阅读说明：本技术 一种油浸式变压器油温预测的概念漂移检测方法 (Conceptual drift detection method for oil temperature prediction of oil-immersed transformer ) 是由 徐健锋 郑智茗 刘斓 胡然 赵志宾 于 2021-05-20 设计创作，主要内容包括：本发明提供了一种油浸式变压器油温预测的概念漂移检测方法,包括3个模块：前置模块,误差检测模块,漂移判定模块；前置模块中将油浸式变压器的工况数据序列输入至油温预测器中预测相应时间的变压器油温；之后将预测到的结果集和对应真实油温序列传入误差检测模块求得相对误差率集合,接着将该集合送入漂移判定模块中；在漂移判定模块中为了检测出变压器油温预测模型是否发生了概念漂移,以γ为警戒值统计集合中预测异常的元素个数以及它们在整个预测结果集中的占比,并设定阈值对占比进行分析,最后得到判定结果。本发明针对油温预测器的概念漂移检测方法能够及时有效的发现油温预测器中概念漂移现象的产生。(The invention provides a conceptual drift detection method for oil temperature prediction of an oil-immersed transformer, which comprises 3 modules: the device comprises a front-end module, an error detection module and a drift judgment module; the working condition data sequence of the oil-immersed transformer is input into an oil temperature predictor in the preposed module to predict the oil temperature of the transformer at the corresponding time; then, the predicted result set and the corresponding real oil temperature sequence are transmitted into an error detection module to obtain a relative error rate set, and then the set is transmitted into a drift judgment module; in order to detect whether the transformer oil temperature prediction model generates conceptual drift or not in the drift determination module, gamma is used as the number of elements which are predicted to be abnormal in the warning value statistical set and the proportion of the elements in the whole prediction result set, a threshold value is set to analyze the proportion, and finally a determination result is obtained. The concept drift detection method for the oil temperature predictor can timely and effectively find the generation of the concept drift phenomenon in the oil temperature predictor.)

1. A conceptual drift detection method for oil temperature prediction of an oil-immersed transformer is characterized by comprising the following steps: the system comprises 3 modules: the device comprises a front-end module, an error detection module and a drift judgment module; inputting a working condition data sequence of the oil-immersed transformer into an oil temperature predictor in the preposed module to predict the oil temperature of the transformer at corresponding time, wherein the working condition data sequence of the oil-immersed transformer consists of oil temperature data acquisition time, active remote measurement of a main transformer high middle low voltage side, reactive remote measurement of the main transformer high middle low voltage side, power factor remote measurement of the main transformer high middle low voltage side, current value remote measurement of the main transformer high middle low voltage side, main transformer A/B/C phase winding temperature, main transformer A/B/C phase oil temperature and external meteorological environment attributes; then, the predicted result set and the corresponding real oil temperature sequence are transmitted into an error detection module to obtain a relative error rate set, and then the set is transmitted into a drift judgment module; in order to detect whether the transformer oil temperature prediction model generates conceptual drift or not in the drift determination module, gamma is used as the number of elements which are predicted to be abnormal in the warning value statistical set and the proportion of the elements in the whole prediction result set, a threshold value is set to analyze the proportion, and finally a determination result is obtained.

2. The conceptual drift detection method for oil temperature prediction of the oil-immersed transformer according to claim 1, wherein the conceptual drift detection method comprises the following steps: the front module will continue for a period of time t₁～t_nN sets of operating condition data sequences V_xInputting the signal into an oil temperature predictor to obtain t₁～t_nOil temperature prediction result set V'_yThen V'_yAnd corresponding real oil temperature sequence V_yIs transmitted to the errorA difference detection module, wherein:

3. the conceptual drift detection method for oil temperature prediction of the oil-immersed transformer according to claim 2, wherein the conceptual drift detection method comprises the following steps: the error detection module utilizes V'_yAnd V_yAnd obtaining relative error rates of oil temperature prediction, and recording a set of the relative error rates as delta, wherein:the method for obtaining the relative error rate comprises the following steps:wherein:

4. the conceptual drift detection method for oil temperature prediction of the oil-immersed transformer according to claim 3, wherein the conceptual drift detection method comprises the following steps: the drift determination module selects elements which are larger than a warning value gamma in a relative error rate set delta and forms the elements into an abnormal set delta', and then determines whether the oil temperature predictor generates concept drift: if | δ '| is the number of elements in the anomaly set δ', and | δ | is the number of elements in the relative error rate set δ, thenIf the value is less than or equal to the threshold value alpha, the concept drift does not occur; if it is notAbove threshold β, a conceptual drift has occurred; otherwise, adding one to the delay decision count variable θ, and comparing with the threshold λ: if θ is less than the threshold λ, then concept drift has occurred; otherwise, the original data is replaced by another continuous timeThe n groups of working condition data sequences and the oil temperature sequences in the corresponding time periods are returned to the front module, and the operation is carried out again, wherein: alpha is more than 0 and less than beta is less than 1, and lambda is a delay judgment time threshold value.

Technical Field

The invention belongs to the technical field of power grid equipment state monitoring and risk early warning, and particularly relates to a conceptual drift detection method for oil temperature prediction of an oil-immersed transformer.

Background

An oil-immersed transformer is one of the core devices in an electrical power system. However, in real life, if the oil-immersed transformer is in an overload state for a long time, the oil temperature is abnormal, so that the performance is reduced, and various fault problems such as unstable power distribution occur. Therefore, the future oil temperature of the transformer is predicted through the working condition data of the oil-immersed transformer, so that the abnormality of the oil-immersed transformer can be found in advance, and the method has important significance for guaranteeing the normal operation of the oil-immersed transformer. With the continuous and deep research on artificial intelligence in recent years, the abnormity diagnosis technology of the oil-immersed transformer is greatly improved. Many people try to analyze historical oil temperature working condition data of the transformer by using a machine learning technology so as to diagnose the abnormity of the transformer. Although the method based on machine learning is more convenient and easier to popularize than the traditional transformer abnormity diagnosis technology, the accuracy of the fault diagnosis method is generally low. Through research, the main reasons that the accuracy of fault diagnosis is not high are usually that the components inside the transformer are continuously oxidized, and the external load of the transformer changes irregularly, which may cause that the oil temperature working condition and the oil temperature data of the transformer easily change unpredictably along with the time, and the concept drift phenomenon refers to that the statistical characteristic of a target variable changes in an unforeseeable manner along with the time, so that the phenomenon is a typical concept drift phenomenon, and the phenomenon has great influence on the existing transformer oil temperature prediction model.

Disclosure of Invention

In the process of predicting the oil temperature of the transformer by using a machine learning method, in order to solve the problem that a conventional method does not consider that a transformer oil temperature predictor possibly has concept drift, a concept drift detection method for oil temperature prediction of an oil-immersed transformer is provided.

In order to achieve the purpose, the invention provides the following technical scheme: a conceptual drift detection method for oil temperature prediction of an oil-immersed transformer comprises 3 modules: the device comprises a front-end module, an error detection module and a drift judgment module; inputting a working condition data sequence of the oil-immersed transformer into an oil temperature predictor in the preposed module to predict the oil temperature of the transformer at corresponding time, wherein the working condition data sequence of the oil-immersed transformer consists of oil temperature data acquisition time, active remote measurement of a main transformer high middle low voltage side, reactive remote measurement of the main transformer high middle low voltage side, power factor remote measurement of the main transformer high middle low voltage side, current value remote measurement of the main transformer high middle low voltage side, main transformer A/B/C phase winding temperature, main transformer A/B/C phase oil temperature and external meteorological environment attributes; then, the predicted result set and the corresponding real oil temperature sequence are transmitted into an error detection module to obtain a relative error rate set, and then the set is transmitted into a drift judgment module; in order to detect whether the transformer oil temperature prediction model generates conceptual drift or not in the drift determination module, gamma is used as the number of elements which are predicted to be abnormal in the warning value statistical set and the proportion of the elements in the whole prediction result set, a threshold value is set to analyze the proportion, and finally a determination result is obtained.

Further, the front module can continue for a period of time t₁～t_nN sets of operating condition data sequences V_xInputting the signal into an oil temperature predictor to obtain t₁～t_nOil temperature prediction result set V_yThen, will V_yAnd corresponding real oil temperature sequence V_yTransmitting to an error detection module, wherein:

further, the error detection module utilizes V_yAnd V_yAnd obtaining relative error rates of oil temperature prediction, and recording a set of the relative error rates as delta, wherein:the method for obtaining the relative error rate comprises the following steps:wherein:

further, the drift determination module selects the error rate set delta to be larger than a warning value gammaAnd elements, and the elements are combined into an abnormal set delta', and then whether the concept drift occurs in the oil temperature predictor is judged: if | δ '| is the number of elements in the anomaly set δ', and | δ | is the number of elements in the relative error rate set δ, thenIf the value is less than or equal to the threshold value alpha, the concept drift does not occur; if it is notAbove threshold β, a conceptual drift has occurred; otherwise, adding one to the delay decision count variable θ, and comparing with the threshold λ: if θ is less than the threshold λ, then concept drift has occurred; otherwise, the original data is replaced by n groups of working condition data sequences in another continuous period and oil temperature sequences in the corresponding period, and then the operation is returned to the front module, and the operation is carried out again, wherein: alpha is more than 0 and less than beta is less than 1, and lambda is a delay judgment time threshold value.

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

the invention provides a concept drift detection method for oil temperature prediction of an oil immersed transformer, which can judge whether the concept drift occurs in an oil temperature predictor, thereby early warning and implementing a corresponding solution. The concept drift detection method for the oil temperature predictor is verified through an example, and the generation of the concept drift phenomenon in the oil temperature predictor can be timely and effectively found.

Drawings

FIG. 1 is a basic flow diagram of the present invention;

FIG. 2 is a front end module flow diagram;

FIG. 3 is a flow diagram of an error detection module;

fig. 4 is a flow chart of a drift determination module.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The embodiments described herein are only for explaining the technical solution of the present invention and are not limited to the present invention.

The invention provides a technical scheme that: as shown in fig. 1, a conceptual drift detection method for oil temperature prediction of an oil-immersed transformer respectively and sequentially executes a pre-module, an error detection module and a drift determination module, wherein the specific steps of each module are as follows:

as shown in fig. 2, the flow of the front module:

step 1: obtaining t₁～t_nN sets of operating condition data sequences V_xAnd t₁～t_nReal oil temperature sequence V_y. And (5) transferring to the step 2.

Step 2: will V_xInputting the oil temperature into an oil temperature predictor based on machine learning, predicting the corresponding oil temperature of the corresponding time period, and then collecting the obtained prediction result as V'_y. And (5) turning to the step 3.

And step 3: will V_yAnd V'_yTogether, input to the error detection module. And (5) turning to the step 4.

As shown in fig. 3, the error detection module flow:

and 4, step 4: and (3) taking the form of a relative error rate for each result predicted by the oil temperature predictor to represent the accuracy of the predicted transformer oil temperature, and recording the obtained result as delta, wherein:the method for obtaining the relative error rate comprises the following steps:wherein:and (5) turning to the step.

And 5: the set of relative error rates δ is communicated to a drift determination module. And (6) turning to the step.

As shown in fig. 4, the drift determination module flow:

step 6: and screening elements exceeding the warning value gamma in the delta by taking the warning value gamma of the relative error rate as a standard, and recording a set formed by the elements as an abnormal set delta'. Proceed to step 7.

And 7: and counting the number of elements in the anomaly set delta' and the relative error rate set delta and solving the ratio of the elements. Proceed to step 8.

And 8: if the ratio is less than or equal to the threshold value alpha, the oil temperature predictor is not subjected to concept drift, and the step 12 is carried out; otherwise, go to step 9.

And step 9: if the ratio is larger than the threshold value beta, the concept drift of the oil temperature predictor is shown, and the step 12 is carried out; otherwise, go to step 10.

Step 10: the delay decision count variable θ is incremented by one and θ is compared with a delay decision number threshold λ. If theta is less than lambda, the concept drift of the oil temperature predictor is shown, and the step 12 is carried out; otherwise, go to step 11.

Step 11: and (4) replacing the working condition data and the real oil temperature data with other continuous time periods, and then switching to the step 1.

Step 12: and outputting a drift judgment result.

Examples

A conceptual drift detection method for oil temperature prediction of an oil-immersed transformer comprises the following steps:

step 1: acquiring real oil temperature data of four 110kv oil immersed transformers in 2018 and 1-12 months and preprocessed working condition data, wherein the data is mainly formed by combining oil temperature working condition data of the transformers and corresponding meteorological data, and the data mainly comprises the following attributes: the method comprises the steps of oil temperature data acquisition time, active telemetry of a main transformer high middle low voltage side, reactive telemetry of the main transformer high middle low voltage side, power factor telemetry of the main transformer high middle low voltage side, current value telemetry of the main transformer high middle low voltage side, main transformer A/B/C phase winding temperature, main transformer A/B/C phase oil temperature and the like and corresponding meteorological attributes.

Step 2: selecting historical working condition oil temperature data from 2018 for 1 month to 2018 for 12 months, dividing the historical working condition oil temperature data into half months, and obtaining a real oil temperature data sequence (Y)₁，Y₂，...，Y₂₄) And operating condition data sequence (X)₁，X₂，...，X₂₄)。

And step 3: working condition data sequence (X)₁，X₂，...，X₂₄) Inputting the oil temperature into the existing transformer oil temperature prediction model for prediction to obtain an oil temperature prediction result set (Y)₁，Y₂，...，Y₂₄)。

And 4, step 4: obtaining a relative error rate set delta (delta) by using the oil temperature prediction result set and the real oil temperature data sequence₁，δ₂，...，δ₂₄) Wherein:

and 5: screening out elements of the relative error rate set delta larger than the alarm value 0.3, and forming the elements into an abnormal set delta', wherein:

step 6: the ratio of the number of elements in the anomaly set δ' to the number of elements in the relative error rate set δ is calculated.

And 7: if the ratio is less than or equal to the threshold value of 0.4, the oil temperature predictor has no concept drift, and the step 9 is entered; if the ratio is larger than the threshold value of 0.6, the oil temperature predictor generates concept drift and then the step 9 is carried out; otherwise, the decision is deemed to need to be delayed, and the process proceeds to step 8.

And 8: adding one to the delay decision counting variable theta, if theta exceeds a threshold value 4 at the moment, replacing a working condition data sequence and a corresponding oil temperature data sequence, and entering a step 3; otherwise, the oil temperature predictor generates concept drift and then the method enters step 9.

And step 9: and outputting a detection result of whether the concept drift occurs to the oil temperature predictor.

Step 10: and repeating the steps, and turning to the step 3 when the concept drift detection is required.

In this example, when the setting parameters γ is 0.3, α is 0.4, β is 0.6, and λ is 4, it is finally detected that the oil temperature predictor has 8 conceptual drifts from 1 month in 2018 to 12 months in 2018. Through manual verification, the number of times of concept drift really occurs is 9, and the recall ratio is 88.9%.

The foregoing merely represents preferred embodiments of the invention, which are described in some detail and detail, and therefore should not be construed as limiting the scope of the invention. It should be noted that, for those skilled in the art, various changes, modifications and substitutions can be made without departing from the spirit of the present invention, and these are all within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.