Active Early Warning Method and Device for Transformer Faults Based on Multi-parameter Comprehensive Analysis

Abstract

The present application discloses a method and device for active warning of transformer faults based on multi-parameter comprehensive analysis, which relates to the technical field of transformer fault monitoring. First, the present application performs fault judgment based on the signal measurement values of each sensor in the fault judgment group, realizing the fault judgment warning and timely removal of the transformer. Then, the primary characteristic parameters and secondary characteristic parameters of the signal sequences of various sensors in the pre-warning group before the fault are calculated and analyzed, realizing the warning and removal of the transformer with breakdown risk, and determining the discharge development stage where the transformer is located. Through the comprehensive judgment and analysis of multiple sensors, the present application improves the hierarchy, timeliness and accuracy of fault detection.

Description

Technical Field

[0001] The present application relates to the technical field of transformer fault monitoring, and particularly to a method and device for active warning of transformer faults based on comprehensive analysis of multiple parameters. Background Art

[0002] Transformer fault monitoring technology realizes the comprehensive evaluation of the operation state of a transformer through a large amount of real-time data such as partial discharge, oil and gas, and temperature monitored online, combined with parameters such as the structure of the transformer and factory tests input offline, and forms an effective evaluation and diagnosis result. It provides a scientific and effective reference basis for operation and maintenance personnel, can effectively extend the operation time of equipment, formulate a reasonable maintenance plan, and prevent the occurrence of sudden faults.

[0003] The existing fault evaluation method uses the method of evaluating each parameter separately. Due to the time difference between the occurrence of a fault and the manifestation of the fault, it will lead to the lag of the fault diagnosis result, and this method will cause some faults to be ignored, which cannot meet the requirements of timeliness and accuracy for transformer fault monitoring. Summary of the Invention

[0004] The purpose of the present application is to provide a method and device for active warning of transformer faults based on comprehensive analysis of multiple parameters to improve the timeliness and accuracy of transformer fault monitoring.

[0005] To achieve the above object, the present application provides the following solutions.

[0006] In the first aspect, the present application provides a method for active warning of transformer faults based on comprehensive analysis of multiple parameters, including:

[0007] Obtaining the signal measurement values of various sensors in the fault judgment group; the fault judgment group includes a magnetic leakage sensor, a flow velocity sensor, and a pressure sensor installed on the transformer;

[0008] When the signal measurement values of various sensors in the fault judgment group meet the first warning condition, determining that the transformer is a breakdown fault transformer and cutting off the breakdown fault transformer; otherwise, performing the following steps:

[0009] Obtaining the signal sequences of various sensors in the pre-fault warning group; the pre-fault warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer;

[0010] Using the sliding window method to divide the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively, and calculating the growth rate of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rate of each primary characteristic parameter of each window of the partial discharge UHF sensor to form a growth rate set;

[0011] When the growth rate set meets the development stage early warning condition, the sliding window method is used to divide the signal sequences of various sensors in the pre-fault early warning group respectively, and the secondary characteristic parameters of each window of various sensors in the pre-fault early warning group are calculated;

[0012] When the secondary characteristic parameters of any window of various sensors in the pre-fault early warning group all meet the second early warning condition, determine that the transformer is a breakdown risk transformer, and cut off the breakdown risk transformer; otherwise, according to the signal sequences of various sensors in the pre-fault early warning group, use the time series weighted decision tree model to determine the discharge development stage of the transformer.

[0013] In a second aspect, the present application provides a transformer fault active early warning device based on multi-parameter comprehensive analysis. The transformer fault active early warning device based on multi-parameter comprehensive analysis applies the above-mentioned transformer fault active early warning method based on multi-parameter comprehensive analysis. The transformer fault active early warning device based on multi-parameter comprehensive analysis includes:

[0014] A first signal acquisition module, configured to obtain signal measurement values of various sensors in the fault judgment group; the fault judgment group includes a leakage magnetic sensor, a flow velocity sensor, and a pressure sensor installed on the transformer;

[0015] A breakdown fault transformer judgment module, configured to determine that the transformer is a breakdown fault transformer and cut off the breakdown fault transformer when the signal measurement values of various sensors in the fault judgment group meet the first early warning condition; otherwise, call the following module:

[0016] A second signal acquisition module, configured to obtain signal sequences of various sensors in the pre-fault early warning group; the pre-fault early warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer;

[0017] A growth rate calculation module, configured to use the sliding window method to divide the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively, and calculate the growth rates of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rates of each primary characteristic parameter of each window of the partial discharge UHF sensor, and form a growth rate set;

[0018] A secondary characteristic parameter calculation module, configured to use the sliding window method to divide the signal sequences of various sensors in the pre-fault early warning group respectively when the growth rate set meets the development stage early warning condition, and calculate the secondary characteristic parameters of each window of various sensors in the pre-fault early warning group;

[0019] A development stage determination module is configured to determine that the transformer is a breakdown risk transformer and cut off the breakdown risk transformer when the secondary characteristic parameters of any window of various sensors in the pre-fault warning group all meet the second warning condition. Otherwise, according to the signal sequences of various sensors in the pre-fault warning group, a time series weighted decision tree model is used to determine the discharge development stage of the transformer.

[0020] According to the specific embodiments provided in the present application, the present application has the following technical effects.

[0021] The present application provides a method and device for active warning of transformer faults based on multi-parameter comprehensive analysis. First, the present application performs a breakdown fault judgment on the transformer according to the signal measurement values of various sensors in the fault judgment group. When a breakdown fault exists, an alarm is given and the transformer is cut off in a timely manner. When no breakdown fault exists, the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor are respectively divided by using the sliding window method, and the growth rates of the primary characteristic parameters of each window of the high-frequency current sensor and the growth rates of the primary characteristic parameters of each window of the partial discharge UHF sensor are calculated to form a growth rate set. When the growth rate set meets the development stage warning condition, the signal sequences of various sensors in the pre-fault warning group are divided by using the sliding window method, and the secondary characteristic parameters of each window of various sensors in the pre-fault warning group are calculated. When the secondary characteristic parameters of any window of various sensors in the pre-fault warning group all meet the second warning condition, it is determined that the transformer is a breakdown risk transformer and the breakdown risk transformer is cut off. Otherwise, according to the signal sequences of various sensors in the pre-fault warning group, a time series weighted decision tree model is used to determine the discharge development stage of the transformer. First, the present application performs a fault judgment according to the signal measurement values of each sensor in the fault judgment group, realizes the fault judgment warning and timely cut-off of the transformer, and then calculates and analyzes the primary and secondary characteristic parameters of the signal sequences of various sensors in the pre-fault warning group to realize the warning and cut-off of the breakdown risk transformer and determine the discharge development stage of the transformer. Through the comprehensive judgment and analysis of multiple sensors, the present application improves the hierarchy, timeliness and accuracy of fault detection. Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1Schematic diagram of a transformer fault active warning method based on multi-parameter comprehensive analysis provided by an embodiment of the present application.

[0024] Figure 2 Schematic diagram of the principle of a transformer fault active warning method based on multi-parameter comprehensive analysis provided by an embodiment of the present application.

[0025] Figure 3 Flow chart of the training of the time series weight decision tree sub-model provided by an embodiment of the present application.

[0026] Figure 4 Schematic diagram of the calculation result of the peak EMA of the partial discharge UHF signal provided by an embodiment of the present application.

[0027] Figure 5 Schematic diagram of the calculation result of the skewness EMA of the high-frequency current signal provided by an embodiment of the present application.

[0028] Figure 6 Schematic diagram of the calculation result of the EMA of the peak coefficient of the acoustic signal provided by an embodiment of the present application.

[0029] Figure 7 Schematic diagram of the feature statistical results within the time windows of three periods provided by an embodiment of the present application. Detailed implementation manners

[0030] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0031] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0032] In an exemplary embodiment, as Figure 1 and Figure 2 shown, a transformer fault active warning method based on multi-parameter comprehensive analysis is provided, including the following steps 101 to 106.

[0033] Step 101, obtain the signal measurement values of various sensors in the fault judgment group; the fault judgment group includes a leakage magnetic sensor, a flow velocity sensor, and a pressure sensor installed on the transformer.

[0034] Step 102, when the signal measurement values of various sensors in the fault judgment group meet the first warning condition, determine that the transformer is a breakdown fault transformer and cut off the breakdown fault transformer; otherwise, execute the following steps 103 - 106.

[0035] Step 103, obtain the signal sequences of various sensors in the pre-fault warning group; the pre-fault warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer.

[0036] Step 104, use the sliding window method to divide the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively, and calculate the growth rates of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rates of each primary characteristic parameter of each window of the partial discharge UHF sensor to form a growth rate set.

[0037] Step 105, when the growth rate set meets the development stage warning condition, use the sliding window method to divide the signal sequences of various sensors in the pre-fault warning group respectively, and calculate the secondary characteristic parameters of each window of various sensors in the pre-fault warning group.

[0038] Step 106, when the secondary characteristic parameters of any window of various sensors in the pre-fault warning group all meet the second warning condition, determine that the transformer is a breakdown risk transformer and cut off the breakdown risk transformer; otherwise, according to the signal sequences of various sensors in the pre-fault warning group, use the time series weight decision tree model to determine the discharge development stage of the transformer.

[0039] Implementing the above steps 101 - 106 can achieve phased warning and timely cutting off of transformer faults. For different discharge development stages, different judgment methods are adopted to improve the hierarchy, timeliness, and accuracy of fault detection.

[0040] The action criterion of the effective pressure action quantity protection uses the pressure information of multiple measuring points under the conditions of normal operation, internal short-circuit fault, and external short-circuit fault of the transformer as characteristic quantities. The action value of the measured point pressure is generally 4 - 5 kPa. For small-turn short circuits, the transformer protection based on effective pressure information can identify and act within hundreds or even dozens of milliseconds, while for large-turn short circuits, the protection device can act within 20 ms to 30 ms. At the same time, this protection method can distinguish between external short-circuit faults and internal short-circuit faults, and its protection criterion is not affected by other internal faults such as overheating faults, with high reliability.

[0041] The oil flow rate is usually applied in the process of judging the operation of the heavy gas. Further, the oil flow rate setting value of the traditional heavy gas protection is 1 - 1.5 m / s.

[0042] When a breakdown fault occurs inside the transformer, causing a short circuit, it will lead to the distortion of the leakage magnetic field distribution inside the transformer.

[0043] In the embodiments of the present application, the high-frequency current sensor, ultrasonic sensor, and partial discharge UHF sensor installed on the transformer are set as the pre-fault warning group, and the leakage magnetic field sensor, flow rate sensor, and pressure sensor installed on the transformer are set as the fault judgment group to comprehensively monitor the transformer fault.

[0044] In another exemplary embodiment, the following settings need to be made before step 101 above.

[0045] Install high-frequency current, ultrasonic, partial discharge UHF, leakage magnetic field, flow rate, and pressure sensors on the transformer. Among them, the high-frequency current, ultrasonic, and partial discharge UHF sensors are classified into the pre-fault warning group, and the leakage magnetic field, flow rate, and pressure sensors are classified into the post-fault severity judgment group, that is, the fault judgment group, to collect the signals during the operation of the transformer in real time.

[0046] In this embodiment, the selected high-frequency current sensor, ultrasonic sensor, partial discharge UHF sensor, flow rate sensor, and pressure sensor are all optical sensors, and the transformer used for testing is a 110 kV full-scale transformer. The installation positions of the sensors are as follows: the high-frequency current sensor is installed on the bushing end shield, the ultrasonic sensor is installed on the top of the oil tank, the partial discharge UHF sensor is installed at the handhole position, the flow rate sensor is installed above the winding, and the pressure sensor is installed at the position facing the tank wall between the two-phase windings. After the installation is completed, all sensors are started.

[0047] In another exemplary embodiment, as Figure 2 shown, in steps 101 - 106 above, if the signal measurement values of various sensors in the fault judgment group meet the first warning condition, it is determined that the transformer has a breakdown fault, and the breakdown fault transformer is promptly cut off; if not all exceed the threshold, but both the high-frequency current and partial discharge UHF sensors have a primary characteristic parameter with an absolute value of growth rate greater than or equal to 20%, then further judge the secondary characteristic parameters of each sensor in the pre-fault warning group. If the second warning condition is met, it is determined that the transformer has a breakdown risk, and the breakdown risk transformer is cut off. If the second warning condition is not met, the discharge development stage of the transformer is determined by using the time series weight decision tree model according to the signal sequence of various sensors in the pre-fault warning group; if the growth rate of all signals in the pre-fault warning group is less than 20%, it is determined that it is interference signal, and continue to monitor the leakage magnetic field, pressure, and flow rate signals of the transformer.

[0048] In another exemplary embodiment, in step 104 above, the pre-fault warning group sensor signals are divided by the sliding window method, and the primary characteristic parameters within each window of the high-frequency current and partial discharge UHF signals are calculated. If there are primary characteristic parameters with an absolute value of the growth rate greater than 20% in both signals, the pre-fault warning group signals are normalized by exponential moving average (EMA) and the secondary characteristic calculation is performed (i.e., step 105 is executed); otherwise, the signals are continuously monitored.

[0049] The transformer collects sound, high-frequency current, and partial discharge UHF signals through sensors, and uses the sliding window partitioning technique to partition the signals according to time. The window duration is set to T seconds, and the sliding step is t seconds. The signals within the window are demodulated and the signal information is statistically analyzed to obtain the frequency-current amplitude information corresponding to the sound signal within the window, and the phase-voltage amplitude information corresponding to the high-frequency current and partial discharge UHF signals.

[0050] The signals within the windows of the high-frequency current and partial discharge UHF sensors are preliminarily processed to obtain the primary characteristic parameters of the high-frequency current and partial discharge UHF sensors.

[0051] Among them, the amplitude average value and the window phase width of the high-frequency current and partial discharge UHF signals are calculated.

[0052] The calculation formula for the amplitude average value is shown as follows.

[0053] .

[0054] Among them, is the amplitude average value of the signals of the type sensor within the th window, is the amplitude of the type sensor at the th data point within the th window. When takes 0, the type sensor is the high-frequency current sensor; when takes 1, the

[0055] type sensor is the UHF sensor; i is the data point number obtained by sampling within the window, i = 1, 2, 3, …, I; I is the number of data points within the window.

[0056] .

[0057] Among them, is the type sensor within the The window phase width within a window is the maximum phase point at which the signal of the th type of sensor is located within the th window, and is the minimum phase point at which the signal of the th type of sensor is located within the

[0058] The formula for calculating the growth rate of the average amplitude and window phase width of two adjacent windows is as follows.

[0059] .

[0060] .

[0061] Among them, is the growth rate of the average amplitude of the th type of sensor in the th window, is the average amplitude of the signal of the th type of sensor within the th window, is the growth rate of the window phase width of the th type of sensor in the th window, is the window phase width of the th type of sensor within the th window.

[0062] For example, for a partial discharge UHF sensor, the average amplitudes of two consecutive windows are 0.54V and 0.98V respectively, and the growth rate is 81.48%. The phase widths are 102° and 198° respectively, and the growth rate is 94.12%. For a high-frequency current sensor, the characteristic amplitudes of two consecutive windows are 0.025A and 0.017A respectively, and the growth rate is -32%. The phase widths are 217° and 270° respectively, and the growth rate is 24.42%.

[0063] The exponential moving average (EMA) is used to normalize the signal characteristic parameters and calculate the normalized characteristic values.

[0064] The signal amplitudes of I consecutive data points within the latest window are obtained in real time, and the quantization index of the current moment is calculated as shown in the following formula.

[0065] .

[0066] Among them, and are the quantization indices of the data within window n and the data within the previous window n - 1 respectively, It is the secondary characteristic parameter of the data within the current window n. is the smoothing constant, which is between 0 and 1. Generally, it takes values from 0.1 to 0.3. In this embodiment, it takes 0.1.

[0067] For the UHF signal of partial discharge, the secondary characteristic parameter is the kurtosis of the maximum amplitude, and the calculation formula is shown as follows.

[0068] .

[0069] Among them, is the kurtosis of the maximum amplitude, is the amplitude of the i-th data point within the window, is the average value of the amplitudes of each data point within the window, is the number of data points within the window.

[0070] For the high-frequency current signal, the secondary characteristic value is the slope of the maximum amplitude, and the calculation formula is shown as follows.

[0071] .

[0072] Among them, is the slope of the maximum amplitude.

[0073] For the ultrasonic signal, the secondary characteristic value is the amplitude peak coefficient, and the calculation formula is shown as follows.

[0074] .

[0075] Among them, is the amplitude peak coefficient, is the maximum amplitude within the window, is the root mean square of the signal amplitudes within the window.

[0076] When setting the threshold interval in the stage approaching breakdown, the peak threshold interval of the UHF signal of partial discharge is 0.638 - 1, the slope threshold interval of the high-frequency current signal is 0 - 0.238, and the peak coefficient threshold interval of the ultrasonic signal is 0.803 - 1.

[0077] In this embodiment, based on a large number of experimental results, the thresholds set for the three sensors are respectively: the upper limit of the peak threshold of the UHF signal of partial discharge is 0.6, the lower limit of the slope threshold of the high-frequency current signal is 0.25, and the upper limit of the peak coefficient threshold of the ultrasonic signal is 0.8. To more clearly illustrate the change trend of the EMA calculation results of the signal characteristic values in each fault development stage, the data of a certain test set (the calculation results of the entire fault development stage) are selected for illustration, as Figures 4 - 6 shown, Figures 4 - 6They are respectively the calculation results of the peak EMA of the partial discharge UHF signal, the skewness EMA of the high-frequency current signal, and the peak coefficient EMA of the acoustic signal.

[0078] Exemplarily, select the latest time window, calculate the kurtosis of the partial discharge UHF signal, the maximum amplitude skewness of the high-frequency current signal, and the peak coefficient of the ultrasonic signal. The calculation results are 0.12, 0.63, and 0.03 respectively, all of which do not exceed the set threshold. Therefore, combine the high-frequency current, the amplitude and phase information of the partial discharge UHF signal, and the amplitude and frequency information of the ultrasonic signal within the latest time window into a feature vector.

[0079] In another exemplary embodiment, in step 106 above, if all the secondary characteristic parameters of the signals of various sensors in the pre-fault warning group are within the set threshold range, it is determined as the stage approaching breakdown (that is, it is determined that the transformer is a transformer at breakdown risk) and an alarm is issued. At the same time, cut off the transformer at breakdown risk. Otherwise, use the time series weight gradient decision tree model to calculate the probability of the discharge development stage of each signal in the pre-fault warning group, use the entropy weight method to allocate the signal addition weights, select the stage with the highest probability as the judgment result, and issue an alarm prompt, and continue to monitor the signals.

[0080] This application proposes a recursive multi-class fault diagnosis overall framework, and constructs a transformer fault diagnosis method based on the gradient boosting decision tree in the overall framework.

[0081] In another exemplary embodiment, in step 106 above, during the fault development stage, the numerical differences of the characteristic signals are not very obvious and there are fluctuations. At the same time, there will also be a silent period, which makes it difficult to accurately identify based on the conventional threshold method. However, through the measured values, it can be found that there are differences in the amplitudes, corresponding phases and frequency information of different signal development stages. Therefore, an algorithm can be used to identify this part of the signals to more accurately determine the fault state.

[0082] If no secondary characteristic parameter of any sensor signal in the pre-fault warning group exceeds the set threshold, that is, the second warning condition is not met, then convert the amplitude, phase or amplitude, frequency information of each sensor signal into a feature vector. Within one time window, the feature vectors of the partial discharge UHF signal, the high-frequency current signal, and the acoustic signal are as follows.

[0083] 。

[0084] 。

[0085] 。

[0086] Among them, U represents the UHF signal of partial discharge; H represents the high-frequency current signal; A represents the acoustic signal; y is the development stage of the signal, y ∈ {1, 2, 3}. When y = 1, the signal is in the discharge initiation stage. When y = 2, the signal is in the severe stage. When y = 3, the signal is in the silent stage. N is the number of windows. In the embodiment of the present application, the discharge development stage is divided into the discharge initiation stage, the severe stage, and the silent stage.

[0087] For the UHF signal of partial discharge and the high-frequency current signal, the maximum value of the signal amplitude and the maximum value of the phase shift within the window are selected as the signal characteristics of the window. The specific calculation formula is shown as follows.

[0088] 。

[0089] Among them, is the maximum value of the phase shift of the UHF signal of partial discharge within window n, is the phase of the i-th data point of the UHF signal of partial discharge within window n, and I is the number of data points within the window.

[0090] 。

[0091] Among them, is the maximum value of the phase shift of the high-frequency current signal within window n, is the phase of the i-th data point of the high-frequency current signal within window n.

[0092] 。

[0093] Among them, is the maximum value of the amplitude of the UHF signal of partial discharge within window n, is the amplitude of the i-th data point of the UHF signal of partial discharge within window n.

[0094] 。

[0095] Among them, is the maximum value of the amplitude of the high-frequency current signal within window n, is the amplitude of the i-th data point of the high-frequency current signal within window n.

[0096] For the acoustic signal, the maximum value of the signal amplitude within the window and its corresponding phase are selected as the signal characteristics of the window.

[0097] 。

[0098] 。

[0099] Among them, is the amplitude of the i-th data point of the acoustic signal within window n, is the maximum amplitude of the acoustic signal within window n, is the phase of the i-th data point of the acoustic signal within window n, is the maximum phase of the acoustic signal within window n.

[0100] In another exemplary embodiment, time dimension information and a time decay function are introduced to balance the timeliness of features.

[0101] The time dimension is represented by the sampling interval to capture the degree of dynamic change of features and provide a basis for subsequent weight calculation. For samples that are far from the current time, weaken their influence on the final result. After adding the time weight, the calculated probability distribution is smoother than that without adding it.

[0102] The formula for calculating the basic time weight is shown as follows.

[0103] .

[0104] Where, is the basic time weight at time t, is the time decay factor; is the difference between the sample time and the current time.

[0105] For the input signal, calculate the basic time weight corresponding to the window it contains, and calculate the average time weight of the same type of signal .

[0106] .

[0107] Where, m is the type of sensor, that is, the signal type, y is the stage where the signal is located, is the number of samples in the signal of the m-th type of sensor that are in the discharge development stage y, is the basic time weight at time t of the samples in the signal of the m-th type of sensor that are in the discharge development stage y.

[0108] Establish a prediction model: .

[0109] Where, is the prediction model obtained from the k-th training, is the prediction model obtained from the (k + 1)-th training, is the gradient coefficient, is the negative gradient of the pre-prediction for the (k + 1)-th training, is the input feature vector.

[0110] Input the initial prediction model: ; where, is the initial prediction model.

[0111] Sample initialization probability distribution: .

[0112] Among them, is the distribution probability of each sample in the k-th training process, , and are the probabilities that each sample belongs to the first, second, and third prediction results in the k-th training process, respectively.

[0113] Calculate the negative gradient before prediction , as the learning target of the new decision tree, ; among them, is the actual probability of the sample. When the sample belongs to category 2, is [0, 1, 0].

[0114] Arrange the signals of the same type in a group of signals in ascending order, calculate the median value of the two data as the splitting point. Taking the maximum value of the amplitude of the partial discharge UHF signal as an example, the calculation formula of this splitting point is shown as follows.

[0115] .

[0116] Among them, is the splitting point vector of the maximum value of the amplitude of the partial discharge UHF signal, N is the number of windows, , , , and are the maximum values of the amplitude of the partial discharge UHF signal in windows 1, 2, 3, N - 1, and N, respectively.

[0117] Calculate the gain and set the gain threshold. Select the splitting point with the maximum gain during each split. If the gain is greater than the threshold, continue to split; otherwise, stop splitting. When the gains are the same, preferentially select the smaller value.

[0118] .

[0119] .

[0120] Among them, is the gain, represents the information entropy of the sample set S of the parent node; |S| is the total number of samples; is the information entropy of the left subtree subset at the splitting point; | | is the number of samples in the left subtree subset; is the information entropy of the right subtree subset at the splitting point; | | is the number of samples in the subset of the right subtree.

[0121] The gain calculation of this application is applied in the growth process of the decision tree to find a balance between model complexity and prediction accuracy. This gain has the same effect as the loss function. Sometimes, even though the loss function is still decreasing, if the gain brought by the split is too small, the split will also stop.

[0122] After selecting the split point, calculate the average negative gradient of the left and right subtrees of the sample split point as the leaf node of the decision tree.

[0123] 。

[0124] 。

[0125] Among them, is the predicted negative gradient of the left subtree of the split point in the (k + 1)-th training, is the predicted negative gradient of the right subtree of the split point in the (k + 1)-th training.

[0126] Input the classification sample, compare it with the decision tree split point, and update the decision tree model using the corresponding leaf node:

[0127] 。

[0128] 。

[0129] Calculate the sample probability using the softmax function with time weights, so that more recent samples have higher influence and improve the adaptability to the current environment.

[0130] 。

[0131] Among them, is the distribution probability of each sample in the (k + 1)-th training process, is the prediction model obtained in the -th training.

[0132] According to the actual situation requirements, input the model parameters, including the number of trees, learning rate, maximum depth, subsampling rate, and minimum number of samples in a leaf.

[0133] To illustrate the specific implementation of the technical solution of this application, the following examples are set in this application.

[0134] Data processing: The partial discharge UHF amplitude and phase information of adjacent 5 time windows are respectively samples 1 - 5: ,Among them, is the input vector of the partial discharge UHF sensor.

[0135] Its corresponding stage label is: ; among which, is the label corresponding to the input vector of the partial discharge UHF sensor.

[0136] Extract data features: , ; among which, is the partial discharge UHF amplitude feature vector, is the partial discharge UHF phase feature vector.

[0137] Calculate the median split point of the data after sorting: , , is the median split point of the partial discharge UHF amplitude feature vector, is the median split point of the partial discharge UHF phase feature vector.

[0138] The two groups of data are arranged in order and correspond to the true labels: , , is the label corresponding to the partial discharge UHF amplitude feature vector, is the label corresponding to the partial discharge UHF phase feature vector.

[0139] Establish an initialization model for the signal of a certain stage : .

[0140] Set the initial values of the probabilities of a certain sample being in four stages : .

[0141] Time decay factor is set to 0.1; the sampling interval is 0.01 s, and calculate the time weights of each sample :

[0142] Sample 5: .

[0143] Sample 4: .

[0144] Sample 3: .

[0145] Sample 2: .

[0146] Sample 1: .

[0147] For each signal, a decision tree needs to be established separately for each stage. Take stage 2 as an example.

[0148] The first round of iteration.

[0149] The calculation formula for the negative gradient (residual) of class 2 for all samples is shown as follows.

[0150] 。

[0151] Among them, is the negative gradient of class 2, is the actual probability of the samples of class 2, is the predicted probability of the sample.

[0152] Calculate the split point gain, and set the gain threshold to 0.1. For the amplitude signal, the information gain of the split point is 0.35.

[0153] The left subtree samples include 1 (0.2), and the corresponding gradient is 。

[0154] The right subtree samples include 4 (0.5, 1.3, 0.75, 2.1), and the corresponding gradients are respectively , , , 。For the parent node, the positive residual probability is: The negative residual probability is: 。

[0155] Parent node entropy: 。

[0156] Left subtree entropy: 。

[0157] Right subtree entropy: 。

[0158] 。

[0159] Similarly, the gain of the split point 0.625 is 0.0202; the gain of the split point 1.025 is 0.4202; the gain of the split point 1.7 is 0.171. By comparison, the information gain of the split point 1.025 is the largest, which is 0.4202. Therefore, the best gain split point for this set of data is 1.025.

[0160] Re-split the samples after splitting according to the amplitude feature: The right subtree after splitting only contains one sample, and the average negative gradient is , and it cannot be split. Split the left subtree. The amplitude samples corresponding to the left subtree are: , and its corresponding split point is: , and calculate the information gain of each point.

[0161] When the splitting point is 23.5, the information gain is 0.3115. When the splitting point is 36.5, the information gain is 0. When the splitting point is 48.5, the information gain is 0.3115. Select 23.5 as the splitting point. After splitting, the average negative gradient of the left subtree is , and the average negative gradient of the right subtree is .

[0162] The decision tree established is as shown in the following series of program segments.

[0163] if , ;

[0164] else if , ;

[0165] else ;

[0166] Among them, is the model output probability corresponding to the UHF amplitude feature vector of partial discharge.

[0167] Input sample 2 belonging to category 2, and the extracted data features are [0.5, 32].

[0168] In the score, the learning rate η is set to 0.1, and it is calculated that ; among them, is the sequential weight decision tree sub-model of the discharge development stage 2 corresponding to the partial discharge UHF sensor after one iteration.

[0169] Similarly, the scores of this sample in other types of decision trees can be obtained.

[0170] .

[0171] .

[0172] Among them, , are the sequential weight decision tree sub-models of the discharge development stages 1 and 3 corresponding to the partial discharge UHF sensor after 1 round of iteration, respectively.

[0173] Use the softmax function considering time weights to calculate the probabilities of sample 2 in each discharge development stage: , , ; among them, , , The probabilities of sample 2 corresponding to the partial discharge UHF sensor after 1 round of iteration being in the discharge development stages 1, 2, and 3 output by the sequential weight decision tree sub - models respectively.

[0174] Exemplarily, according to the actual situation requirements, input the model parameters, and the initial model is , the initial probabilities of the three stages of the sample , the gain threshold Gain is (greater than) 0.1, the learning rate η is 0.05, and the maximum depth of the leaf nodes is 5. The gradient decision tree will be trained using the historical experimental data of each signal and each period of the 110 kV full - scale transformer used in the experiment. The high - frequency current, ultrasonic, and partial discharge UHF signal data of each stage of the historical faults of transformers with the same voltage level are divided into three parts: 80% is the training set, 10% is the validation set, and 10% is the test set. Use the training set data to train the sequential weight gradient decision tree sub - models of the high - frequency current, ultrasonic, and partial discharge UHF signals in the three discharge development stages of the discharge starting stage, severe period, and silent period respectively, and use the validation set and the test set to test the accuracy of the sequential weight gradient decision tree sub - models.

[0175] The entropy value method is used to calculate the weights of the judgment results of the sound, high - frequency current, and partial discharge UHF sensor signals in the total results. Considering the information of the three signals comprehensively, the specific calculation formula is shown as follows.

[0176] .

[0177] .

[0178] .

[0179] .

[0180] .

[0181] .

[0182] .

[0183] In the above formula, is the final probability result, is the probability that the transformer belongs to the discharge development stage y, is the weight of the discharge development stage y, is the probability that the signal sequence of the m - th type of sensor in the pre - fault warning group belongs to the discharge development stage y, is the information entropy redundancy of the discharge development stage y, is the discharge development stage of the information entropy redundancy, is the information entropy of the discharge development stage y. is the proportion of the samples for training the time - series weighted decision tree sub - model corresponding to the m - th type of sensor in the pre - fault warning group for the discharge development stage y. is the number of samples for training the time - series weighted decision tree sub - model corresponding to the m - th type of sensor in the pre - fault warning group for the discharge development stage y. is for training the time - series weighted decision tree sub - model corresponding to the m - th type of sensor in the pre - fault warning group for the discharge development stage of the number of samples of the time - series weighted decision tree sub - model. is the normalization factor. is the total number of categories.

[0184] Exemplarily, the specific training process is described in the following embodiments. As Figure 3 shown, it specifically includes the following steps.

[0185] Input the partial discharge UHF signal into the three time - series weighted gradient decision tree sub - models of the partial discharge UHF signal discharge initiation stage, the partial discharge UHF signal intense period, and the partial discharge UHF signal silent period, and calculate the probabilities of the signal for the three stages, which are 37%, 46%, and 27% respectively.

[0186] Input the high - frequency current signal into the three time - series weighted gradient decision tree sub - models of the high - frequency current signal discharge initiation stage, the high - frequency current signal intense period, and the high - frequency current signal silent period, and calculate the probabilities of the signal for the three stages, which are 11%, 49%, and 44% respectively.

[0187] Input the acoustic signal into the three time - series weighted gradient decision tree sub - models of the acoustic signal discharge initiation stage, the acoustic signal intense period, and the acoustic signal silent period, and calculate the probabilities of the signal for the three stages, which are 51%, 63%, and 21% respectively.

[0188] Use the entropy method to calculate the weights of the judgment results of the partial discharge UHF sensor signal, the high - frequency current sensor signal, and the sound sensor signal in the total result, and comprehensively consider the information of the three signals.

[0189] Calculate the weight of the partial discharge UHF sensor signal as: .

[0190] Calculate the weight of the high - frequency current sensor signal as: .

[0191] Calculate the weight of the sound sensor signal as: .

[0192] Add the three signals according to the calculated weights, and the calculation results are as follows.

[0193] P1 = 37%×31.54% + 11%×48.79% + 51%×19.67% = 27.07%.

[0194] P2 = 46%×31.54% + 49%×48.79% + 63%×19.67% = 50.81%.

[0195] P3 = 27%×31.54% + 44%×48.79% + 21%×19.67% = 34.13%.

[0196] Among them, P1, P2, and P3 are the probabilities of the partial discharge UHF sensor being in the discharge initiation stage, intense period, and silent period respectively.

[0197] Using the formula P = max(P1, P2, P3) = 50.81%, where P is the final judgment result, select the intense period as the final judgment result and issue an alarm.

[0198] To illustrate the differences in the fault amplitude characteristics at different fault development stages, the characteristic statistical results within the three - period time windows after visualizing part of the data in a certain test set are as Figure 7 shown.

[0199] In another exemplary embodiment, step 202 above specifically includes the following steps.

[0200] During the whole process of fault monitoring, after starting the sensor, if the monitored value of any one of the signals of the magnetic flux leakage, pressure, and flow rate sensors in the fault severity judgment group is greater than 0, the signal needs to be further judged, otherwise continuous monitoring is carried out.

[0201] When all three signal sensors exceed the threshold, the fault is judged as the severe fault stage, and this transformer is a breakdown fault transformer, and the breakdown fault transformer is cut off in time.

[0202] Optionally, the upper limit value range of the magnetic flux leakage sensor threshold is 1.5 - 2.0 B / mT, the upper limit value range of the flow rate sensor threshold is 0 - 1 m / s, and the upper limit value range of the pressure sensor threshold is 0 - 2 kPa.

[0203] In this embodiment, the upper limit value range of the magnetic flux leakage sensor threshold is set to 1.5 B / mT, the upper limit value range of the flow rate sensor threshold is 0.1 m / s, and the upper limit value range of the pressure sensor threshold is 0.5 kPa. The measured values of the pressure and flow rate sensors do not exceed the threshold.

[0204] In the method of the above embodiments of the present application, high-frequency current, ultrasonic, partial discharge UHF, leakage magnetic flux, flow rate, and pressure sensors are installed on the transformer, and the sensors are divided into a fault judgment group and a pre-fault warning group. If the signal measurement values of various sensors in the fault judgment group do not meet the first warning condition, the signals of the sensors in the pre-fault warning group are divided by the sliding window method, and the preliminary characteristic values in each window of the high-frequency current and partial discharge UHF signals are calculated. If there is a primary characteristic parameter with an absolute value of the growth rate greater than 20% in both signals, the signals in the pre-fault warning group are normalized by EMA and the secondary characteristics are calculated. Otherwise, the signals are continuously monitored. If all the secondary characteristic parameters of the signals in the pre-fault warning group are within the set threshold range, it is determined as a transformer with breakdown risk and an alarm is issued. At the same time, the transformer with breakdown risk is disconnected. Otherwise, the probability of the discharge development stage of each signal in the pre-fault warning group is calculated using the time series weighted gradient decision tree model, the signal addition weights are assigned using the entropy weight method, the stage with the highest probability is selected as the judgment result, an alarm prompt is issued, and the signals are continuously monitored.

[0205] Through comprehensive judgment of multiple sensors and algorithm analysis, this method realizes phased early warning and timely disconnection of transformer faults. For different discharge development stages, different judgment methods are adopted, which improves the hierarchy, timeliness, and accuracy of fault detection.

[0206] Based on the same inventive concept, an embodiment of the present application also provides a transformer fault active warning device based on multi-parameter comprehensive analysis for implementing the above-mentioned transformer fault active warning method based on multi-parameter comprehensive analysis. The implementation solutions provided by this device to solve problems are similar to the implementation solutions recorded in the above method. Therefore, the specific limitations in one or more embodiments of the following transformer fault active warning device based on multi-parameter comprehensive analysis can refer to the limitations on the transformer fault active warning method based on multi-parameter comprehensive analysis in the above text, and will not be repeated here.

[0207] In an exemplary embodiment, a transformer fault active warning device based on multi-parameter comprehensive analysis is provided, including the following modules.

[0208] The first signal acquisition module is used to obtain the signal measurement values of various sensors in the fault judgment group; the fault judgment group includes a leakage magnetic flux sensor, a flow rate sensor, and a pressure sensor installed on the transformer.

[0209] The breakdown fault transformer judgment module is used to determine the transformer as a breakdown fault transformer and disconnect the breakdown fault transformer when the signal measurement values of various sensors in the fault judgment group meet the first warning condition. Otherwise, the following modules are called.

[0210] The second signal acquisition module is used to obtain the signal sequences of various sensors within the pre-fault warning group; the pre-fault warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer.

[0211] The growth rate calculation module is used to divide the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively by using the sliding window method, and calculate the growth rates of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rates of each primary characteristic parameter of each window of the partial discharge UHF sensor, forming a growth rate set.

[0212] The secondary characteristic parameter calculation module is used to, when the growth rate set meets the development stage warning condition, divide the signal sequences of various sensors within the pre-fault warning group respectively by using the sliding window method, and calculate the secondary characteristic parameters of each window of various sensors within the pre-fault warning group.

[0213] The development stage determination module is used to, when the secondary characteristic parameters of any window of various sensors within the pre-fault warning group all meet the second warning condition, determine that the transformer is a breakdown risk transformer, and cut off the breakdown risk transformer; otherwise, according to the signal sequences of various sensors within the pre-fault warning group, use the time series weighted decision tree model to determine the discharge development stage of the transformer.

[0214] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0215] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0216] Specific examples are used in this article to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, based on the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for active early warning of transformer faults based on comprehensive analysis of multiple parameters, characterized in that, Including: Obtain the signal measurement values of various sensors in the fault judgment group; The fault judgment group includes a leakage magnetic sensor, a flow velocity sensor, and a pressure sensor installed on the transformer; When the signal measurement values of various sensors in the fault judgment group meet the first warning condition, determine that the transformer is a breakdown fault transformer and cut off the breakdown fault transformer; otherwise, perform the following steps: Obtain the signal sequences of various sensors in the pre-fault warning group; the pre-fault warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer; Use the sliding window method to divide the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively, and calculate the growth rates of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rates of each primary characteristic parameter of each window of the partial discharge UHF sensor, and form a growth rate set; When the growth rate set meets the development stage warning condition, use the sliding window method to divide the signal sequences of various sensors in the pre-fault warning group respectively, and calculate the secondary characteristic parameters of each window of various sensors in the pre-fault warning group; When the secondary characteristic parameters of any window of various sensors in the pre-fault warning group all meet the second warning condition, determine that the transformer is a breakdown risk transformer and cut off the breakdown risk transformer; otherwise, according to the signal sequences of various sensors in the pre-fault warning group, use the time series weighted decision tree model to determine the discharge development stage of the transformer; Each primary characteristic parameter includes an amplitude average value and a window phase width; The secondary characteristic parameter of the high-frequency current sensor is the maximum amplitude slope; The secondary characteristic parameter of the ultrasonic sensor is the amplitude peak coefficient; The secondary characteristic parameter of the partial discharge UHF sensor is the maximum amplitude kurtosis.

2. The active warning method for transformer faults based on multi-parameter comprehensive analysis according to claim 1, wherein The first warning condition is that the signal measurement value of the leakage magnetic sensor is greater than the leakage magnetic threshold, the signal measurement value of the flow velocity sensor is greater than the flow velocity threshold, and the signal measurement value of the pressure sensor is greater than the pressure threshold; The development stage warning condition is that there are primary characteristic parameters with an absolute value of the growth rate greater than the growth rate threshold in adjacent windows for both the high-frequency current sensor and the partial discharge UHF sensor; The second warning condition is that the maximum amplitude slope is within the slope range in the near-breakdown stage, the amplitude peak coefficient is within the coefficient range in the near-breakdown stage, and the maximum amplitude kurtosis is within the kurtosis range in the near-breakdown stage.

3. The active warning method for transformer faults based on multi-parameter comprehensive analysis according to claim 1, characterized in that According to the signal sequences of various sensors in the pre-fault warning group, using the time series weighted decision tree model to determine the discharge development stage of the transformer, specifically including: According to the signal sequences of various sensors in the pre-fault warning group, construct the feature vectors of the signal sequences of various sensors in the pre-fault warning group; Respectively input the feature vectors of the signal sequences of various sensors in the pre-fault warning group into the time series weighted decision tree sub-models of each discharge development stage corresponding to various sensors, and determine the probabilities that the signal sequences of various sensors in the pre-fault warning group belong to each discharge development stage; Using the following formula, the probabilities of the signal sequences of various sensors in the pre-fault warning group belonging to each discharge development stage are weighted and summed to determine the probability of the transformer belonging to each discharge development stage; ； Among them, is the probability that the transformer belongs to the discharge development stage y, is the weight of the discharge development stage y, is the probability that the signal sequence of the m-th type of sensor in the pre-fault warning group belongs to the discharge development stage y; Based on the probability of the transformer belonging to each discharge development stage, determine the discharge development stage of the transformer.

4. The method for active early warning of transformer faults based on multi-parameter comprehensive analysis according to claim 3, characterized in that, The calculation formula for the weight of the discharge development stage y is: ； ； ； ； Among them, is the information entropy redundancy of the discharge development stage y, is the discharge development stage 's information entropy redundancy, is the information entropy of the discharge development stage y, is the proportion of samples for training the time-series weighted decision tree sub-model corresponding to the m-th type of sensor in the pre-fault warning group for the discharge development stage y, is the number of samples for training the time-series weighted decision tree sub-model corresponding to the m-th type of sensor in the pre-fault warning group for the discharge development stage y, is for training the time-series weighted decision tree sub-model corresponding to the m-th type of sensor in the pre-fault warning group for the discharge development stage 's number of samples of the time-series weighted decision tree sub-model, is the normalization factor.

5. The method for active early warning of transformer faults based on multi-parameter comprehensive analysis according to claim 3, characterized in that, The time-series weight decision tree sub-models corresponding to various sensors in the pre-fault warning group for each discharge development stage are obtained by training in the following manner: Construct a target sample set; the target sample set is a set for training the time-series weight decision tree sub-model of the target discharge development stage. The target discharge development stage is any one of the various discharge development stages, and the target sensor is any one of the various sensors in the pre-fault warning group; Based on the target sample set, train a decision tree model to obtain the time-series weight decision tree sub-model of the target discharge development stage.

6. An active warning device for transformer faults based on multi-parameter comprehensive analysis, characterized in that, The transformer fault active warning device based on multi-parameter comprehensive analysis applies the transformer fault active warning method based on multi-parameter comprehensive analysis according to any one of claims 1-5. The transformer fault active warning device based on multi-parameter comprehensive analysis includes: A first signal acquisition module for obtaining the signal measurement values of various sensors in the fault judgment group; the fault judgment group includes a leakage magnetic sensor, a flow rate sensor, and a pressure sensor installed on the transformer; A breakdown fault transformer judgment module for determining that the transformer is a breakdown fault transformer and cutting off the breakdown fault transformer when the signal measurement values of various sensors in the fault judgment group meet the first warning condition. Otherwise, call the following module: A second signal acquisition module for obtaining the signal sequences of various sensors in the pre-fault warning group; the pre-fault warning group includes a high-frequency current sensor, an ultrasonic sensor, and a partial discharge UHF sensor installed on the transformer; A growth rate calculation module for dividing the signal sequences of the high-frequency current sensor and the partial discharge UHF sensor respectively by using the sliding window method, and calculating the growth rates of each primary characteristic parameter of each window of the high-frequency current sensor and the growth rates of each primary characteristic parameter of each window of the partial discharge UHF sensor to form a growth rate set; A secondary characteristic parameter calculation module for dividing the signal sequences of various sensors in the pre-fault warning group respectively by using the sliding window method and calculating the secondary characteristic parameters of each window of various sensors in the pre-fault warning group when the growth rate set meets the development stage warning condition; A development stage determination module for determining that the transformer is a breakdown risk transformer and cutting off the breakdown risk transformer when the secondary characteristic parameters of any window of various sensors in the pre-fault warning group all meet the second warning condition. Otherwise, determine the discharge development stage of the transformer by using the time-series weight decision tree model according to the signal sequences of various sensors in the pre-fault warning group.

Images