A method for detecting the rated capacity of a transformer

Abstract

The present application relates to a kind of transformer rated capacity detection method, comprising: S1 constructs transformer rated capacity detection model;S2 obtains the test sample of transformer nameplate, test sample includes transformer rated capacity, transformer no-load loss, load loss and short-circuit impedance parameter;S3 is input to the transformer rated capacity detection model in test sample;S4: compare whether the capacity value of the transformer rated capacity detection model output is consistent with the rated capacity value of this transformer, when the comparison result is consistent, then identify the output capacity value is real capacity value;When the comparison result is inconsistent, then execute step S5;S5 returns step S2, the capacity value of the transformer rated capacity detection model output is used as new transformer rated capacity value, reacquire transformer no-load loss, load loss and short-circuit impedance parameter, continue to execute transformer rated capacity detection.

Description

Technical Field

[0001] This invention relates to the field of transformer technology, specifically to a method for detecting the rated capacity of a transformer. Background Technology

[0002] my country implements a two-part electricity pricing system for large industrial enterprises, which divides the electricity price into a basic price and a per-unit price. The basic price is calculated based on the transformer capacity (KVA) or maximum demand (KW) of the industrial enterprise, and is not affected by the actual amount of electricity consumed. The per-unit price is charged based on the actual amount of electricity used by the industrial enterprise.

[0003] Since transformer capacity is the basis for charging the basic electricity price, and the basic way to obtain transformer capacity is to check the parameters marked on the transformer nameplate, some users deliberately change the parameters marked on the transformer nameplate to reduce the transformer capacity in order to pay less electricity bill. Therefore, various transformer capacity testing methods have been developed to accurately obtain the transformer capacity.

[0004] Existing methods for transformer capacity testing mainly include the loss comparison method, the impedance voltage method, and the three-element method. The loss comparison method primarily determines the actual capacity of the transformer by measuring its no-load loss and load loss and comparing them with the specified values ​​in national standards. The impedance voltage method calculates the transformer capacity by measuring the short-circuit impedance and combining it with the impedance voltage. The three-element method first calculates the transformer capacity using the impedance voltage method, and then makes a comprehensive judgment based on the no-load loss and load loss at that capacity. The no-load loss is mainly measured through a transformer no-load test, where a sinusoidal waveform, rated voltage at the rated frequency is applied to the low-voltage winding, and the no-load loss is measured with the other windings open-circuited. The load loss is measured through a transformer short-circuit test, where a rated current at the rated frequency is applied to the high-voltage winding, and the transformer load loss is measured with the other windings short-circuited.

[0005] However, existing methods for transformer capacity testing all have some shortcomings. For example, the loss comparison method requires applying the rated current when measuring load loss offline. If the transformer nameplate is replaced, the applied current may not be the rated current, leading to inaccurate load loss measurements. This makes it impossible to determine the capacity when comparing with national standards, and may even require multiple repeated measurements, resulting in low efficiency, low accuracy, and high time costs. The impedance voltage method requires ensuring that the rated voltage on the high-voltage side of the transformer and the impedance voltage value indicated on the nameplate are accurate and valid, making its application conditions quite strict. This method is not suitable when the transformer nameplate parameters have been tampered with. The loss comparison method, impedance voltage method, and three-element method all require knowledge of the transformer's rated current when obtaining load loss parameters, and tampering with the transformer nameplate parameters will lead to inaccurate load loss measurements.

[0006] Furthermore, the existing methods mentioned above all require manual comparison with national standard values ​​when determining capacity. Different energy efficiency series of transformers correspond to different national standard values. The different standard values ​​of different energy efficiency series lead to ambiguity in the no-load loss and load loss parameters of adjacent capacity levels. This results in a slow, inefficient and inaccurate process for judging transformer capacity. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method for detecting the rated capacity of a transformer.

[0008] The method for detecting the rated capacity of the transformer includes:

[0009] S1: Construct a transformer rated capacity detection model;

[0010] S2: Obtain test samples from the transformer nameplate, including but not limited to the transformer's rated capacity, no-load loss, load loss, and short-circuit impedance parameters;

[0011] S3: Input the obtained test samples into the transformer rated capacity detection model;

[0012] S4: Compare the capacity value output by the transformer rated capacity detection model with the rated capacity value of the transformer. If the comparison result is consistent, the output capacity value is considered to be the true capacity value; if the comparison result is inconsistent, proceed to step S5.

[0013] S5: Return to step S2, take the capacity value output by the transformer rated capacity detection model as the new transformer rated capacity value, re-acquire the transformer no-load loss, load loss and short-circuit impedance parameters, and continue to perform transformer rated capacity detection until the output capacity value of the transformer rated capacity detection model is equal to the transformer rated capacity value input to the capacity detection model.

[0014] In step S1, the steps for constructing the transformer rated capacity detection model include:

[0015] S1-1: Determine the initial training set for the capacity detection model. This initial training set contains M samples, and each sample contains T transformer parameter features.

[0016] S1-2: Randomly select N subsets of training data from the initial training set using the bagging method with replacement. Each subset contains the same number of samples, which is less than the number of samples in the initial training set.

[0017] S1-3: Use the random forest algorithm to construct a decision tree for each sub-training set. In the process of constructing each decision tree, randomly select t transformer parameter features as node fields, where t≤T;

[0018] S1-4: The voting method is used to output the voting results for the random forest model composed of multiple decision trees.

[0019] In step S1-1, the transformer parameter characteristics include, but are not limited to, no-load loss, load loss, short-circuit impedance, rated capacity, capacity level, and energy efficiency series.

[0020] Wherein: In steps S1-2, the number of samples contained in each sub-training set is 75% of the number of samples contained in the initial training set.

[0021] In step S1-3, no-load loss, load loss and short-circuit impedance parameters are selected as node fields.

[0022] In step S1-4, the voting method is to use the principle of majority rule to determine the voting result; when the output values ​​of each decision tree are different, a result value is randomly selected as the voting result output.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] This invention provides a transformer rated capacity detection method that fully utilizes the advantages of the random forest method, such as fast training speed and strong generalization ability. Addressing the problem of inaccurate rated capacity and load loss characteristics due to nameplate parameter tampering, this method employs a multi-iteration judgment process for capacity detection, effectively improving the speed and accuracy of transformer capacity detection. Results show that this method is unaffected by nameplate parameter tampering, achieving a capacity detection accuracy of 96.6%, significantly improving accuracy compared to manual table lookup methods. Furthermore, the automatic capacity detection ensures fast judgment speed. This invention also greatly improves the problem of low accuracy and slow judgment speed caused by the blurred boundaries of technical parameters between adjacent capacity levels due to different standard values ​​for different energy efficiency series. Attached Figure Description

[0025] Figure 1 A flowchart illustrating the workflow of the transformer rated capacity detection model based on the random forest algorithm provided by this invention.

[0026] Figure 2 The flowchart of a transformer rated capacity detection method provided by the present invention. Detailed Implementation

[0027] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Figure 2 This is a flowchart illustrating a transformer rated capacity detection method provided by the present invention. Figure 2 As shown, the transformer rated capacity detection method provided by the present invention includes:

[0029] S1: Construct a transformer rated capacity detection model.

[0030] Figure 1 The flowchart illustrates the workflow of the transformer rated capacity detection model built using the random forest algorithm provided by this invention. The specific process of building the transformer rated capacity detection model includes the following steps:

[0031] S1-1: Determine the initial training set for the capacity detection model. This initial training set contains M samples, each containing T transformer parameter features. The samples are various transformers and the information recorded on their nameplates. Transformer parameter features include, but are not limited to, no-load loss, load loss, short-circuit impedance, rated capacity, capacity rating, and energy efficiency series. During training, no-load loss, load loss, and short-circuit impedance are preferably used as judgment features, while capacity rating can be used as a label feature.

[0032] S1-2: N subsets of training data are randomly selected from the initial training set using the bagging method with replacement. Each subset contains the same number of samples, but less than the number of samples in the initial training set. Randomly selecting subsets with replacement means randomly selecting L samples from the initial training set containing M samples using the bagging method to form subsets. The selected L samples are returned to the initial training set, rather than the initial training set decreasing by L samples after the selection. Here, L < M, and the number of samples L in the subset is approximately 75% of the original training set's sample size M.

[0033] S1-3: A CART (Classification and Regression Tree) decision tree is constructed for each sub-training set using the random forest algorithm. During the construction of each CART decision tree, t (t≤T) transformer parameter features are randomly selected as node fields. Here, no-load loss, load loss, and short-circuit impedance parameters are preferably used as node fields.

[0034] The advantage of using the random forest algorithm to construct decision trees is that each decision tree is unique. This invention involves two random processes in constructing the capacity detection model: first, random sampling with replacement is used when drawing the training set. This process results in each decision tree having the same number of samples in its training set, but different sample types. The second random process involves randomly selecting feature parameters from the training set for training. This process results in each training set having the same number of samples, but different sample types, and each sample having different selected feature parameters. These two processes, to a certain extent, determine the uniqueness of each decision tree.

[0035] Here, the number of sub-training sets can be selected as 500 or more, and the no-load loss, load loss and short-circuit impedance parameters can exist in multiple node fields. For example, suppose there are 200 decision tree node fields that record the no-load loss parameter.

[0036] These nodes serve as feature parameters for determining the capacity level of the transformer sample. Each CART decision tree derives a capacity value based on the input data. Different decision trees randomly select which feature parameter to use as the decision factor, resulting in different capacity values. When numerous decision trees are combined, various feature combinations are generated, thus evaluating the sample from multiple dimensions. This ensures the algorithm's generalization ability and improves the accuracy of capacity determination.

[0037] Furthermore, in constructing CART decision trees for each sub-training set, this invention, because it uses the Random Forest algorithm, does not perform the "pruning" operation common in other algorithms. This pruning involves evaluating the priority of all features of the object and removing feature nodes with minimal impact on the result from subtree growth (i.e., no longer using those features for depth-based decision-making). The Random Forest algorithm utilizes a large number of decision trees, and unlike traditional decision trees that utilize all feature nodes of the sample, it randomly selects a small number of feature points for decision-making. This simplifies the decision model and saves time. Simultaneously, the multiple decision trees and the random selection of feature points without pruning effectively ensure the generalization ability of the Random Forest.

[0038] S1-4: The voting method is used to output the voting results for the random forest model composed of multiple decision trees.

[0039] Each decision tree outputs a result value. The voting method uses the majority rule to determine the outcome. When all decision trees output different results, a result value is randomly selected as the vote. That is, if N decision trees output N results, the result value with the most identical values ​​is output as the vote. For example, suppose there are 10 decision trees each outputting 10 results. If two results are identical and the other eight are different, the two identical values ​​are selected as the vote. In the extreme case where all 10 decision trees output 10 different results, a result value is randomly selected as the vote.

[0040] S2: Obtain test samples from the transformer nameplate. Test samples include, but are not limited to, transformer rated capacity, transformer no-load loss, load loss, and short-circuit impedance parameters.

[0041] In the initial state, the rated capacity of the transformer is taken as the rated capacity marked on the transformer nameplate.

[0042] Here, the methods for obtaining transformer no-load loss, load loss, and short-circuit impedance parameters are mainly through offline measurements, typically obtained by reading wattmeters. For example, transformer no-load loss can be measured through a no-load test, i.e., applying a sinusoidal waveform, rated voltage at rated frequency, to the low-voltage winding while measuring the no-load loss with other windings open-circuited. Load loss and short-circuit impedance are measured through a short-circuit test, i.e., applying a rated current at rated frequency to the high-voltage winding while measuring the load loss with other windings short-circuited; similarly, short-circuit impedance is measured under short-circuit conditions.

[0043] The rated voltage needs to be applied during no-load testing, while the rated current needs to be applied during load testing. When the rated current value is not given on the transformer nameplate, it can be calculated from the rated capacity and rated voltage given on the nameplate.

[0044] S3: Input the obtained test sample into the transformer rated capacity detection model. The test sample contains the obtained transformer rated capacity, transformer no-load loss, load loss and short-circuit impedance parameter values. In other words, the obtained transformer rated capacity, transformer no-load loss, load loss and short-circuit impedance parameter values ​​are input into the transformer rated capacity detection model.

[0045] S4: Compare the capacity value output by the transformer rated capacity detection model with the rated capacity value of the transformer. If the comparison results are consistent, the output capacity value is considered to be the true capacity value; if the comparison results are inconsistent, proceed to step S5.

[0046] The transformer rated capacity detection model calculates a transformer capacity value based on the input transformer no-load loss, load loss, and short-circuit impedance parameters. This capacity detection model compares its calculated transformer capacity value with the previously input rated transformer capacity value. If the comparison results match, the output capacity value is considered the true capacity value; if the comparison results do not match, step S5 is executed. In other words, when the input transformer rated capacity value is the rated capacity value on the transformer nameplate, if the comparison results match, it means the transformer nameplate has not been tampered with; if the comparison results do not match, it means the transformer nameplate has been tampered with. In this case, through the iterative detection process provided by this method, the true capacity value can finally be obtained.

[0047] S5: Return to step S2, take the capacity value output by the transformer rated capacity detection model as the new transformer rated capacity value, re-acquire the transformer no-load loss, load loss and short-circuit impedance parameters, and continue to perform transformer rated capacity detection until the output capacity value of the transformer rated capacity detection model is equal to the transformer rated capacity value input to the capacity detection model.

[0048] Here's an explanation of steps 4 and 5. When the rated capacity value marked on the transformer nameplate is altered (i.e., not the true value), the capacity value output by the transformer rated capacity detection model will be inconsistent with the input rated capacity value. The process returns to step 2, where the capacity value output by the capacity detection model is used as the transformer's rated capacity value (new rated capacity value), replacing the original value. The transformer's rated capacity, no-load loss, load loss, and short-circuit impedance parameters are then re-acquired. When a load test is performed, the rated current is applied and calculated based on the new rated capacity value. Continuing with step 3, the new rated capacity value, the re-acquired no-load loss, load loss, and short-circuit impedance parameters are input into the transformer rated capacity detection model. The output capacity value of the model is compared with the new rated capacity value, thus performing iterative detection until the output capacity value of the model equals the input rated capacity value. In this way, even if the rated capacity value marked on the transformer nameplate is altered to a non-true value, the true rated capacity value of the transformer can be obtained through multiple iterative detections provided by this invention.

[0049] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection defined by the claims submitted herein.

Claims

1. A method for detecting the rated capacity of a transformer, comprising: S1: Construct a transformer rated capacity detection model; S2: Obtain test samples from the transformer nameplate. The test samples include: transformer rated capacity, transformer no-load loss, load loss, and short-circuit impedance parameters. S3: Input the obtained test samples into the transformer rated capacity detection model; S4: The transformer rated capacity detection model calculates a transformer capacity value based on the input transformer no-load loss, load loss and short-circuit impedance parameter values. It compares the capacity value output by the transformer rated capacity detection model with the transformer rated capacity value input in step S3. If the comparison result is consistent, the output capacity value is considered to be the true capacity value; if the comparison result is inconsistent, step S5 is executed. S5: Return to step S2, take the capacity value output by the transformer rated capacity detection model as the new transformer rated capacity value, re-acquire the transformer no-load loss, load loss and short-circuit impedance parameters, and continue to perform transformer rated capacity detection. When performing a load test, a rated current needs to be applied. Calculate the rated current value based on the replaced new rated capacity value. Input the replaced new rated capacity value, the re-acquired transformer no-load loss, load loss and short-circuit impedance parameters into the transformer rated capacity detection model, and compare whether the capacity value output by the transformer rated capacity detection model is consistent with the input new rated capacity value of the transformer. This iterative detection continues until the output capacity value of the transformer rated capacity detection model is equal to the transformer rated capacity value input into the transformer rated capacity detection model, and finally, the true rated capacity value of the transformer is obtained.

2. The transformer rated capacity testing method according to claim 1, wherein: In step S1, the steps for constructing the transformer rated capacity detection model include: S1-1: Determine the initial training set for the capacity detection model. This initial training set contains M samples, and each sample contains T transformer parameter features. S1-2: Randomly select N subsets of training data from the initial training set using the bagging method with replacement. Each subset contains the same number of samples, which is less than the number of samples in the initial training set. S1-3: Use the random forest algorithm to construct a decision tree for each sub-training set. In the process of constructing each decision tree, randomly select t transformer parameter features as node fields, where t≤T; S1-4: The voting method is used to output the voting results for the random forest model composed of multiple decision trees.

3. The transformer rated capacity testing method according to claim 2, wherein: In step S1-1, the transformer parameter characteristics include, but are not limited to, no-load loss, load loss, short-circuit impedance, rated capacity, capacity class, and energy efficiency series.

4. The transformer rated capacity testing method according to claim 2, wherein: In steps S1-2, each sub-training set contains 75% of the number of samples contained in the initial training set.

5. The transformer rated capacity testing method according to claim 2, wherein: In steps S1-3, no-load loss, load loss, and short-circuit impedance parameters are selected as node fields.

6. The transformer rated capacity testing method according to claim 2, wherein: In steps S1-4, the voting method is to use the principle of majority rule to determine the voting result; when the output values ​​of each decision tree are different, a result value is randomly selected as the voting result output.

Images