A method for predicting the furfural content of a transformer

Abstract

The application discloses a transformer furfural content prediction method, comprising the following steps: (1) collecting the historical information of furfural content, meteorological data and load data in oil; (2) taking the average temperature, average power, sampling interval and furfural content of the previous sampling as input, and the furfural content of the next sampling as output to establish a furfural prediction model; (3) based on the furfural prediction model, the furfural content of the next sampling is predicted for any two samplings; (4) an integrated learning model is established based on the prediction result; (5) the furfural content of other sampling time is predicted by means of the furfural prediction model; (6) the integrated learning model is used for calculation, and the final prediction value is obtained by anti-normalizing the calculation value. The present application predicts the present furfural content through the previous sampling information, meteorological data and load data since sampling, so as to make up for the problem of inaccurate furfural content measurement caused by oil replacement.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power equipment life prediction, and in particular to a transformer furfural content prediction method. BACKGROUND

[0002] Oil-immersed transformer is a relatively important equipment in power system, and the prediction of its life can help to better improve the economic benefits of transformer use.

[0003] However, the traditional life prediction method needs to use the empirical expression of furfural and polymerization degree, and on this basis, the segmented linear relationship between polymerization degree and life is used to predict the life of the transformer. It is noted that the acquisition of polymerization degree needs to be lifted in the heart of the transformer, which is more laborious than the sampling of furfural. However, for furfural, the sampling of furfural has the disadvantage of inaccuracy due to oil change operation.

[0004] The present method considers that in order to solve the problem of life prediction, the measurement of polymerization degree should be abandoned, and the problem of large error of furfural sampling with oil change operation should be further handled. Therefore, the present method also studies the sampling prediction problem with long interval time. Since the prediction effect is less accurate with larger interval time, the prediction value is fused by using ensemble learning. SUMMARY

[0005] Based on the above technical problems, the purpose of the present application is to overcome the shortcomings of the prior art and provide a transformer furfural content prediction method. The present application uses BP neural network to roughly establish a furfural prediction model, and then uses ensemble learning to fuse the prediction results of the furfural prediction model, which lays a certain foundation for subsequent life prediction.

[0006] To solve the above calculation problem, the present application adopts the following technical scheme:

[0007] A transformer furfural content prediction method comprises the following steps:

[0008] Step 1. Collect the historical information of furfural content in oil, meteorological data and load data;

[0009] Step 2. Take the furfural content h i of the i-th sampling, the average temperature T i,j of the interval between the i-th sampling and the j-th sampling, the average power P i,j , and the furfural content sampling interval I i,j as input, and the furfural content h j of the j-th sampling as output, and use BP neural network to establish a furfural prediction model, wherein

[0010] Step 3. Based on the furfural prediction model, the furfural content of the i-th furfural sampling, the average temperature, the average power and the sampling interval between the i-th furfural sampling and the j-th furfural sampling are taken as inputs to predict the furfural content of the j-th furfural sampling, and the prediction result is defined as

[0011] Step 4. Taking i = j-k, k = 1, 2, 3, …, l, let the input vector be The input matrix U = {G l+1 ; G l+2 ; …; G j ; …} and the output vector V = {h l+1 ; h l+2 ; …; h j ; …} are randomly shuffled to establish an ensemble learning model.

[0012] Step 5. Taking the furfural content h a of the a-th sampling, the average temperature T a,b , the average power P a,b and the furfural sampling interval I a,b between the a-th sampling and the b-th sampling as inputs, the furfural content of the b-th sampling is predicted by means of the furfural prediction model, and the predicted value is a = b-l, …, b-3, b-2, b-1.

[0013] Step 6. After normalization processing of , the ensemble learning model is used for calculation, and the final prediction value is obtained by anti-normalization of the calculated value.

[0014] Further, the meteorological data refers to outdoor temperature.

[0015] Further, the sampling time intervals of the historical information of furfural content, meteorological data and load data are uniform.

[0016] The average temperature T i,j is the average value of the operating temperature in the sampling interval; and the average power P i,j is the average value of the load in the sampling interval.

[0017] The average temperature T i,j is defined as:

[0018]

[0019] t1, t2, …, t m , …, t M are the temperature sampling time points between the i-th furfural sampling and the j-th furfural sampling, t1, t2, …, t m , …, t MThe corresponding temperatures are τ1, τ2, …, τ m , …, τ M , M is the number of temperature sampling times, wherein α and β are the time of the i-th furfural sampling and the j-th furfural sampling, that is, α < t1 < t2 < … < t m < … < t M < β.

[0020] Further, the average power P i,j is defined as:

[0021]

[0022] s1, s2, …, s n , …, s N are the power sampling time points between the i-th furfural sampling and the j-th furfural sampling, s1, s2, …, s n , …, s N are the power sampling time points between the i-th furfural sampling and the j-th furfural sampling, s1, s2, …, s n , …, s N , N is the number of power sampling times, that is, α < s1 < s2 < … < s n < … < s N < β.

[0023] Further, the furfural sampling interval I i,j can be defined as

[0024] I i,j = β-α.

[0025] Further, in step 4, the data sets U and V are randomly shuffled and divided into a training input set U train , a training output set V train , a verification input set U valid , and a verification output set V valid , and on the basis of U train and V train , normalization processing is performed and integrated learning is used for training, U valid and V valid are used for parameter adjustment, so as to establish an integrated learning model.

[0026] Compared with the prior art, the present application has the following beneficial effects:

[0027] 1. The traditional method is based on the functional relationship between the aggregation degree and the service life of the transformer, and the aggregation degree is obtained by hoisting the heart sampling, which is labor-intensive and complex to operate due to the need for power-off operation, so it is less developed, and the technology adopted by the present application avoids such heavy work of hoisting the heart, and is more operable and economical;

[0028] 2. When transformers have been in operation for a long time, oil changes are performed, which can lead to inaccurate furfural sampling. The technology used in this application makes full use of rich historical furfural content to make more accurate predictions of future furfural content, thereby laying a foundation for life prediction. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method for predicting furfural content in transformers according to the present invention. Detailed Implementation

[0030] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings, embodiments and comparative examples.

[0031] Example

[0032] like Figure 1 As shown in the figure, this embodiment provides a method for predicting the furfural content of transformers, including the following steps:

[0033] Step 1. Collect historical information on furfural content in oil, meteorological data, and load data;

[0034] Step 2. Using the furfural content h from the i-th sample... i The average temperature T between the i-th and j-th sampling times. i,j Average power P i,j Sampling interval I for furfural content i,j As input, the furfural content h of the j-th sample. j As output, a furfural prediction model is established using a BP neural network, whereby...

[0035] In a further preferred embodiment, the meteorological data refers to the outdoor temperature.

[0036] In a further preferred embodiment, the historical information sampling time intervals for furfural content, meteorological data, and load data are uniform;

[0037] Average temperature T i,j The average operating temperature over the two sampling intervals; average power P i,j This is the average load over the two sampling intervals;

[0038] Average temperature T i,j Defined as:

[0039]

[0040] t1, t2, ..., t m 、…、t M Let t1, t2, ..., tj be the temperature sampling times between the i-th and j-th furfural samplings. mt1, t2, …, t M τ1, τ2, …, τ m t1, t2, …, t M , M is the number of temperature sampling, where α and β are the time of the i-th furfural sampling and the j-th furfural sampling, i.e. α < t1 < t2 < … < t m < … < t M < β.

[0041] In a further preferred embodiment, the average power P i,j is defined as:

[0042]

[0043] s1, s2, …, s n , s2, …, s N is the power sampling time interval between the i-th furfural sampling and the j-th furfural sampling, s1, s2, …, s n , s2, …, s N , s2, …, s n , s2, …, s N , N is the number of power sampling, i.e. α < s1 < s2 < … < s n < … < s N < β.

[0044] In a further preferred embodiment, the furfural sampling interval I i,j is defined as

[0045] I i,j = β - α.

[0046] Step 3. Based on the furfural prediction model, the furfural content of the i-th furfural sampling, the average temperature of the interval between the i-th furfural sampling and the j-th furfural sampling, the average power, and the sampling interval are taken as inputs to predict the furfural content of the j-th furfural sampling, and the prediction result is defined as

[0047] In a further preferred embodiment, in the step 4, the data sets U and V are randomly shuffled and divided into a training input set U train , a training output set V train , a verification input set U valid , and a verification output set V valid , based on U train and V train , normalization processing is performed and training is performed using ensemble learning, U valid and V valid are used for parameter adjustment, and thus an ensemble learning model is established.

[0048] The following table shows the next sampling time of furfural prediction value based on the previous furfural sampling, the average temperature and the average power of the two sampling intervals.

[0049]

[0050] For example, it represents the prediction value of the furfural content at the second sampling time based on the first sampling value, the first second sampling interval and its average temperature and average power.

[0051] Based on the above prediction results, integrated learning is used for learning to establish an integrated model, as follows:

[0052] Step 4. Take i = j-k, k = 1, 2, 3, …, l, and let the input vector be The input matrix U = {G l+1 ; G l+2 ; …; G j ; …}, and the output vector V = {h l+1 ; h l+2 ; …; h j ; …}, the data set U and V are randomly shuffled and divided into training input set U train , training output set V train and validation input set U valid , validation output set V valid On the basis of U train and V train , normalize them and train them using integrated learning. Further, U valid and V valid are used for parameter tuning to establish an integrated learning model.

[0053] Step 5. Take the furfural content h a of the a-th sampling, the average temperature T a,b and the average power P a,b of the a-th sampling and the b-th sampling interval, and the furfural sampling interval I a,b as input, and use the furfural prediction model to predict the furfural content of the b-th sampling. The prediction value is a = b-l, …, b-3, b-2, b-1;

[0054] Step 6. After normalizing , use the integrated learning model to calculate the value, and then de-normalize the calculated value to obtain the final prediction value.

[0055] ​The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications, etc. made without departing from the spirit and principles of the present application should be equivalent replacement manners and should be included in the protection scope of the present application.

Claims

1. A method of transformer furfural content prediction, characterized by, The method comprises the following steps: Step 1. Collecting the history information of the furfural content in oil, meteorological data and load data; Step 2. The furfural content of the first i sample h i , the i sampling interval j average temperature between the first and second sampling T i,j , the average power P i,j , the furfural content sampling interval I i,j As input, the furfural content of the first j sample h j As output, a furfural prediction model is established using a BP neural network, wherein ; Step 3. Predict the furfural content of the nth furfural sampling based on the furfural prediction model with the furfural content of the (n-1)th furfural sampling, the average temperature, the average power, and the sampling interval between the (n-1)th furfural sampling and the nth furfural sampling as inputs, and the prediction result is defined as i F(n). i Step 4. Calculate the prediction error of the nth furfural sampling based on the prediction result of the nth furfural sampling and the actual furfural content of the nth furfural sampling, and the prediction error is defined as j E(n) = F(n) - F(n). j Step 5. Calculate the prediction accuracy of the nth furfural sampling based on the prediction result of the nth furfural sampling and the actual furfural content of the nth furfural sampling, and the prediction accuracy is defined as A(n) = 1 - E(n) / Step 4. Take i = j - k, k = 1, 2, 3, …, l , let the input vector , input matrix , output vector , data set U and V randomly shuffled and divided into training input set U train , training output set V train and validation input set U valid , validation output set V valid , on the basis of U train and V train , normalize and train with ensemble learning, and use U valid and V valid for parameter tuning, thereby establishing an ensemble learning model; Step 5. (The sentence is incomplete and requires more context.) a furfural content in the second sample h a , No. a The second sampling and the first b Average temperature between sampling intervals T a,b Average power P a,b Furfural sampling interval I a,b As input, the furfural prediction model is used to predict the first... b The furfural content of the second sample was predicted, and the predicted value was... , ; Step 6. The following formula is used to calculate the final prediction value: After normalization, the integrated learning model is used for calculation, and the final prediction value is obtained by anti-normalization of the calculated value.

2. The method of claim 1, wherein: The meteorological data refer to outdoor temperature.

3. The method of claim 1, wherein: The sampling time intervals of the history information of the furfural content in oil, meteorological data and load data are uniform. average temperature T i,j is the average of the operating temperatures in the two sampling intervals; average power P i,j is the average of the loads in the two sampling intervals; Average temperature T i,j is defined as: t 1 、t 2 、…、t m 、…、t M For the i rd furfural sampling and the j th furfural sampling interval, t 1 、t 2 、…、 t m 、…、t M The corresponding temperature is τ 1 , τ 2 , …, τ m , …, τ M , M is the number of temperature sampling times, where α and β are the times of the i rd furfural sampling and the j th furfural sampling, i.e. α<t 1 <t 2 、…<t m <…<t M <β .

4. The method of claim 3, wherein: Average power P i,j is defined as: s 1 、s 2 、…、s n 、…、s N for the first i time and the second j time, s 1 、s 2 、…、 s n 、…、s N corresponding temperature is ρ 1 , ρ 2 , …, ρ n , …, ρ N ，N number of power sampling, i.e. α<s 1 <s 2 、…<s n <…<s N <β .

5. The method for predicting the furfural content in a transformer according to claim 3, characterized in that: Furfural sampling interval I i,j may be defined as 。

Images