit the curve q.

[0044] Step S4. Calculate the similarity between the fitted curve q and the fitted curve p;

[0045] Step S5. Repeat steps S3 and S4 until the similarity between the m-th fitted curve and the (m-1)-th fitted curve first decreases;

[0046] Step S6. Use the (m-1)th fitted curve to predict dissolved gases in the oil and provide fault warnings.

[0047] In the above embodiments, the dissolved gases in the transformer oil in step S1 mainly include H2, CH4, C2H2, C2H4 and C2H6, which can be obtained by taking oil samples online using an online oil chromatography monitoring device.

[0048] For any gas, perform the following operation:

[0049] Step 2.1: Set the i+1 time points to -i, -i+1, -i+2, ..., -2, -1, 0 respectively;

[0050] Step 2.2: The gas concentration at time i+1 is fitted using an i-th degree polynomial, and the following system of equations is established:

[0051]

[0052] In the formula, X, B, and Y are the independent variable matrix, coefficient matrix, and dependent variable matrix, respectively, and n is the number of polynomial terms; x -i For time -i, β0, β1, β2, ..., β n y represents the coefficients of the constant term, linear term, quadratic term, ..., nth degree term of the polynomial, respectively. -i Let B be the gas concentration at time -i; B = (X T X) -1 X T Y, thus the fitted curve p can be obtained.

[0053] Further, step S3 includes:

[0054] Step 3.1: Let i = i+1, and use a polynomial of degree i+1 to fit the curve q.

[0055] When fitting curve p, in this embodiment we let i = 2, then the system of equations for fitting curve p is as follows:

[0056]

[0057] In the formula, Let be the independent variable matrix, coefficient matrix, and dependent variable matrix when i = 2, respectively, and n be the number of polynomial terms. Calculate B = (X... T X) -1 X T Y, thus the fitted curve p can be obtained.

[0058] When fitting curve q, in this embodiment, we let i = 3. Then, the system of equations for fitting curve q is as follows:

[0059]

[0060] In the formula, Let n be the independent variable matrix, coefficient matrix, and dependent variable matrix when i = 3, and n be the number of polynomial terms. Calculate B = (X... T X) -1 X T Y, thus obtaining a new fitted curve q.

[0061] Step S4 includes the following sub-steps:

[0062] Step 4.1: Define the curve coefficients of the fitted curves p and q as follows:

[0063]

[0064] Where G = {y p -i-1 y p -i y p -i+1 ...,y p 0} represents the i-th order polynomial fitting of the gas concentration sampling values ​​at times -i, -i+1, ..., 0, and the resulting fitted curve p predicts the gas concentration y at times -i-1, -i, -i+1, ..., 0. p -i-1 y p -i y p -i+1 ... y p 0, H = {y q - i-1 y q - i y q - i+1 ... y0 q} represents the (i+1)th degree polynomial fitting of the gas concentration sampling values ​​at times -i-1, -i, -i+1, ..., 0, and the resulting fitted curve q predicts the gas concentration y at times -i-1, -i, -i+1, ..., 0. q -i-1 y q -i y q -i+1 ... y0 q μ p μ q These are the means of G and H, respectively;

[0065] Step 4.2: Define the similarity between the fitted curve p and the fitted curve q, specifically as follows:

[0066]

[0067] In the above embodiment, when calculating the similarity between the composite curve p and the fitted curve q, the time when q was used to fit the curve is selected, and the old fitted curve is used for prediction.

[0068] When fitting curve p, in this embodiment, let i = 2, then a second-order polynomial is used for fitting, and the sampled values ​​at times -2, -1, and 0 are used to fit curve p. When fitting curve q, a third-order polynomial is used for fitting, and the sampled values ​​at times -3, -2, -1, and 0 are used to fit curve q. To ensure that sequences G and H have the same length, a concentration value at time -3 needs to be added to sequence G. Therefore, the sampled fitting curve p predicts the concentration values ​​at times -3, -2, -1, and 0, i.e., G = {y} p -3 y p -2 y p -1 ...,y p 0}, H={y q -3、y q -2、...、y0 q}, in sequence G, y p -3 y p -2 y p -1 ...,y p 0 represents the concentration prediction values ​​at times -3, -2, -1, and 0 using the fitted curve p, and y represents... q -3 y q -2 ... y0 q These are the predicted concentration values ​​at times -3, -2, -1, and 0 in curve q, respectively.

[0069] In a further preferred embodiment, step S5 includes:

[0070] Step 5.1: Initialize S0 = 0, and based on step 2.2, obtain the fitted curve p;

[0071] Step 5.2: Let i = i + 1, and based on step 2.2, obtain the fitted curve q;

[0072] Step 5.3: Combine step S4 to calculate the similarity S between the fitted curve p and the fitted curve q, i.e., between sequence G and sequence H. Determine whether S has decreased compared to S0. If it has decreased, select the fitting equation of the fitted curve p as the equation to be predicted; otherwise, set S0 = S, rename the fitted curve q to the fitted curve p, and repeat steps 5.1 to 5.3.

[0073] Further, step S6 specifically includes:

[0074] The concentration prediction value of dissolved gas in oil is obtained by predicting the concentration of dissolved gas in oil using the equation to be predicted. Based on the concentration prediction value of dissolved gas in oil, the three-ratio method or the four-ratio method is used for fault diagnosis and fault early warning.

[0075] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for transformer fault early warning based on DGA, characterized in that, Includes the following steps: Step S1. Collect historical concentration data of dissolved gases in transformer oil based on DGA technology to form a concentration sequence; Step S2. Select the nearest i+1 The gas concentration at each moment, using i Fitting the polynomial of degree to obtain the fitted curve p It includes the following sub-steps: Step S2.1, i+1 Each time point is set as follows: -i , -i+1 , -i+2 … -2 , -1 , 0 ; Step S2.2 i+1 The gas concentration at each moment, using i By fitting a polynomial of degree 1, the following system of equations is established: In the formula, , , , X , B , Y These are the independent variable matrix, coefficient matrix, and dependent variable matrix, respectively. n The number of terms in the polynomial; for- i time, , , ... They represent the constant term, linear term, quadratic term, ..., of the polynomial, respectively. n The coefficient of the second term, for- i The gas concentration at any given time; Thus, the fitted curve can be obtained. p ; Step S3. Select the most recent i+2 The gas concentration at each moment, using i+1 The fitting curve is obtained by fitting a polynomial of degree n. q ; Step S4. Calculate the fitted curve q With fitted curve p Similarity; including the following sub-steps: Step S4.1: Define the fitted curve p and fitted curve q The curve coefficients are: in, Indicates the use of - i - i Gas concentration sampling values ​​at times +1, ..., 0 were performed. i Polynomial fitting of degree 1000 yields the fitted curve p right- i -1、- i - i Predicted gas concentration values ​​at times +1, ..., 0 , , ... , Indicates the use of - i- 1、- i - i Gas concentration sampling values ​​at times +1, ..., 0 were performed. i+ First-order polynomial fitting yields the fitted curve. q right- i -1、- i - i Predicted gas concentration values ​​at times +1, ..., 0 , They are respectively G and H The mean; Step S4.2: Define the fitted curve p and fitted curve q Similarity, specifically: Step S5. Repeat steps S3 and S4 until the... m Fitted curve and the first m -1 indicates the first decrease in the similarity of the fitted curves; including: Step S5.1, Initialization S 0=0, based on step S2.2, obtain the fitted curve. p ; Step S5.2, let i=i +1, Based on step S2.2, obtain the fitted curve. q ; Step S5.3: Combine with step S4 to calculate the fitted curve. p With fitted curve q i.e. sequence G with sequence H similarity S ,judge S Comparison S Has 0 decreased? If so, select the fitted curve. p The fitted equation is the equation to be predicted; otherwise, it will be... S 0 =S and the fitted curve q Rename to Fitted Curve p， And repeat steps S5.1 to S5.3; Step S6. Use the first m -1 fitting curves are used to predict dissolved gases in oil and to provide fault warnings.

2. The method for transformer fault early warning based on DGA according to claim 1, characterized in that: The historical concentration data of dissolved gases in step S1 refers to the concentrations of H2, CH4, C2H2, C2H4, and C2H6 dissolved in transformer oil. These gas concentrations are obtained by taking oil samples online using an online oil chromatography monitoring device.

3. The method for transformer fault early warning based on DGA according to claim 1, characterized in that: Step S3 includes: Step S3.1, let i = i +1, adopt i+1 The fitting curve is obtained by fitting a polynomial of degree n. q .

4. The method for transformer fault early warning based on DGA according to claim 1, characterized in that: Step S6 specifically includes: The concentration prediction value of dissolved gas in oil is obtained by predicting the concentration of dissolved gas in oil using the equation to be predicted. Based on the concentration prediction value of dissolved gas in oil, the three-ratio method or the four-ratio method is used for fault diagnosis and fault early warning.

Images