Transformer performance evaluation method based on data analysis and application

Abstract

The invention relates to the technical field of transformer performance evaluation, and particularly discloses a transformer performance evaluation method based on data analysis and application, and the method comprises the following steps: grouping transformers based on the change condition of environmental factors, and marking the transformers with the same transformation mode in the group as target transformers; obtaining a time point t1 at which the target transformer starts to be transformed and a time point t2 at which the transformation is completed, and setting a plurality of first time points and second time points based on the time point t1 and the time point t2; and obtaining a first score and a second score, classifying the target transformer based on the first score and the second score, determining a representative transformer in a single classification, and evaluating the reconstruction feasibility of the target transformer based on the first score and the second score of the representative transformer. According to the invention, the feasibility of transformer transformation can be evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer performance evaluation, and particularly relates to a method and application for evaluating transformer performance based on data analysis. Background Art

[0002] A transformer is an electrical device that uses the principle of electromagnetic induction to convert AC voltage and current proportionally, mainly consisting of an iron core and a winding. When an alternating current is applied to the primary winding, a changing magnetic flux is generated, which is conducted through the iron core to the secondary winding, inducing a corresponding voltage thereon, thereby achieving the function of stepping up or stepping down the voltage.

[0003] In order to meet the requirements of different voltage regulation and operating conditions, the transformer can be modified. Common modification methods include: 1) Rewinding the original no-load tap winding or adding an on-load tap changer, determining the number of turns, wire diameter, and structure of each tap according to calculations, and reasonably leading out the tap leads, thereby achieving on-line voltage regulation and fine-tuning the output voltage without shutting down; 2) Drilling holes, welding, and sealing an on-load tap changer on the side plate of the oil tank or the oil conservator, simultaneously machining the tank size and sealing structure as needed, and installing a special oil tank, thereby providing a reliable mechanical support and good cooling and sealing environment for the on-load voltage regulation device and extending the life of the switch; 3) Selecting an on-load tap changer model with a step difference ≤ 2.5% and an appropriate rated current according to the voltage level and rated capacity of the transformer, and performing insulation and withstand voltage tests on the overall system to ensure voltage regulation accuracy and electrical safety, so that the modified transformer can operate stably and reliably under various loads and conditions.

[0004] However, the modification of the transformer is not necessarily feasible: If the design of the winding turns or wire diameter is improper, it may cause resonance, local overheating of the iron core, or even increase the no-load loss; If the machining or welding of the oil tank is not tightly sealed, it often leads to oil leakage or moisture absorption, reducing the insulation performance and increasing the breakdown risk; If the selection of the tap changer model is incorrect or the installation is inaccurate, there may be poor contact, overload heating, resulting in malfunction of the tap operation, or even damage to the switch structure. In the above cases, a series of problems such as increased noise, decreased efficiency, and increased temperature rise will occur during the operation of the originally excellent transformer, and in severe cases, it may even shorten the equipment life. Therefore, how to evaluate the feasibility of transformer modification has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and application for evaluating transformer performance based on data analysis to solve the above technical problems.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for evaluating transformer performance based on data analysis includes the following steps:

[0008] Collect the environmental factors around the transformer, group the transformers based on the changes in the environmental factors, and for the transformers in the same group, mark the transformers with the same transformation method as the target transformers;

[0009] Obtain the start time point t1 and the completion time point t2 of the transformation of the target transformer, set the first period and the second period based on the time point t1 and the time point t2, and set a number of first time points and second time points within the first period and the second period respectively;

[0010] Obtain the performance scores of the target transformer at the first time point and the second time point, denoted as the first score and the second score respectively. Classify the target transformers based on the first score and the second score. For a single classification, determine the representative transformer based on the first score and the second score of the target transformers in the classification, and evaluate the transformation feasibility of the target transformer based on the first score and the second score of the representative transformer.

[0011] Preferably, grouping the transformers includes:

[0012] Determine the target factors in the environmental factors that have an impact on the transformer performance exceeding a preset value based on deep learning technology, preprocess the target factors, and the preprocessing includes cleaning, denoising, and normalization, and fill in the missing values based on the predicted values of the machine learning algorithm;

[0013] Construct an n + 1 - dimensional coordinate system, where n represents the types of target factors, and generate coordinate points (A1, A2,..., A n , S), A n represents the nth preprocessed target factor, and S represents the time point for collecting environmental factors;

[0014] Fit the coordinate points to obtain a fitting curve. In the same group, the similarity degree between the fitting curves corresponding to any two transformers is greater than a preset first similarity degree threshold.

[0015] Preferably, setting the first time point and the second time point includes:

[0016] Set the first period [t1 - T, t1] and the second period [t2, t2 + T], where T represents a preset duration;

[0017] Starting from the starting points of the first period and the second period respectively, select a number of time points within the first period and the second period at a preset time interval, and denote them as the first time point and the second time point respectively.

[0018] Preferably, obtaining the performance score of the target transformer includes:

[0019] Collect the operation data of the target transformer, and perform a first processing on the operation data. The first processing includes cleaning, denoising, and normalization processing;

[0020] Apply signal processing and machine learning algorithms to extract the influencing parameters from the operation data after the first processing. The influencing parameters are characteristic parameters that affect the performance state of the transformer;

[0021] Build a transformer performance evaluation model based on the influencing parameters, input the current influencing parameters into the transformer performance evaluation model, and output the performance score of the target transformer.

[0022] Preferably, determining the representative transformer includes:

[0023] Draw a curve A1 of the first score changing with time and a curve A2 of the second score changing with time;

[0024] In the same classification, the first similarity between any two target transformers is greater than a preset second similarity threshold and the second similarity between any two target transformers is greater than a preset second similarity threshold. The first similarity represents the similarity degree between the curves A1, and the second similarity represents the similarity degree between the curves A2;

[0025] Calculate the selection value of the target transformer i , p1 j and p2 j respectively represent the first similarity and the second similarity between the target transformer i and the target transformer j, and m represents the total number of target transformers other than the target transformer i in the classification;

[0026] Take the target transformer corresponding to the maximum selection value as the representative transformer.

[0027] Preferably, evaluating the transformation feasibility of the target transformer includes:

[0028] Respectively draw a curve d1 of the first score changing with time and a curve d2 of the second score changing with time of the representative transformer. Let the time points corresponding to the starting points of the curve d1 and the curve d2 be the origin to obtain new curves, which are respectively denoted as curve D1 and curve D2;

[0029] Obtain curve D3 = D2 - D1, count the proportion X of the domain of the part of curve D3 below the x-axis in the domain of curve D3, and calculate the evaluation score , where N represents the number of intersections of curve D3 and the x-axis.

[0030] Preferably, evaluating the transformation feasibility of the target transformer further includes:

[0031] Set an evaluation score threshold K1;

[0032] If the evaluation score K ≥ K1, it is determined that the retrofit feasibility of all target transformers in the classification is poor; otherwise, it is determined that the retrofit feasibility of all target transformers in the classification is good.

[0033] An application of the transformer performance evaluation method based on data analysis as described in a running transformer, in which a remote central computer collects the running data of the transformer through SCADA and applies it to the evaluation of the retrofit feasibility of the transformer.

[0034] Advantages of the present invention: Compared with the prior art:

[0035] 1) Firstly, the present invention performs high-dimensional fitting on the collected environmental factor curves and automatically groups the transformers according to the curve similarity, so that the devices in the same group are in almost the same environmental background, thereby shielding the influence of environmental factors. This is because even if the running data is all the same, different performance evaluation results may be generated due to different environmental factors. Subsequently, only the trend of the performance scores before and after the retrofit is compared within the same group, completely isolating the masking and amplifying effects of environmental differences on the results, ensuring that the analysis conclusion truly reflects the influence of retrofit measures such as winding design, tank sealing, and tap-changer selection. With the help of this mechanism, the operation and maintenance personnel can obtain a more objective, repeatable, and traceable evaluation basis, avoiding the risk of misjudging the retrofit feasibility due to environmental factors from the source.

[0036] 2) The present invention selects the most representative devices in each category, and then only needs to perform in-depth curve difference and trend analysis on these representative transformers to infer the retrofit effects of all devices in the same category. This strategy significantly reduces the number of data cleaning, feature extraction, and curve comparison, shortens the overall evaluation cycle, and avoids the system load caused by repeated operations. At the same time, the operation and maintenance team can quickly complete the screening of the retrofit feasibility of large-scale transformers with limited human and computing resources, realizing an efficient and economical intelligent decision-making process.

[0037] 3) The present invention aligns the performance score curves of the representative transformers before and after the retrofit by time and generates a difference curve to visually present the gain or deterioration brought by the retrofit. When the difference curve generally tends to rise and the fluctuation is stable, it indicates that the retrofit plan has a positive contribution to efficiency, temperature rise, and noise control; if the difference is in a long-term downward trend or violent oscillation, it indicates that there are potential problems in the design or process. The operation and maintenance personnel can quickly identify the retrofit reliability and potential risks from the curve shape without additional disassembly and inspection, and then decide to promote, optimize, or terminate the subsequent work, realizing precise pre-control and reducing operation and maintenance risks and economic losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below with reference to the accompanying drawings.

[0039] Figure 1 It is a schematic flow chart of a method for evaluating the performance of a transformer based on data analysis according to the present invention. Specific embodiments

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0041] Please refer to Figure 1 As shown, the present invention is a method for evaluating the performance of a transformer based on data analysis, including the following steps:

[0042] Deploy multiple types of sensors on site to collect environmental factors around the transformer in real time, including environmental temperature, relative humidity, altitude pressure, etc. After synchronizing the time of these original data streams, they are used for subsequent grouping of the transformer. After the grouping is completed, the system reads the renovation records in the operation and maintenance management platform, marks the equipment with rewound windings and the tap changer replaced with a solid-sealed type as one type of target transformer, and marks the equipment with only the oil tank replaced and an online monitoring interface added as another type of target transformer, forming clear labels for the same renovation method within the group (only these three renovation methods are taken as examples here, and other renovation methods will not be listed one by one).

[0043] It should be noted that the interference of the natural environment is shielded in advance, so that the subsequent performance comparison really focuses on the renovation measures themselves; after grouping, the equipment with the same renovation method is uniformly marked, which can avoid repeated retrieval in large-scale data, help quickly allocate computing resources and shorten the evaluation process, and at the same time lay a foundation for the selection of subsequent representative samples. Ultimately, it can help the operation and maintenance team accurately judge the actual value of different renovation plans in a complex background, reduce decision-making risks and improve management efficiency.

[0044] A preferred embodiment of the present invention for grouping transformers includes:

[0045] Retrieve historical operation and maintenance records, collect multi-dimensional environmental curves such as temperature and humidity, salt fog concentration, partial discharge noise, etc. and synchronous performance indicators. After slicing by time window, input them into a convolutional neural network integrated with an attention mechanism model calibrated in the laboratory. Extract the contribution degree of each dimension to performance fluctuations through gradient inversion, and automatically register the dimensions with contribution degrees exceeding the preset threshold as target factors. Subsequently, for these target factors, call the built-in anomaly detection module of the edge gateway to label the anomaly points, and then use sliding median filtering and wavelet soft threshold denoising to clean burrs and high-frequency noise, and normalize the calibrated sensor signals to zero mean unit variance per hour. If the data is missing in a certain period, use the front and back sequences of the same factor as input, and use the trained bidirectional gated recurrent network to predict the reasonable value of the missing measurement position in real time and supplement it to the time series library. After preprocessing, the system constructs a coordinate system with the number of target factors n plus one-dimensional time stamp, and maps the normalized temperature and humidity A1, salt fog index A2, etc. at the same sampling moment into points (A1, A2,..., S). For all coordinate points of a single transformer within the selected observation period, use cubic B-spline fitting to generate a smooth curve, and calculate the cosine similarity after aligning the curves with the dynamic time warping algorithm. Group the transformers with similarity higher than the threshold into the same group.

[0046] It can be understood that such processing can automatically lock the environmental dimension that can best explain the performance difference among a large number of measurement points, avoiding interference from irrelevant features on the evaluation results; systematic cleaning, denoising and deep prediction for filling values ensure the integrity and consistency of the input data, providing a reliable basis for subsequent modeling; after expanding to an n+1-dimensional space with time coordinates and then performing curve fitting, the discrete monitoring sequence can be transformed into an object with stable geometric shape and easy to measure. Using a high similarity threshold for screening within the group further ensures that the internal environmental background of the group is almost the same, helping the evaluation process to eliminate the heterogeneity of external factors at an early stage, and finally enabling the performance difference before and after transformation to truly focus on technical measures such as winding rewinding, tank sealing or tap changer selection itself, providing a solid and credible objective basis for the conclusion of the transformation feasibility.

[0047] Obtain the start time point t1 and the completion time point t2 of the transformation of the target transformer, set the first period and the second period based on the time point t1 and the time point t2, and set a number of first time points and second time points within the first period and the second period respectively.

[0048] In another preferred embodiment of the present invention, setting the first time point and the second time point includes:

[0049] In the asset management platform, call the API to retrieve the renovation work order for each target transformer, directly read the "Construction Start" and "Completion Acceptance" fields, and parse the dates into time points t1 and t2. The project team sets T to one month in the parameter configuration, and the system automatically generates the first cycle [t1-30d, t1] and the second cycle [t2, t2+30d] according to the length of 30 days. For example, if the renovation date is 2023-04-01 to 2023-08-15, the first cycle is 2023-03-02 to 2023-04-01, and the second cycle is 2023-08-15 to 2023-09 -14; then the system evenly slices the time axis according to the monitoring granularity: if routine monitoring is once every half day, then in the first cycle, time tags are written every half day from 2023-03-02 to 2023-04-01, forming the first time point sequence. Similarly, in the second cycle, from 2023-08-15, the same interval is calculated to 2023-09-14 to obtain the second time point sequence; the system writes the two time tags together with the transformer ID into the task queue. The subsequent scoring model uses this to pull the same amount and the same interval of operation, electrical inspection and environmental data to ensure that symmetrical samples of a full month are taken before and after the transformation.

[0050] It should be noted that the use of a one-month window can filter out sudden short-term fluctuations and avoid the load and climate mismatch caused by cross-seasonal factors, so that the comparison before and after the transformation can focus on the process changes of the equipment itself; the shortened cycle makes the data volume more streamlined, and the computing resource usage and storage pressure are reduced accordingly, which facilitates the rapid issuance of results during batch evaluation; the unified thirty-day interval and fixed sampling rhythm allow transformers of different regions and types to be on the same time scale, reducing sample voids caused by differences in operation and maintenance cycles, and improving the structural consistency of the feature matrix, thereby providing a stable and repeatable input basis for subsequent performance scoring and transformation feasibility judgment.

[0051] Performance scores of the target transformer are obtained at a first time point and a second time point, respectively recorded as a first score and a second score. The target transformer is classified based on the first score and the second score. For a single classification, a representative transformer is determined based on the first score and the second score of the target transformer in the classification. The feasibility of transformation of the target transformer is evaluated based on the first score and the second score of the representative transformer.

[0052] In another preferred embodiment of the present invention, obtaining the performance score of the target transformer includes:

[0053] After synchronously capturing the original operation data such as load current, phase voltage, oil temperature, winding hot spot temperature, gas concentration in oil, partial discharge level, vibration acceleration, etc. through the on-site side online monitoring terminal and the SCADA link, the system first filters out the invalid channels according to the health identification carried by the sensors, then uses the sliding window H-ampel detection to eliminate the outliers. For the short-term missing report segments, the time series K-NN interpolation is called. After that, wavelet soft threshold denoising is used to suppress the high-frequency spikes and multi-point linear interpolation is used to align the timestamps. Finally, z-score normalization is implemented according to each range; immediately, the cleaned multi-channel signals are fed into the signal processing pipeline: the fast Fourier transform is performed on the current and voltage to extract the fundamental wave amplitude and harmonic content, the sliding mean square gradient is calculated for the temperature sequence to capture the heat dissipation change, the continuous wavelet transform is used for the partial discharge and vibration sequences to obtain the energy density distribution, and statistical indicators such as peak value, kurtosis, and spectral entropy are extracted to form the influence parameter vector describing the winding, core, insulation, and mechanical states; subsequently, a gradient boosting tree-gated recurrent network hybrid model is established based on the historical working conditions and maintenance records. First, the decision tree partitions the static features, and then the recurrent network aggregates the dynamic features to output the comprehensive score. After the model is offline trained, it is packaged as a service and deployed on the edge server. Whenever a new batch of influence parameters is generated, it is immediately pushed to the model endpoint, and the current performance score of the target transformer is returned and written into the time series database for subsequent evaluation and call;

[0054] It can be understood that such processing can eliminate the interference caused by instrument failures and environmental noises at the data entry stage, keeping the input structurally complete and temporally consistent; the systematic feature extraction compresses the original signals with a large number of dimensions into a small number of parameters with clear physical meanings, reducing the complexity of subsequent algorithms and providing an intuitive basis for expert review; the machine learning model integrates static working conditions and dynamic trends, and can keep the evaluation criteria consistent under different brands and different load scenarios, improving the objectivity and transferability of the evaluation; the finally output performance score serves as a unified metric for comparison before and after subsequent transformation and feasibility judgment, helping the operation and maintenance team quickly locate performance hidden dangers and process optimization spaces in the decision-making process;

[0055] A preferred embodiment of the present invention, determining the representative transformer includes:

[0056] Under the same classification, the system first assembles the first scores and the second scores of each target transformer into sequences of equal length in chronological order, and calls the data visualization component to draw two broken lines at once: A1 takes the first time point to the end, and A2 takes the second time point to the end. The horizontal axis uniformly uses the relative number of days from the renovation reference date to ensure curve alignment. Subsequently, the dynamic time warping algorithm is used to perform elastic registration on the A1 curves of the two transformers, and then the cosine similarity of the registered sequences is calculated as the first similarity. Similarly, the second similarity is obtained by repeating the operation on the A2 curves. If both the first similarity and the second similarity between the two curves exceed the second similarity degree threshold determined by the empirical method in the training stage, these two transformers are retained in the current classification. In this way, all combinations are traversed and the similarity matrix is output. Then, the system sums the i-th row in the matrix, and accumulates p1 j and p2 j pair by pair to obtain the selection value. After looping through all i, the row with the largest selection value is found, and the associated transformer is written into the database and marked as the representative transformer. Its ID is recorded together with the selection value for subsequent performance difference and renovation feasibility evaluation calls;

[0057] It should be noted that a strict homogeneity screening mechanism is established using the similarity of the dual-channel scoring curves to keep the operation performance within the classification highly consistent. On this basis, the device that can best comprehensively represent the overall curve shape is selected through the sum maximization strategy. The subsequent in-depth analysis of one device can replace the repeated calculation of the entire class of devices, thereby reducing the computing power consumption and data scheduling volume during evaluation. At the same time, it ensures that the state change of the representative device can accurately map the actual renovation effect of the entire class of transformers, providing reliable sample support for the subsequent conclusion of the rise or fall of the difference curve, and helping the operation and maintenance personnel to efficiently obtain reliable feasibility judgments in the large-scale asset management scenario.

[0058] Another preferred embodiment of the present invention for evaluating the renovation feasibility of the target transformer includes:

[0059] Retrieve from the database the complete scoring curves d1 and d2 formed by the transformer at the first and second time point sequences, and synchronize the two curves at the completion date t2 of the transformation: remap the time stamp corresponding to t2 to t = 0, and subtract this time stamp from each of the remaining sampling moments, so that the two curves are aligned only on the horizontal axis without changing the vertical values; denote the curves after horizontal axis synchronization as D1 and D2 respectively, and then subtract them point by point to form the difference curve D3 = D2 - D1; the system scans the sign changes along D3, determines the intersection positions with the x-axis and the cumulative interval lengths for each interval where D3 < 0 through linear interpolation, divides by the total observation duration to obtain the proportion X, and simultaneously counts the number of intersections N of the entire D3 with the x-axis; finally, synthesize the evaluation score K according to the weighted formula in the configuration file, and compare it with the threshold K1 to label all target transformers within the classification as "good feasibility" or "poor feasibility";

[0060] Align the time axis rather than the vertical values, so that the two scoring curves maintain their original amplitudes, and can truthfully reflect the absolute gap in scores before and after the transformation; the difference curve D3 directly presents the increase and decrease relationship between the two curves at the same time node. The negative area proportion X captures the persistence of the score after the transformation being lower than that before the transformation, and the number of intersections N depicts the frequency of fluctuations. The synthesized evaluation score takes both amplitude and stability into account; use a unified threshold to convert the continuous signal analysis result into an executable binary label, which can quickly locate the transformation batches that need to be reviewed or optimized in a large-scale asset management scenario, providing a clear and reliable basis for operation and maintenance decisions.

[0061] An application of a transformer performance evaluation method based on data analysis in an operating transformer, which uses a remote central computer to collect the operating data of the transformer through SCADA and applies it to the evaluation of the feasibility of transformer transformation.

[0062] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0063] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the present invention.

Claims

1. A transformer performance evaluation method based on data analysis, characterized in that It includes the following steps: Collect the environmental factors around the transformer, group the transformers based on the changes in the environmental factors. For the transformers in the same group, mark the transformers with the same transformation method as the target transformers; Obtain the start time point t1 and the completion time point t2 of the transformation of the target transformers, set the first period and the second period based on the time point t1 and the time point t2, and set a number of first time points and second time points within the first period and the second period respectively; Obtain the performance scores of the target transformers at the first time point and the second time point, denoted as the first score and the second score respectively. Classify the target transformers based on the first score and the second score. For a single classification, determine the representative transformer based on the first score and the second score of the target transformers in the classification, and evaluate the transformation feasibility of the target transformers based on the first score and the second score of the representative transformer.

2. The method for evaluating the performance of a transformer based on data analysis according to claim 1, wherein Grouping the transformers includes: Determine the target factors in the environmental factors that have an impact on the transformer performance exceeding a preset value based on deep learning technology, perform preprocessing on the target factors, and the preprocessing includes cleaning, denoising and normalization, and fill in the missing values based on the predicted values of the machine learning algorithm; Construct an n+1-dimensional coordinate system, where n represents the types of target factors, and generate coordinate points (A1, A2, …, A n , S), where A n represents the nth preprocessed target factor, and S represents the time point for collecting environmental factors; Fit the coordinate points to obtain a fitting curve. In the same group, the similarity degree between the fitting curves corresponding to any two transformers is greater than a preset first similarity degree threshold.

3. A method for evaluating the performance of a transformer based on data analysis according to claim 1, characterized in that Setting the first time point and the second time point includes: Set the first period [t1 - T, t1] and the second period [t2, t2 + T], where T represents a preset duration; Starting from the starting points of the first period and the second period respectively, select a number of time points within the first period and the second period at a preset time interval, and denote them as the first time point and the second time point respectively.

4. A method for evaluating the performance of a transformer based on data analysis according to claim 1, characterized in that, Obtaining the performance score of the target transformer includes: Collect the operation data of the target transformer, and perform the first processing on the operation data, and the first processing includes cleaning, denoising and normalization processing; Apply signal processing and machine learning algorithms to extract the influence parameters from the operation data after the first processing, and the influence parameters are the characteristic parameters that affect the performance state of the transformer; Construct a transformer performance evaluation model based on the influence parameters, input the current influence parameters into the transformer performance evaluation model, and output the performance score of the target transformer.

5. The method for evaluating the performance of a transformer based on data analysis according to claim 1, wherein Determining the representative transformer includes: Draw a curve A1 of the first score changing with time and a curve A2 of the second score changing with time; In the same classification, the first similarity degree between any two target transformers is greater than a preset second similarity degree threshold and the second similarity degree between any two target transformers is greater than a preset second similarity degree threshold. The first similarity degree represents the similarity degree between the curves A1, and the second similarity degree represents the similarity degree between the curves A2; Calculate the selection value of the target transformer i , p1 j and p2 j respectively represent the first similarity and the second similarity between the target transformer i and the target transformer j, and m represents the total number of target transformers other than the target transformer i in the classification; Take the target transformer corresponding to the maximum selection value as the representative transformer.

6. The method for evaluating the performance of a transformer based on data analysis according to claim 1, wherein, Evaluating the transformation feasibility of the target transformer includes: Draw a curve d1 of the first score changing with time and a curve d2 of the second score changing with time of the representative transformer respectively. Let the time point corresponding to the starting point of the curve d1 and the curve d2 be the origin to obtain new curves, denoted as curve D1 and curve D2 respectively; Obtain the curve D3 = D2 - D1, count the proportion X of the domain of the part of the curve D3 below the x-axis in the domain of the curve D3, and calculate the evaluation score , where N represents the number of intersections of the curve D3 with the x-axis.

7. A method for evaluating the performance of a transformer based on data analysis according to claim 6, characterized in that Evaluating the retrofit feasibility of the target transformer also includes: Setting the evaluation score threshold K1; If the evaluation score K ≥ K1, it is determined that the retrofit feasibility of all target transformers in the classification is poor; otherwise, it is determined that the retrofit feasibility of all target transformers in the classification is good.

8. Application of the transformer performance evaluation method based on data analysis according to any one of claims 1-7 in an operating transformer, characterized in that, Using the remote central computer to collect the operating data of the transformer through SCADA for use in evaluating the retrofit feasibility of the transformer.

Images