阅读说明：本技术 基于XGBoost算法的变压器故障诊断与健康评估系统及方法 (XGboost algorithm-based transformer fault diagnosis and health assessment system and method ) 是由 周光翀 王祺 谢宁 王承民 于 2021-07-08 设计创作，主要内容包括：本发明提供了一种基于XGBoost算法的变压器故障诊断与健康评估系统及方法,包括如下模块：故障诊断模型模块：分别从故障表现和故障程度两个角度出发定义故障状态,丰富故障集合,基于XGBoost算法建立故障诊断模型；健康评估模型模块：通过KMeans聚类建立状态区间,使输出的分数具有区分度,基于XGBoost算法建立健康评估模型,对处于正常和故障状态之间的变压器给出评分,通过状态区间呈现变压器健康评估结果。本发明使用XGBoost算法,故障诊断准确可靠,健康评估客观合理,有很强的可行性和有效性,对电力变压器的状态评估具有指导意义；尤其适用于对电力变压器可靠性和安全性要求较高的电力系统。(The invention provides a transformer fault diagnosis and health evaluation system and method based on an XGboost algorithm, which comprises the following modules: a fault diagnosis model module: defining fault states from two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm; a health assessment model module: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval. The XGboost algorithm is used, so that the fault diagnosis is accurate and reliable, the health evaluation is objective and reasonable, the feasibility and the effectiveness are very strong, and the XGboost algorithm has guiding significance on the state evaluation of the power transformer; the method is particularly suitable for power systems with high requirements on reliability and safety of power transformers.)

1. A transformer fault diagnosis and health assessment system based on an XGboost algorithm is characterized by comprising the following modules:

a fault diagnosis model module: defining fault states from two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm;

a health assessment model module: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval.

2. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 1, further comprising a data preprocessing module: and receiving input data obtained by the fault diagnosis model, wherein the input data comprises gas absolute content data, converting the gas absolute content data into gas ratio data through data processing, adding the gas ratio data into the fault diagnosis model to participate in fault diagnosis, and adding the total gas content into the input data of the fault diagnosis model when various special conditions that the gas ratio data are similar but the total gas amount is far away occur.

3. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 2, wherein the fault diagnosis model takes the ratio between dissolved gases in oil and the total content of gases as input data; the input data of the health assessment model comprises gas absolute content data, the health assessment model extracts the XGboost model intermediate layer probability to score the health condition of the transformer, and the health assessment result of the transformer is presented through a state interval.

4. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 1, wherein fault manifestations include heating or heat release; the fault level comprises low energy or high energy, and the state of the transformer comprises normal, low energy heating, high energy heating, low energy discharging and high energy discharging which are respectively represented by five category sets.

5. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 4, wherein for a multi-classification task, the XGboost converts the multi-classification task into a plurality of bi-classification tasks; the fault diagnosis model consists of five submodels, and each submodel outputs a score value to judge whether the input sample belongs to the category; the two classification tasks include transformer normal or fault.

6. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 1, wherein the fault diagnosis model uses back propagation for gradient boosting with a minimization loss function as an optimization objective.

7. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 1, wherein the health assessment model automatically generates explicit state intervals according to scores output by the health assessment model.

8. The XGboost algorithm-based transformer fault diagnosis and health assessment system according to claim 1, wherein the health assessment model analyzes monitoring data of dissolved gas in oil to assess the health status of the whole transformer; the health assessment model trains a classification model aiming at a two-classification task of judging whether the transformer belongs to a normal or fault state according to monitoring data of dissolved gas in the transformer oil; training a transformer fault discrimination model, and when the accuracy of the transformer fault discrimination model is high enough, learning fault characteristics by a health evaluation model; the health evaluation model analyzes fault information contained in the transformer monitoring data, the more the fault information is, the lower the probability that an output sample is positive is, the worse the health condition of the transformer is, and the probability output by the health evaluation model is adopted as the health evaluation score of the transformer.

9. A transformer fault diagnosis and health assessment method based on an XGboost algorithm is characterized by comprising the following steps:

a fault diagnosis model step: defining fault states from two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm;

a health evaluation model step: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval.

10. The XGboost algorithm-based transformer fault diagnosis and health assessment method according to claim 9, further comprising a data preprocessing step of: and receiving input data obtained by the fault diagnosis model, wherein the input data comprises gas absolute content data, converting the gas absolute content data into gas ratio data through data processing, adding the gas ratio data into the fault diagnosis model to participate in fault diagnosis, and adding the total gas content into the input data of the fault diagnosis model when various special conditions that the gas ratio data are similar but the total gas amount is far away occur.

Technical Field

The invention relates to the technical field of transformer fault diagnosis and health assessment, in particular to a transformer fault diagnosis and health assessment system and method based on an XGboost algorithm.

Background

As one of the most important electrical devices in an electrical power system, the health of a power transformer is directly related to the safe and stable operation of the power grid. At present, a power grid company mainly adopts a strategy of regular maintenance on a power transformer, and the problems of too frequent maintenance and untimely maintenance exist. Compared with a regular maintenance strategy, the maintenance cost per year can be reduced by 25-50% and the power failure time can be reduced by 75% by evaluating the implementation state of the transformer. Therefore, the state evaluation of the power transformer can not only find faults in time and improve the safe reliability of the operation of the power system, but also reduce the waste of manpower and material resources and generate better economic benefit. The state evaluation task covers subtasks of fault diagnosis, health evaluation, life prediction and the like.

The theoretical method for diagnosing the fault of the oil-immersed power transformer is a Dissolved Gas Analysis (DGA) method in oil, and the method is based on the principle that insulating oil and insulating paper can crack to release Gas when the oil-immersed power transformer has the fault, takes the content of organic Gas in the transformer oil as a research object, and calculates the Gas ratio by adopting a three-ratio method, a four-ratio method, a David triangle method and the like to obtain a corresponding fault code, so as to diagnose the fault type. However, the method has the defects of few fault codes, large diagnosis granularity and over-absolute boundary, so that some faults can not be judged or are judged fuzziness. With the development and popularization of data mining and artificial intelligence technologies, machine learning and deep learning methods are applied to the field of power transformer fault diagnosis. However, the traditional machine learning method has limited capability of fitting a function, so that the diagnosis accuracy rate is difficult to further improve; the deep learning method has strict requirements on the quantity and quality of data, has high training cost, and is easy to generate an overfitting phenomenon. The method for evaluating the health of the oil-immersed power transformer is mainly characterized in that the health index of the transformer is calculated by weighting and scoring each influence factor according to the contents of gas and furfural in oil and combining methods such as hierarchical analysis, fuzzy evaluation, evidence theory and the like. In such evaluation methods, the scores and corresponding weights of the influencing factors are assigned by individuals according to experience, and therefore are more dominant than subjective factors and less objective and accurate.

The chinese patent publication CN108414684A discloses a method for transformer state evaluation and fault diagnosis, which includes the following steps: when the transformer normally operates, acquiring the original total hydrocarbon content in dissolved gas in insulating oil; monitoring the real-time total hydrocarbon content of dissolved gas in the insulating oil in real time; analyzing the real-time total hydrocarbon content and the original total hydrocarbon content, and calculating the absolute gas production rate of the total hydrocarbons; when the real-time total hydrocarbon content and the absolute gas production rate of the total hydrocarbons reach certain limits, corresponding alarms are sent out, and diagnosis is made.

With respect to the related art in the above, the inventors consider that with the existing fault diagnosis method: the traditional machine learning method has limited capability of fitting a function, so that the diagnosis accuracy rate is difficult to further improve; the deep learning method has strict requirements on the quantity and quality of data, has high training cost, and is easy to generate an overfitting phenomenon. For the existing health assessment methods: the scores and the corresponding weights of the influencing factors are assigned by individuals according to experience, so that the scores and the corresponding weights are more dominant than subjective factors and are not objective and accurate enough.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide a transformer fault diagnosis and health evaluation system and method based on an XGboost algorithm.

The invention provides a transformer fault diagnosis and health evaluation system based on an XGboost algorithm, which comprises the following modules:

a fault diagnosis model module: defining fault states from two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm;

a health assessment model module: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval.

Preferably, the system further comprises a data preprocessing module: and receiving input data obtained by the fault diagnosis model, wherein the input data comprises gas absolute content data, converting the gas absolute content data into gas ratio data through data processing, adding the gas ratio data into the fault diagnosis model to participate in fault diagnosis, and adding the total gas content into the input data of the fault diagnosis model when various special conditions that the gas ratio data are similar but the total gas amount is far away occur.

Preferably, the fault diagnosis model takes the ratio of dissolved gases in oil and the total content of gases as input data; the input data of the health assessment model comprises gas absolute content data, the health assessment model extracts the XGboost model intermediate layer probability to score the health condition of the transformer, and the health assessment result of the transformer is presented through a state interval.

Preferably, the failure manifestation includes heating or exothermia; the fault level comprises low energy or high energy, and the state of the transformer comprises normal, low energy heating, high energy heating, low energy discharging and high energy discharging which are respectively represented by five category sets.

Preferably, for the multi-classification task, the XGboost converts the multi-classification task into a plurality of two-classification tasks; the fault diagnosis model consists of five submodels, and each submodel outputs a score value to judge whether the input sample belongs to the category; the two classification tasks include transformer normal or fault.

Preferably, the fault diagnosis model uses back propagation to perform gradient lifting with a minimization loss function as an optimization target.

Preferably, the health assessment model automatically generates an explicit state interval according to the score output by the health assessment model.

Preferably, the health assessment model analyzes monitoring data of dissolved gas in oil to assess the overall health state of the transformer; the health assessment model trains a classification model aiming at a two-classification task of judging whether the transformer belongs to a normal or fault state according to monitoring data of dissolved gas in the transformer oil; training a transformer fault discrimination model, and when the accuracy of the transformer fault discrimination model is high enough, learning fault characteristics by a health evaluation model; the health evaluation model analyzes fault information contained in the transformer monitoring data, the more the fault information is, the lower the probability that an output sample is positive is, the worse the health condition of the transformer is, and the probability output by the health evaluation model is adopted as the health evaluation score of the transformer.

The invention provides a transformer fault diagnosis and health assessment method based on an XGboost algorithm, which comprises the following steps:

a fault diagnosis model step: defining fault states from two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm;

a health evaluation model step: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval.

Preferably, the method further comprises the following data preprocessing steps: and receiving input data obtained by the fault diagnosis model, wherein the input data comprises gas absolute content data, converting the gas absolute content data into gas ratio data through data processing, adding the gas ratio data into the fault diagnosis model to participate in fault diagnosis, and adding the total gas content into the input data of the fault diagnosis model when various special conditions that the gas ratio data are similar but the total gas amount is far away occur.

Compared with the prior art, the invention has the following beneficial effects:

1. the fault state is defined from the two angles of fault expression and fault degree in the fault diagnosis model, the contained fault types are relatively comprehensive, and the diagnosis result is more accurate;

2. the XGboost algorithm is adopted in the fault diagnosis model, the diagnosis accuracy exceeding that of other traditional machine learning methods is obtained, and the method has lower training cost, good generalization and excellent interpretability;

3. in the data preprocessing module, the absolute gas content data is converted into gas ratio data to participate in fault diagnosis, so that the capability of fitting the data with a real fault function is stronger, the diagnosis granularity is finer, and the diagnosis effect is better;

4. in the data preprocessing module, the total gas content is added into the input characteristics, and the special conditions that the ratio of various gases is similar but the total amount is far from each other are considered;

5. in the health evaluation model, by introducing a state interval, the output score has discrimination, and a proper score can be given to the transformer between a normal state and a fault state;

6. the XGboost algorithm and the KMeans clustering are adopted in the health assessment model, subjectivity caused by artificial scoring and empowerment in the assessment process is eliminated, and the assessment result is more accurate and objective.

Drawings

Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:

FIG. 1 is a flow chart of a fault diagnosis based on the XGboost algorithm;

FIG. 2 is a flow chart of health assessment based on the XGboost algorithm.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.

The embodiment of the invention discloses a system and a method for diagnosing transformer faults and evaluating health based on an XGboost algorithm, as shown in figures 1 and 2, the system comprises the following modules: a fault diagnosis model module: and respectively defining fault states from the two aspects of fault expression and fault degree, enriching a fault set, and establishing a fault diagnosis model based on an XGboost algorithm. The fault diagnosis model takes the ratio of dissolved gases in oil and the total content of gases as input data. Failure manifestations include heating or exothermia; the fault level includes low energy or high energy, and the state of the transformer includes normal, low energy heating, high energy heating, low energy discharging and high energy discharging, which are respectively represented by five kinds of category sets. For multi-classification tasks, XGboost converts the multi-classification task into a plurality of bi-classification tasks. The fault diagnosis model consists of five submodels, and each submodel outputs a score value to judge whether the input sample belongs to the category. The classification task includes transformer normal or fault. The fault diagnosis model takes a minimum loss function as an optimization target and uses back propagation to carry out gradient lifting.

The XGboost-based fault diagnosis model module defines the fault state from the aspects of fault expression and fault degree, enriches the fault set, establishes a fault diagnosis model based on the XGboost algorithm, improves the diagnosis accuracy and the training efficiency, and establishes a theoretical basis for subsequent modules.

The XGboost-based fault diagnosis model module mainly analyzes the principle of the XGboost method, introduces the specific structure of the transformer fault diagnosis model, and shows the construction flow of the model by combining characters and images.

XGboost belongs to an ensemble learning method in machine learning. The core idea of the integration method comprises the following steps: after inputting a training data set and initializing an output function, carrying out iterative training on the data set by taking a reduced residual error between a model prediction result and a real label as a target, generating a plurality of classes and regression trees as weak classifiers, and finally combining the classes and the regression trees into a strong classifier. Compared with other traditional machine learning methods, the XGboost method has higher accuracy; compared with a deep learning method, the training cost is lower, and the generalization performance is better.

The fault state is defined from the two aspects of fault expression (heating or discharging) and fault degree (low energy or high energy), and the transformer has five types of states: normal, low-energy heating, high-energy heating, low-energy discharging, and high-energy discharging, which are respectively represented by five category sets. For the multi-classification task, a common processing method of XGBoost is to convert the multi-classification task into a plurality of tasks such as "is the transformer normal? "two classification tasks. Therefore, the model is composed of five submodels, and each submodel respectively outputs a scoring value for judging whether the input sample belongs to the category. The fault diagnosis model outputs the state of the power transformer, and the state comprises five types of normal, low-energy heating, high-energy heating, low-energy discharging and high-energy discharging. The fault diagnosis model is a classification model based on the XGboost algorithm.

The model takes a minimization loss function as an optimization target and uses back propagation to carry out gradient lifting.

A data preprocessing module: and receiving input data obtained by the fault diagnosis model, wherein the input data comprises gas absolute content data, converting the gas absolute content data into gas ratio data through data processing, adding the gas ratio data into the fault diagnosis model to participate in fault diagnosis, and adding the total gas content into the input data of the fault diagnosis model when various special conditions that the gas ratio data are similar but the total gas amount is far away occur. The fault diagnosis model takes the ratio of dissolved gases in oil and the total content of gases as input data after pretreatment; the specific meaning of data preprocessing is that: and basically acquiring the absolute content of each gas, and selecting the ratio of 7 gases and the total content of the gases as input data of a fault diagnosis model.

The data preprocessing module based on the gas ratio method receives input data obtained by the XGboost-based fault diagnosis model module, converts absolute gas content data into gas ratio data to participate in fault diagnosis through data processing, considers special conditions that various gas ratios are close but total amounts are far apart, and adds total gas content into input characteristics.

The data preprocessing module based on the gas ratio method mainly improves the method of directly using the absolute content of gas in other machine learning diagnosis methods, and converts the absolute content data of the gas into a relative gas ratio for fault diagnosis by using a three-ratio method for reference.

The data preprocessing process of the invention mainly makes the following improvements: 1. in order to reduce data redundancy and input characteristic dimension, the invention deletes part of gas ratio and selects H₂/CH₄,C₂H_l/C₂H₄,C₂H₄/C₂H₂,H₂/(H₂+C₁+C₂),CH₄/(C₁+C₂),C₂H₆/(C₁+C₂),C₂H₄/(C₁+C₂) A total of seven linearly independent ratios are used as inputs. 2. To distinguish between samples with similar gas ratios but widely different total amounts, the total gas content is added to the input signature. 3. When a certain gas content is 0, 10 is used^-3Instead. H₂Represents hydrogen; CH (CH)₄Represents methane; c₂H₆Represents ethane; c₂H₄Represents ethylene; c₂H₂Represents acetylene; c₁+C₂Represents a total hydrocarbon of CH₄、C₂H₆、C₂H₄And C₂H₂The total amount of the four gases. C₁Represents an organic matter containing one carbon, i.e., methane. C₂Representing organic species containing two carbons, namely ethane, ethylene and acetylene.

A health assessment model module: and establishing a state interval through KMeans clustering to enable the output scores to have discrimination, establishing a health assessment model based on an XGboost algorithm, giving scores to the transformers between normal and fault states, and presenting the health assessment results of the transformers through the state interval. The input data to the health assessment model includes absolute gas content data including hydrogen, methane, ethane, ethylene, and acetylene. And the XGboost model intermediate layer probability is extracted by the health evaluation model to score the health condition of the transformer, and the health evaluation result of the transformer is presented through a state interval. The health assessment model automatically generates an explicit state interval according to the score output by the health assessment model. The input data to the health assessment model is the absolute content of the various gases. The input data of the health assessment model comprises absolute content data of gases such as hydrogen, methane, ethane, ethylene and acetylene.

And analyzing the monitoring data of the dissolved gas in the oil by the health evaluation model to evaluate the overall health state of the transformer. The health assessment model trains a classification model aiming at a two-classification task of judging whether the transformer belongs to a normal or fault state according to monitoring data of dissolved gas in the transformer oil. And training a transformer fault discrimination model, and when the accuracy of the transformer fault discrimination model is high enough, learning fault characteristics by the health evaluation model. The health evaluation model analyzes fault information contained in the transformer monitoring data, the more the fault information is, the lower the probability that an output sample is positive is, the worse the health condition of the transformer is, and the probability output by the health evaluation model is adopted as the health evaluation score of the transformer. The fault diagnosis model outputs specific fault states, and the total number is five. The fault discrimination model judges whether the transformer is normal or fault probabilistically. Only two states are output.

The XGboost and KMeans-based health assessment model module establishes a state interval through KMeans clustering, so that output scores have discrimination, and a proper score can be given to a transformer in a normal state and a fault state. A health evaluation model is established based on the XGboost algorithm, subjectivity caused by artificial scoring and empowerment in an evaluation process is eliminated, and an evaluation result is more accurate and objective.

The XGboost and KMeans-based health assessment model module mainly analyzes the principle of a health assessment model, introduces the specific structure of the health assessment model, determines the specific parameters of the model by using an experiment and analysis method, and automatically generates an explicit state interval according to the score output by the model.

Principle of the health assessment model: the deterioration of various factors related to the health condition of the transformer leads to the deterioration of the overall health condition of the transformer and is reflected in the monitoring data of gas in transformer oil. The health state of the whole transformer can be evaluated as long as the monitoring data of the dissolved gas in the oil is analyzed.

The specific structure of the health assessment model: and (3) training a classification model aiming at a two-classification task of judging whether the transformer belongs to a normal/fault state by using monitoring data of dissolved gas in the transformer oil. Firstly, a transformer fault discrimination model is trained, and if the model accuracy is high enough, the model can be considered to learn the fault characteristics. The fault information contained in the transformer monitoring data is analyzed by the model, the more the fault information is, the lower the probability that the output sample is positive is, and the worse the health condition of the transformer is, and the probability output by the model is taken into consideration as the health evaluation score of the transformer.

Specific parameters of the model: each regression tree has different depths and has different numbers of branch nodes and leaf nodes. According to the principle of the Okamm razor, a model as simple as possible should be selected on the premise of meeting the requirements of the model. The maximum depth of the regression tree is limited to 2, and the model can be greatly simplified under the condition of small accuracy loss. And taking the regression tree with the number of 10, wherein the leaf node number is 40, and the distribution condition of the model test set accuracy and the score is not inferior to that of the regression tree with the number of 12. Therefore, the number of regression trees of the finally selected model is 10, and the maximum depth is 2.

Determining a state interval: in order to overcome the objectivity introduced by artificially setting a state boundary threshold value for the model in the traditional scoring method, the KMeans clustering method is adopted in the model, and the boundary points of the health scores in different states are independently learned based on data. The transformer states are divided into the following 4 levels: normal, attention, abnormal and severe. The state intervals for the four classes were determined by cluster analysis as follows: normal- [0.8,1], Note- [0.43,0.8 ], abnormal- [0.15,0.43 ], Severe- [0,0.15 ].

The fault diagnosis model flowchart and the health assessment model flowchart of the present invention are shown in fig. 1 and 2. Firstly, collecting and obtaining detection data of dissolved gas in oil of the oil-immersed power transformer as a DGA data set, and performing the following steps according to the data of 7: the scale of 3 is divided into a training set and a test set. The sample is divided into five states of normal, low-temperature heating, high-temperature heating, low-energy discharging and high-energy discharging, and the five states are marked as 0 th, 1 st, 2 nd, 3 rd and 4 th types of states in sequence. The regression tree is used as a weak classifier in the model, the maximum depth of the tree is set to be 10, the number of iteration rounds is 210 rounds, and the model comprises 1050 regression trees in total, namely 210 × 5 regression trees. Model training adopts a 10-fold cross validation method, learning results are compared with traditional machine learning methods such as random forests, AdaBoost, decision trees and the like, and the results are shown in Table 1. Therefore, the XGboost model is superior to other models in the indexes such as diagnosis accuracy, AUC value and the like.

TABLE 1 comparison of various machine learning model performances

And then, considering a plurality of indexes such as gas content in the transformer oil, the degradation degree of the insulating oil, the degradation degree of insulating paper, winding dielectric loss, iron core grounding current, polarization index and the like, and comprehensively analyzing to obtain a transformer health evaluation result. The evaluation results of the present invention were compared with the evaluation results obtained by the SVM classifier, as shown in table 2.

TABLE 2 comparison of SVM classifier with model evaluation results of the present invention

Comparing the evaluation results of different models can find that, although the former can correctly evaluate the state of the transformer as serious, the model considers that the transformer state is normal with the same high probability, and actually the normal state and the serious state are far from each other, so the former model has a large risk. The model not only correctly evaluates the state of the transformer, but also gives very low health score, is far from the corresponding score intervals of other states, proves the evaluation result, reduces the risk of evaluation errors, and has very strong guiding significance.

The method has high fault diagnosis accuracy and strong health assessment objectivity, and is suitable for the power system with higher requirements on reliability and safety. The invention relates to transformer fault diagnosis and health assessment based on an Extreme Gradient Boosting (XGboost) algorithm. The system comprises a fault diagnosis model module based on XGboost, a data preprocessing module based on a gas ratio method, a health assessment model module based on XGboost and KMeans and the like. The fault diagnosis model considers that the ratio of dissolved gases in oil and the total amount of gases are used as input, the health assessment model considers that the XGboost model intermediate layer probability is extracted to score the health condition of the transformer, and the health assessment result of the transformer is presented through a state interval.

The invention relates to a state evaluation technology of electrical equipment, which is suitable for an electric power system with higher requirements on reliability and safety. The invention relates to a transformer fault diagnosis and health evaluation method based on an XGboost algorithm. The fault diagnosis model considers that the ratio of dissolved gases in oil and the total amount of gases are used as input, the health assessment model considers that the XGboost model intermediate layer probability is extracted to score the health condition of the transformer, and the health assessment result of the transformer is presented through a state interval.

The invention is realized by the following technical scheme. The invention comprises the following steps: the system comprises a fault diagnosis model module based on XGboost, a data preprocessing module based on a gas ratio method, a health assessment model module based on XGboost and KMeans and the like. Wherein: the XGboost-based fault diagnosis model module mainly analyzes the principle of the XGboost method, introduces the specific structure of the transformer fault diagnosis model, and shows the construction flow of the model by combining characters and images. The data preprocessing module based on the gas ratio method mainly improves the method of directly using the absolute content of the gas in other machine learning diagnosis methods, and converts the absolute content data of the gas into a relative gas ratio for fault diagnosis. The XGboost and KMeans-based health assessment model module mainly analyzes the principle of a health assessment model, introduces the specific structure of the health assessment model, determines the specific parameters of the model by using an experiment and analysis method, and automatically generates an explicit state interval according to the score output by the model.

The XGboost algorithm is adopted in the method, so that the fault diagnosis is accurate and reliable, the health assessment is objective and reasonable, the feasibility and the effectiveness are strong, and the method has guiding significance for the state assessment of the power transformer. The method is particularly suitable for power systems with high requirements on reliability and safety of power transformers. The invention can accurately diagnose the fault and objectively evaluate the state evaluation of the power transformer of the health condition.

Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, units provided by the present invention as pure computer readable program code, the system and its various devices, modules, units provided by the present invention can be fully implemented by logically programming method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system and various devices, modules and units thereof provided by the invention can be regarded as a hardware component, and the devices, modules and units included in the system for realizing various functions can also be regarded as structures in the hardware component; means, modules, units for performing the various functions may also be regarded as structures within both software modules and hardware components for performing the method.

The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.