A transformer fault early warning method based on gas concentration time series data

Abstract

The application provides a transformer fault early warning method based on gas concentration time series data, and the method adopts the dissolved gas concentration in oil under the normal state of a transformer to be diagnosed to construct concentration time series data, obtains a mutual information value sequence based on the concentration time series data, trains a mutual information value prediction model through the mutual information value sequence, predicts the upper and lower threshold values of the mutual information value of the dissolved gas in oil of the transformer to be diagnosed at the next monitoring time through the mutual information value prediction model, judges whether the mutual information value of the concentration time series data and the time variable at the next time is within the upper and lower threshold value range, if yes, the transformer to be diagnosed at the next monitoring time is normal, and if no, the transformer to be diagnosed at the next monitoring time has a fault, and the method diagnoses the fault condition of the transformer to be diagnosed at the next monitoring time through the concentration data under the normal state of the transformer to be diagnosed, thereby solving the problems of low transformer fault diagnosis accuracy and false fault alarm of the transformer caused by individual differences.

Description

Technical Field

[0001] This invention relates to the field of transformer fault diagnosis technology, and more specifically, to a transformer fault early warning method based on gas concentration time-series data. Background Technology

[0002] Power transformers are responsible for power transmission and voltage conversion, and are one of the most critical pieces of equipment in a power system. Their operating status affects the reliability of the entire system. Dissolved gas analysis (DGA) has proven to be an effective method for diagnosing internal faults in oil-immersed power transformers worldwide, and can sensitively identify internal faults in oil-immersed transformers.

[0003] The gas concentration time series data of a transformer will show different changing patterns under normal operation and fault conditions. By exploring the changing patterns of gas concentration data under normal operation, and issuing a signal when the pattern changes, the purpose of early warning can be achieved.

[0004] Existing transformer fault early warning methods all use the dissolved gas concentrations of multiple transformers to train the fault prediction model, thereby obtaining the predicted faults of the transformer and issuing alarms. However, since each transformer is an independent entity and the dissolved gas concentrations inside different transformers are different, the fault prediction model trained using the dissolved gas concentrations of other transformers has the problems of low accuracy in transformer fault diagnosis and false alarms for transformer faults. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a transformer fault early warning method based on gas concentration time-series data, which solves the problems of low accuracy in transformer fault diagnosis and false alarms for transformer faults in existing transformer fault early warning methods.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: a transformer fault early warning method based on gas concentration time-series data, comprising the following steps:

[0007] S1. Collect the concentration of dissolved gas in the oil of the transformer under normal conditions and obtain the time series data of the concentration of dissolved gas in each type of oil.

[0008] S2. Calculate the mutual information value between the time series data of dissolved gas concentration in each type of oil and the time variable to obtain the mutual information value sequence corresponding to each type of oil.

[0009] S3. Based on the mutual information value sequence corresponding to each type of dissolved gas in oil, construct and optimize the mutual information value prediction model to predict the upper and lower thresholds of the mutual information value of each type of dissolved gas in oil at the next monitoring time.

[0010] S4. After obtaining the concentration of dissolved gases in each type of oil at the next monitoring time, construct time series data of the concentration of the monitoring gases.

[0011] S5. Calculate the mutual information value between the time series data of the monitored gas concentration and the time variable. When the mutual information value between the time series data of the monitored gas concentration and the time variable exceeds the upper and lower threshold ranges, a transformer fault alarm is triggered.

[0012] Further, the dissolved gas concentrations in the oil in step S1 include: H2 concentration, CH4 concentration, C2H6 concentration, C2H4 concentration, and C2H2 concentration;

[0013] Step S1 specifically involves:

[0014] Starting at time t, 100 concentration values ​​of dissolved gases were collected for each type of oil. These 100 concentration values ​​were divided into 10 groups, forming the time series data of dissolved gas concentrations for each type of oil: C i，j ={C i，j，1 C i，j，n C i，j，10}, where C i，j Let j be the time series data of the concentration of dissolved gas in the i-th type of oil, i = 1, 2, 3, 4, 5, j = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, when i = 1, C 1，j For the j-th set of time series data on the concentration of dissolved H2 in oil, when i = 2, C 2，j This is the time series data of the concentration of dissolved CH4 in oil for the j-th group. When i = 3, C 3，j This is the time series data of the concentration of C2H6 dissolved in oil for the j-th group. When i = 4, C 4，j This is the time series data of the concentration of C2H4 dissolved in oil. When i = 5, C 5，j This represents the j-th set of time-series concentration data of dissolved C2H2 in oil, where n is the number marking the time when each concentration data point was collected. i，j，1 C represents the concentration value of dissolved gas in the i-th type of oil collected at the first time step in the j-th group of time-series concentration data. i，j，n C represents the concentration value of dissolved gases in the i-th type of oil collected at the n-th time step in the j-th group of time-series concentration data. i，j，10 The concentration value of dissolved gas in the i-th type of oil is collected at the 10th time step in the j-th group of time series data.

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

[0016] S21. Calculate the mutual information value between the time series data of dissolved gas concentration and the time variable for each type of oil;

[0017] S22. Construct a mutual information value sequence from multiple mutual information values ​​belonging to the same type of dissolved gas in the oil. The mutual information value sequence is specifically {MI}. i，1 , ..., MI i，j ,...MI i，10}, where {MI i，1 , ..., MI i，j ,...MI i，10} represents the sequence of mutual information values ​​constructed for the mutual information values ​​corresponding to the dissolved gases in the i-th type of oil, MI. i，1 MI represents the mutual information value of the first set of time-series concentration data of dissolved gases in the i-th type of oil. i，j MI is the mutual information value of the j-th group of time-series concentration data of dissolved gases in the i-th type of oil. i，10 is the mutual information value of the 10th set of time series data of dissolved gas concentration in the i-th type of oil.

[0018] Furthermore, the formula for calculating the mutual information value in step S21 is as follows:

[0019]

[0020] Among them, MI i，j C represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The mutual information value with the time variable, p(C) i，j M) is C i，j The joint probability density function of M, where M is a time variable, M = [1, 2, 3, ..., 10].

[0021] Furthermore, the formula for calculating the mutual information value in step S21 is as follows:

[0022] MI i，j =H(C i，j )+H(M)-H(C i，j H)

[0023] Among them, MI i，j C represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The mutual information value with the time variable, H(C) i，j C represents the time series data of the concentration of the j-th group of dissolved gases in the i-th type of oil. i，j Information entropy, H(M) is the information entropy of time variable M, H(C) is the information entropy of time variable M. i，j H) represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The joint entropy of the time variable M, M = [1, 2, 3, ..., 10].

[0024] Furthermore, step S3 includes the following sub-steps:

[0025] S31. Using nonparametric regression, the mutual information value sequence corresponding to each type of oil is fitted to obtain the upper limit sequence and lower limit sequence of mutual information value of each type of oil.

[0026] S32. Construct a training dataset by matching the upper and lower bound sequences of mutual information values ​​of dissolved gases in each type of oil.

[0027] S33. Train the kernel extreme learning machine using the training dataset to obtain the mutual information value prediction model;

[0028] S34. The parameters of the mutual information value prediction model are optimized using the group search optimization method to obtain the optimized mutual information value prediction model.

[0029] S35. Use an optimized mutual information value prediction model to predict the upper and lower threshold values ​​of the mutual information values ​​between the concentration of dissolved gas in oil and time for various types of oil at the next monitoring time.

[0030] Furthermore, step S5 includes the following sub-steps:

[0031] S51. Calculate the mutual information value between the time series data of the concentration of the monitored gas and the time variable;

[0032] S52. When the mutual information value between the time series data of the monitored gas concentration and the time variable is greater than the upper threshold or less than the lower threshold, a transformer fault early warning shall be issued.

[0033] Furthermore, the formula for calculating the mutual information value in step S51 is as follows:

[0034]

[0035] Among them, MI′ i，j C′ is the time series data of the j-th concentration of the gas being monitored in the i-th type of oil. i，j The mutual information value with the time variable M, p(C′) i，j M) is C′ i，j The joint probability density function of M, where M is a time variable, M = [1, 2, 3, ..., 10].

[0036] Furthermore, the formula for calculating the mutual information value in step S51 is as follows:

[0037] MI′ i，j =H(C′) i，j )+H(M)-H(C′ i，j H)

[0038] Among them, MI′ i，jC′ is the time series data of the j-th concentration of the gas being monitored in the i-th type of oil. i，j The mutual information value with the time variable M, H(C′) i，j C′ represents the time series data of the j-th concentration of the monitored gas in the i-th type of oil. i，j Information entropy, H(M) is the information entropy of time variable M, H(C′) i，j H) represents the time series data of the j-th concentration of the monitored gas in the i-th type of oil, C′. i，j The joint entropy of the time variable M, M = [1, 2, 3, ..., 10].

[0039] The technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:

[0040] This invention uses the dissolved gas concentration in the oil of a transformer under normal conditions to construct concentration time-series data. Based on this data, a mutual information value sequence is obtained. A mutual information value prediction model is trained using this sequence. The model then predicts the upper and lower thresholds of the mutual information value of the dissolved gas in the oil at the next monitoring time. It determines whether the mutual information value between the concentration time-series data and the time variable at the next monitoring time falls within these thresholds. If so, the transformer under diagnosis at the next monitoring time is normal; otherwise, it has a fault. By using the concentration data of the transformer under normal conditions to diagnose the fault status at the next monitoring time, this invention solves the problem of low accuracy in transformer fault diagnosis and false alarms due to individual differences. Attached Figure Description

[0041] Figure 1 This is a flowchart of a transformer fault early warning method based on gas concentration time series data. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0043] like Figure 1 As shown, a transformer fault early warning method based on gas concentration time-series data includes the following steps:

[0044] S1. Collect the concentration of dissolved gas in the oil of the transformer under normal conditions and obtain the time series data of the concentration of dissolved gas in each type of oil.

[0045] The dissolved gas concentrations in the oil during step S1 include: H2 concentration, CH4 concentration, C2H6 concentration, C2H4 concentration, and C2H2 concentration;

[0046] Step S1 specifically involves:

[0047] Starting at time t, 100 concentration values ​​of dissolved gases were collected for each type of oil. These 100 concentration values ​​were divided into 10 groups, forming the time series data of dissolved gas concentrations for each type of oil: C i，j ={C i，j，1 C i，j，n C i，j，10}, where C i，j Let j be the time series data of the concentration of dissolved gas in the i-th type of oil, i = 1, 2, 3, 4, 5, j = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, when i = 1, C 1，j For the j-th set of time series data on the concentration of dissolved H2 in oil, when i = 2, C 2，j This is the time series data of the concentration of dissolved CH4 in oil for the j-th group. When i = 3, C 3，j This is the time series data of the concentration of C2H6 dissolved in oil for the j-th group. When i = 4, C 4，j This is the time series data of the concentration of C2H4 dissolved in oil. When i = 5, C 5，j This represents the j-th set of time-series concentration data of dissolved C2H2 in oil, where n is the number marking the time when each concentration data point was collected. i，j，1 C represents the concentration value of dissolved gas in the i-th type of oil collected at the first time step in the j-th group of time-series concentration data. i，j，n C represents the concentration value of dissolved gases in the i-th type of oil collected at the n-th time step in the j-th group of time-series concentration data. i，j，10 The concentration value of dissolved gas in the i-th type of oil is collected at the 10th time step in the j-th group of time series data.

[0048] S2. Calculate the mutual information value between the time series data of dissolved gas concentration in each type of oil and the time variable to obtain the mutual information value sequence corresponding to each type of oil.

[0049] Step S2 includes the following sub-steps:

[0050] S21. Calculate the mutual information value between the time series data of dissolved gas concentration and the time variable for each type of oil;

[0051] In this embodiment, the mutual information value includes two calculation methods, the first being:

[0052] The formula for calculating the mutual information value in step S21 is as follows:

[0053]

[0054] Among them, MI i，j C represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The mutual information value with the time variable, p(C) i，j M) is C i，j The joint probability density function of M, where M is a time variable, M = [1, 2, 3, ..., 10].

[0055] The second type:

[0056] The formula for calculating the mutual information value in step S21 is as follows:

[0057] MI i，j =H(C i，j )+H(M)-H(C i，j H)

[0058] Among them, MI i，j C represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The mutual information value with the time variable, H(C) i，j C represents the time series data of the concentration of the j-th group of dissolved gases in the i-th type of oil. i，j Information entropy, H(M) is the information entropy of time variable M, H(C) is the information entropy of time variable M. i，j H) represents the time series data of the j-th group of dissolved gas concentrations in the i-th type of oil. i，j The joint entropy of the time variable M, M = [1, 2, 3, ..., 10].

[0059] S22. Construct a mutual information value sequence from multiple mutual information values ​​belonging to the same type of dissolved gas in the oil. The mutual information value sequence is specifically {MI}. i，1 , ..., MI i，j ,...MI i，10}, where {MI i，1 , ..., MI i，j ,...MI i，10} represents the sequence of mutual information values ​​constructed for the mutual information values ​​corresponding to the dissolved gases in the i-th type of oil, MI. i，1 MI represents the mutual information value of the first set of time-series concentration data of dissolved gases in the i-th type of oil. i，j MI is the mutual information value of the j-th group of time-series concentration data of dissolved gases in the i-th type of oil. i，10 is the mutual information value of the 10th set of time series data of dissolved gas concentration in the i-th type of oil.

[0060] S3. Based on the mutual information value sequence corresponding to each type of dissolved gas in oil, construct and optimize the mutual information value prediction model to predict the upper and lower thresholds of the mutual information value of each type of dissolved gas in oil at the next monitoring time.

[0061] Step S3 includes the following sub-steps:

[0062] S31. Using nonparametric regression, the mutual information value sequence corresponding to each type of oil is fitted to obtain the upper limit sequence and lower limit sequence of mutual information value of each type of oil.

[0063] In this embodiment, after the fitting process, a fitting curve is obtained. By setting a dividing line, the fitting curve is divided. The data corresponding to the points above the dividing line constitute the upper limit sequence of mutual information values, and the data corresponding to the points below the dividing line constitute the lower limit sequence of mutual information values.

[0064] S32. Construct a training dataset by matching the upper and lower bound sequences of mutual information values ​​of dissolved gases in each type of oil.

[0065] S33. Train the kernel extreme learning machine using the training dataset to obtain the mutual information value prediction model;

[0066] S34. The parameters of the mutual information value prediction model are optimized using the group search optimization method to obtain the optimized mutual information value prediction model.

[0067] S35. Use an optimized mutual information value prediction model to predict the upper and lower threshold values ​​of the mutual information values ​​between the concentration of dissolved gas in oil and time for various types of oil at the next monitoring time.

[0068] S4. After obtaining the concentration of dissolved gases in each type of oil at the next monitoring time, construct time series data of the concentration of the monitoring gases.

[0069] S5. Calculate the mutual information value between the time series data of the monitored gas concentration and the time variable. When the mutual information value between the time series data of the monitored gas concentration and the time variable exceeds the upper and lower threshold ranges, a transformer fault alarm is triggered.

[0070] Step S5 includes the following sub-steps:

[0071] S51. Calculate the mutual information value between the time series data of the concentration of the monitored gas and the time variable;

[0072] In this embodiment, the mutual information value includes two calculation methods, the first being:

[0073] The formula for calculating the mutual information value in step S51 is as follows:

[0074]

[0075] Among them, MI′ i，j C′ is the time series data of the j-th concentration of the gas being monitored in the i-th type of oil. i，j The mutual information value with the time variable M, p(C′) i，j M) is C′ i，j The joint probability density function of M, where M is a time variable, M = [1, 2, 3, ..., 10].

[0076] The second type:

[0077] The formula for calculating the mutual information value in step S51 is as follows:

[0078] MI′ i，j =H(C′) i，j )+H(M)-H(C′ i，j H)

[0079] Among them, MI′ i，j C′ is the time series data of the j-th concentration of the gas being monitored in the i-th type of oil. i，j The mutual information value with the time variable M, H(C′) i，j C′ represents the time series data of the j-th concentration of the monitored gas in the i-th type of oil. i，j Information entropy, H(M) is the information entropy of time variable M, H(C′) i，j H) represents the time series data of the j-th concentration of the monitored gas in the o-th type of oil, C′. i，j The joint entropy of the time variable M, M = [1, 2, 3, ..., 10].

[0080] S52. When the mutual information value between the time series data of the monitored gas concentration and the time variable is greater than the upper threshold or less than the lower threshold, a transformer fault early warning shall be issued.

[0081] The technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:

[0082] This invention uses the dissolved gas concentration in the oil of a transformer under normal conditions to construct concentration time-series data. Based on this data, a mutual information value sequence is obtained. A mutual information value prediction model is trained using this sequence. The model then predicts the upper and lower thresholds of the mutual information value of the dissolved gas in the oil at the next monitoring time. It determines whether the mutual information value between the concentration time-series data and the time variable at the next monitoring time falls within these thresholds. If so, the transformer under diagnosis at the next monitoring time is normal; otherwise, it has a fault. By using the concentration data of the transformer under normal conditions to diagnose the fault status at the next monitoring time, this invention solves the problem of low accuracy in transformer fault diagnosis and false alarms due to individual differences.

[0083] Experiment: Taking a transformer equipped with an online oil chromatography monitoring device as an example, data from 30 consecutive days (sampling once per day) from September 1st to September 30th, 2020 were selected. Data was taken backwards from the data sampled on September 30th, with 10 data points grouped together. These groups were processed using the aforementioned early warning method, with H2 selected as the analysis object. The predicted upper and lower limits of the mutual information value after adding data from October 1st were 0.016 and 0.009, respectively. However, the actual mutual information value after adding data from October 1st was 0.019, exceeding the upper limit threshold, thus triggering an alarm signal. Subsequent additions of data from October 2nd to October 7th also resulted in exceeding the limits, triggering alarm signals. The actual operating condition of the transformer was as follows: the local power company detected a total hydrocarbon concentration exceeding the warning value of 150 μL / L on November 15th, and diagnosed it as a medium-temperature fault (300-700℃) using the three-ratio method. Upon return to the factory for disassembly and inspection, defects such as poor soldering and obvious signs of overheating and carbonization were found in the lower half of the B-phase winding. Using the early warning method described in this paper, compared to the gas concentration threshold, a latent fault could be detected one and a half months in advance in this case, demonstrating the effectiveness of the early warning method of this invention.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A transformer fault early warning method based on gas concentration time-series data, characterized in that, Includes the following steps: S1. Collect the concentration of dissolved gas in the oil of the transformer under normal conditions and obtain the time series data of the concentration of dissolved gas in each type of oil. S2. Calculate the mutual information value between the time series data of dissolved gas concentration in each type of oil and the time variable to obtain the mutual information value sequence corresponding to each type of oil. S3. Based on the mutual information value sequence corresponding to each type of dissolved gas in oil, construct and optimize the mutual information value prediction model to predict the upper and lower thresholds of the mutual information value of each type of dissolved gas in oil at the next monitoring time. S4. After obtaining the concentration of dissolved gases in each type of oil at the next monitoring time, construct time series data of the concentration of the monitoring gases. S5. Calculate the mutual information value between the time series data of the concentration of the monitored gas and the time variable. When the mutual information value between the time series data of the concentration of the monitored gas and the time variable exceeds the upper and lower threshold ranges, a transformer fault alarm is triggered. Step S5 includes the following sub-steps: S51. Calculate the mutual information value between the time series data of the monitored gas concentration and the time variable; the formula for calculating the mutual information value is: in, For the first The first type of gas monitoring in oil Group concentration time series data With time variables mutual information value, for and The joint probability density function, For time variables, ; S52. When the mutual information value between the time series data of the monitored gas concentration and the time variable is greater than the upper threshold or less than the lower threshold, a transformer fault early warning shall be issued.

2. The transformer fault early warning method based on gas concentration time-series data according to claim 1, characterized in that, The dissolved gas concentration in the oil during step S1 includes: concentration, concentration, concentration, Concentration and concentration; Step S1 specifically involves: In time Starting at a given time, 100 concentration values ​​of dissolved gases were collected from each type of oil. These 100 concentration values ​​were divided into 10 groups, forming the time-series data of dissolved gas concentrations for each type of oil: ,in, For the first The first type of dissolved gas in oil Group concentration time series data, , , hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, To mark the time when each concentration data point was collected, For the first The first type of dissolved gas in oil The concentration value collected at the first time step in the group concentration time series data. For the first The first type of dissolved gas in oil The first group of concentration time series data Concentration values ​​collected at each time point, For the first The first type of dissolved gas in oil The concentration value collected at the 10th time step in the time series data of the group concentration.

3. The transformer fault early warning method based on gas concentration time-series data according to claim 2, characterized in that, Step S2 includes the following sub-steps: S21. Calculate the mutual information value between the time series data of dissolved gas concentration and the time variable for each type of oil; S22. Construct a mutual information value sequence from multiple mutual information values ​​belonging to the same type of dissolved gas in the same oil. The mutual information value sequence is specifically as follows: ,in, For the first A sequence of mutual information values ​​constructed from the mutual information values ​​of dissolved gases in different types of oil. For the first The mutual information values ​​of the first set of time-series concentration data of dissolved gases in various types of oil. For the first The first type of dissolved gas in oil Mutual information values ​​of group concentration time series data For the first The mutual information value of the 10th set of time series data of dissolved gas concentrations in various types of oil.

4. The transformer fault early warning method based on gas concentration time-series data according to claim 3, characterized in that, The formula for calculating the mutual information value in step S21 is as follows: in, For the first The first type of dissolved gas in oil Group concentration time series data Mutual information value with time variable, for and The joint probability density function, For time variables, .

5. The transformer fault early warning method based on gas concentration time-series data according to claim 3, characterized in that, The formula for calculating the mutual information value in step S21 is as follows: in, For the first The first type of dissolved gas in oil Group concentration time series data Mutual information value with time variable, For the first The first type of dissolved gas in oil Group concentration time series data Information entropy For time variables Information entropy For the first The first type of dissolved gas in oil Group concentration time series data and time variables joint entropy, .

6. The transformer fault early warning method based on gas concentration time-series data according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Using nonparametric regression, the mutual information value sequence corresponding to each type of oil is fitted to obtain the upper limit sequence and lower limit sequence of mutual information value of each type of oil. S32. Construct a training dataset by matching the upper and lower bound sequences of mutual information values ​​of dissolved gases in each type of oil. S33. Train the kernel extreme learning machine using the training dataset to obtain the mutual information value prediction model; S34. The parameters of the mutual information value prediction model are optimized using the group search optimization method to obtain the optimized mutual information value prediction model. S35. Use an optimized mutual information value prediction model to predict the upper and lower threshold values ​​of the mutual information values ​​between the concentration of dissolved gas in oil and time for various types of oil at the next monitoring time.

7. A transformer fault early warning method based on gas concentration time-series data, characterized in that, Includes the following steps: S1. Collect the concentration of dissolved gas in the oil of the transformer under normal conditions and obtain the time series data of the concentration of dissolved gas in each type of oil. S2. Calculate the mutual information value between the time series data of dissolved gas concentration in each type of oil and the time variable to obtain the mutual information value sequence corresponding to each type of oil. S3. Based on the mutual information value sequence corresponding to each type of dissolved gas in oil, construct and optimize the mutual information value prediction model to predict the upper and lower thresholds of the mutual information value of each type of dissolved gas in oil at the next monitoring time. S4. After obtaining the concentration of dissolved gases in each type of oil at the next monitoring time, construct time series data of the concentration of the monitoring gases. S5. Calculate the mutual information value between the time series data of the concentration of the monitored gas and the time variable. When the mutual information value between the time series data of the concentration of the monitored gas and the time variable exceeds the upper and lower threshold ranges, a transformer fault alarm is triggered. Step S5 includes the following sub-steps: S51. Calculate the mutual information value between the time series data of the monitored gas concentration and the time variable; the formula for calculating the mutual information value is: in, For the first The first type of gas monitoring in oil Group concentration time series data With time variables mutual information value, For the first The first type of gas monitoring in oil Group concentration time series data Information entropy For time variables Information entropy For the first The first type of gas monitoring in oil Group concentration time series data and time variables joint entropy, ; S52. When the mutual information value between the time series data of the monitored gas concentration and the time variable is greater than the upper threshold or less than the lower threshold, a transformer fault early warning shall be issued.

8. The transformer fault early warning method based on gas concentration time-series data according to claim 7, characterized in that, The dissolved gas concentration in the oil during step S1 includes: concentration, concentration, concentration, Concentration and concentration; Step S1 specifically involves: In time Starting at a given time, 100 concentration values ​​of dissolved gases were collected from each type of oil. These 100 concentration values ​​were divided into 10 groups, forming the time-series data of dissolved gas concentrations for each type of oil: ,in, For the first The first type of dissolved gas in oil Group concentration time series data, , , hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, hour, Dissolved in oil The Group concentration time series data, To mark the time when each concentration data point was collected, For the first The first type of dissolved gas in oil The concentration value collected at the first time step in the group concentration time series data. For the first The first type of dissolved gas in oil The first group of concentration time series data Concentration values ​​collected at each time point, For the first The first type of dissolved gas in oil The concentration value collected at the 10th time step in the time series data of the group concentration.

9. The transformer fault early warning method based on gas concentration time-series data according to claim 8, characterized in that, Step S2 includes the following sub-steps: S21. Calculate the mutual information value between the time series data of dissolved gas concentration and the time variable for each type of oil; S22. Construct a mutual information value sequence from multiple mutual information values ​​belonging to the same type of dissolved gas in the same oil. The mutual information value sequence is specifically as follows: ,in, For the first A sequence of mutual information values ​​constructed from the mutual information values ​​of dissolved gases in different types of oil. For the first The mutual information values ​​of the first set of time-series concentration data of dissolved gases in various types of oil. For the first The first type of dissolved gas in oil Mutual information values ​​of group concentration time series data For the first The mutual information value of the 10th set of time series data of dissolved gas concentrations in various types of oil.

10. The transformer fault early warning method based on gas concentration time-series data according to claim 9, characterized in that, The formula for calculating the mutual information value in step S21 is as follows: in, For the first The first type of dissolved gas in oil Group concentration time series data Mutual information value with time variable, for and The joint probability density function, For time variables, .

11. The transformer fault early warning method based on gas concentration time series data according to claim 9, characterized in that, The formula for calculating the mutual information value in step S21 is as follows: in, For the first The first type of dissolved gas in oil Group concentration time series data Mutual information value with time variable, For the first The first type of dissolved gas in oil Group concentration time series data Information entropy For time variables Information entropy For the first The first type of dissolved gas in oil Group concentration time series data and time variables joint entropy, .

12. The transformer fault early warning method based on gas concentration time-series data according to claim 7, characterized in that, Step S3 includes the following sub-steps: S31. Using nonparametric regression, the mutual information value sequence corresponding to each type of oil is fitted to obtain the upper limit sequence and lower limit sequence of mutual information value of each type of oil. S32. Construct a training dataset by matching the upper and lower bound sequences of mutual information values ​​of dissolved gases in each type of oil. S33. Train the kernel extreme learning machine using the training dataset to obtain the mutual information value prediction model; S34. The parameters of the mutual information value prediction model are optimized using the group search optimization method to obtain the optimized mutual information value prediction model. S35. Use an optimized mutual information value prediction model to predict the upper and lower threshold values ​​of the mutual information values ​​between the concentration of dissolved gas in oil and time for various types of oil at the next monitoring time.

Images