CN113987909B

CN113987909B - Oilpaper insulation aging prediction method, device, computer equipment and storage medium

Info

Publication number: CN113987909B
Application number: CN202111097032.4A
Authority: CN
Inventors: 朱晨; 李光茂; 乔胜亚; 王勇; 杨森; 周鸿铃; 郑服利; 邓剑平
Original assignee: Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Current assignee: Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Priority date: 2021-09-18
Filing date: 2021-09-18
Publication date: 2023-05-02
Anticipated expiration: 2041-09-18
Also published as: CN113987909A

Abstract

The application relates to a oiled paper insulation aging prediction method, a device, computer equipment and a storage medium. The method comprises the steps of obtaining ageing characteristic parameters to be predicted corresponding to oil paper insulation to be predicted, inputting the ageing characteristic parameters to be predicted into a target ageing prediction model obtained by training based on a genetic bias least square method, sample ageing characteristic parameters of a plurality of sample oil paper insulation and sample real ageing parameters, obtaining the ageing prediction parameters output by the target ageing prediction model, and determining a ageing prediction stage corresponding to the oil paper insulation to be predicted according to the ageing prediction parameters. Compared with the traditional method for observing the aging degree of the oilpaper insulation by manpower. According to the scheme, the target aging prediction model obtained based on the genetic bias least square method training is utilized, and the aging stage of the oil paper insulation is predicted according to the aging characteristics of the oil paper insulation, so that the accuracy of prediction is improved.

Description

Oilpaper insulation aging prediction method, device, computer equipment and storage medium

Technical Field

The present disclosure relates to the field of power technologies, and in particular, to a method and apparatus for predicting insulation aging of oilpaper, a computer device, and a storage medium.

Background

The transformer bears important tasks of energy conversion in the power transmission and distribution process in the power grid, and the safe operation of the transformer is a core for guaranteeing the safety and stability of a power system. The oil paper insulation is arranged in the oil immersed transformer to ensure the safety of the oil immersed transformer in operation. However, in the long-term operation process of the oil immersed transformer, the oil paper insulation can be influenced by multiple factors such as electricity, heat, environment and the like, so that the electrical and mechanical properties of the oil immersed transformer are reduced. Therefore, the method accurately diagnoses the insulating aging state of the oil paper of the oil-immersed electrical equipment, prevents the oil paper from happening, and is an important technical support for ensuring the safe and reliable operation of the large-scale oil-immersed electrical equipment. The current aging prediction of oiled paper insulation is usually observed manually. However, the prediction results are easily inaccurate by manually performing observation.

Therefore, the current oilpaper insulation aging prediction method has the defect of low prediction accuracy.

Disclosure of Invention

In view of the foregoing, it is desirable to provide a oiled paper insulation degradation prediction method, apparatus, computer device, and storage medium capable of improving prediction accuracy.

A method of predicting oiled paper insulation aging, the method comprising:

Obtaining aging characteristic parameters to be predicted corresponding to the insulation of the oil paper to be predicted;

inputting the aging characteristic parameters to be predicted into a target aging prediction model, and obtaining predicted aging parameters output by the target aging prediction model; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters;

and determining a predicted aging stage corresponding to the oiled paper insulation to be predicted according to the predicted aging parameter.

In one embodiment, the method further comprises:

obtaining a plurality of sample aging characteristic parameters corresponding to the insulation of a plurality of sample oilpapers and obtaining a plurality of sample real aging parameters corresponding to the insulation of the plurality of sample oilpapers;

obtaining an aging prediction model to be trained; the aging prediction model to be trained comprises an input layer, a hidden layer and an output layer; each layer in the aging prediction model to be trained comprises a corresponding threshold value, and each layer comprises a corresponding weight;

inputting the sample aging characteristic parameters into the input layer, and acquiring the sample aging prediction parameters output by the to-be-trained aging prediction model based on the hidden layer and the output layer;

Acquiring an error value of a predicted aging parameter of each sample and a real aging parameter of a corresponding sample, and judging whether the error value is smaller than a preset error threshold value or not;

if not, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained according to the error value, and returning to the step of inputting the aging characteristic parameters of the plurality of samples into the input layer;

if yes, ending the circulation, and obtaining the target aging prediction model according to the weight among layers in the aging prediction model to be trained and the threshold value of each layer when the current training is ended.

In one embodiment, the obtaining the plurality of sample real aging parameters corresponding to the plurality of sample oiled paper insulation includes:

aiming at each sample oilpaper insulation, acquiring a Raman spectrum of the sample oilpaper insulation;

and carrying out the treatments of removing the spectrum base line, removing the peak and reducing the noise on the Raman spectrogram to obtain a target Raman spectrogram corresponding to the sample oilpaper insulation, wherein the target Raman spectrogram is used as the real aging parameter of the sample.

In one embodiment, the inputting the plurality of sample aging characteristic parameters into the input layer, and obtaining the aging prediction model to be trained to predict the aging parameters based on the plurality of samples output by the hidden layer and the output layer includes:

Acquiring a first weight corresponding to a first node in the input layer and a second node in the hidden layer, and a first threshold of the second node;

acquiring a second weight corresponding to a second node in the hidden layer and a third node in the output layer, and a second threshold of the third node;

inputting the sample aging characteristic parameters into the input layer aiming at each sample aging characteristic parameter, and acquiring a first output result of the sample aging characteristic parameters in the hidden layer according to the first weight, the first threshold and a preset activation function;

and obtaining a second output result of the sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, and obtaining a sample prediction aging parameter corresponding to the sample aging characteristic parameter.

In one embodiment, obtaining an error value of each sample predicted aging parameter and a corresponding sample true aging parameter includes:

obtaining root mean square error values of the sample predicted aging parameters and the corresponding real aging parameters;

obtaining a corresponding prediction set correlation coefficient according to the square of the difference between the sample prediction aging parameter and the sample real aging parameter and the square of the difference between the sample real aging parameter and the average value of a plurality of sample real aging parameters;

And taking the root mean square error value and the prediction set correlation coefficient as the error value.

In one embodiment, the adjusting the weights between layers and the thresholds of the layers in the aging prediction model to be trained according to the error value includes:

obtaining a corresponding error function according to the sample prediction aging parameter and the sample real aging parameter;

and according to the error function and a preset activation function, adjusting the first weight, the first threshold, the second weight and the second threshold based on a Widrow-Hoff learning rule and a momentum gradient descent algorithm.

In one embodiment, the predicted aging parameter is a predicted raman spectrum corresponding to the aging characteristic parameter to be predicted;

the step of determining the predicted aging stage corresponding to the oiled paper insulation to be predicted according to the predicted aging parameter comprises the following steps:

inquiring an aging stage table according to the predicted Raman spectrogram, and determining a predicted aging stage corresponding to the insulation of the oil paper to be predicted; the aging stage table comprises a plurality of correspondence relations between oil paper insulation aging stages and a Raman spectrogram.

An oiled paper insulation degradation prediction apparatus, the apparatus comprising:

The acquisition module is used for acquiring aging characteristic parameters to be predicted corresponding to the insulation of the oil paper to be predicted;

the input module is used for inputting the aging characteristic parameters to be predicted into a target aging prediction model and obtaining predicted aging parameters output by the target aging prediction model; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters;

and the prediction module is used for determining a predicted aging stage corresponding to the oil paper insulation to be predicted according to the predicted aging parameter.

A computer device comprising a memory storing a computer program and a processor implementing the steps of the method described above when the processor executes the computer program.

A computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method described above.

According to the oil paper insulation aging prediction method, the device, the computer equipment and the storage medium, the aging characteristic parameters to be predicted corresponding to the oil paper insulation to be predicted are obtained, the aging characteristic parameters to be predicted are input into the target aging prediction model which is obtained based on the genetic bias least square method, the sample aging characteristic parameters of the plurality of sample oil paper insulation and the sample real aging parameters, and the predicted aging parameters output by the target aging prediction model are obtained, so that the predicted aging stage corresponding to the oil paper insulation to be predicted is determined according to the predicted aging parameters. Compared with the traditional method for observing the aging degree of the oilpaper insulation by manpower. According to the scheme, the target aging prediction model obtained based on the genetic bias least square method training is utilized, and the aging stage of the oil paper insulation is predicted according to the aging characteristics of the oil paper insulation, so that the accuracy of prediction is improved.

Drawings

FIG. 1 is an application environment diagram of an oiled paper insulation aging prediction method in one embodiment;

FIG. 2 is a flow chart of a method for predicting the aging of oiled paper insulation according to one embodiment;

FIG. 3 is a schematic diagram of an aging prediction model in one embodiment;

FIG. 4 is a schematic flow diagram of a genetic partial least squares method in one embodiment;

FIG. 5 is a block diagram of an oiled paper insulation degradation prediction device in one embodiment;

fig. 6 is an internal structural diagram of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

The method for predicting the insulation aging of the oiled paper can be applied to an application environment shown in the figure 1. The terminal 102 can extract the aging characteristic parameters to be predicted corresponding to the oil paper insulation to be predicted, and input the aging characteristic parameters to be predicted into the target aging prediction model to obtain the predicted aging parameters output by the target aging prediction model, so that the terminal 102 determines the aging stage of the oil paper insulation to be predicted according to the predicted aging parameters. Additionally, in some embodiments, a server 104 is also included. Wherein the terminal 102 communicates with the server 104 via a network. The aging characteristic parameters to be predicted may be obtained by the terminal 102 from a database of the server 104. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, and tablet computers, and the server 104 may be implemented as a stand-alone server or a server cluster composed of a plurality of servers.

In one embodiment, as shown in fig. 2, there is provided a method for predicting insulation aging of oilpaper, which is described by taking an example that the method is applied to a terminal in fig. 1, and includes the following steps:

step S202, obtaining aging characteristic parameters to be predicted corresponding to the insulation of the oil paper to be predicted.

The oil paper insulation can be a component in the oil-immersed transformer, and the oil paper insulation can be influenced by multiple factors such as electricity, heat, environment and the like to cause the reduction of the electrical and mechanical properties of the oil-immersed transformer in the long-term operation process, so that the aging condition of the oil paper insulation needs to be predicted in order to ensure the safety of electrical equipment. The oiled paper insulation to be predicted may be oiled paper insulation for which aging prediction is required. The terminal 102 may obtain the aging characteristic parameter to be predicted for the oiled paper insulation to be predicted. The aging characteristic parameter to be predicted can represent the aging characteristic of the oil paper insulation to be predicted, and the aging characteristic parameter to be predicted can be an effective characteristic extracted from a Raman spectrogram of the oil paper insulation to be predicted. For example, the terminal 102 may perform raman spectrum detection on the insulation of the oil paper to be predicted in advance to obtain a corresponding raman spectrum, so that the terminal 102 may extract effective features in the raman spectrum of the insulation of the oil paper to be predicted as the aging feature parameters to be predicted. Among them, raman spectrum (Raman spectra) is a kind of scattering spectrum. The raman spectroscopy is an analysis method for analyzing a scattering spectrum different from the frequency of incident light based on the raman scattering effect to obtain information on vibration and rotation of molecules, and is applied to molecular structure research.

Step S204, inputting the aging characteristic parameters to be predicted into a target aging prediction model to obtain the predicted aging parameters output by the target aging prediction model; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters.

The target aging prediction model can be a neural network model for predicting the aging condition of the oilpaper insulation, and can be obtained by training a plurality of sample aging characteristic parameters and sample real aging parameters of the sample oilpaper insulation based on a genetic partial least square method. The terminal 102 may input the extracted aging characteristic parameter to be predicted into the target aging prediction model, so that the target aging prediction model may output a corresponding predicted aging parameter according to the input aging characteristic parameter to be predicted. The predicted aging parameters can be Raman spectrograms obtained by predicting the oil paper insulation by the target aging prediction model, the sample aging characteristics can be characteristics of an input model when the aging prediction model is trained, and the sample aging characteristics can be obtained by extracting from the Raman spectrograms of the sample oil paper insulation; the real aging parameters of the sample can be a real raman spectrum obtained after the terminal 102 performs raman spectrum detection on the sample oilpaper insulation; the genetic partial least square method can be a GA-PLS neural network, wherein the GA (Genetic Algorithm ) algorithm is designed and proposed according to the organism evolution rule in the nature, is a calculation model of the biological evolution process simulating the natural selection and genetic mechanism of the Darling biological evolution theory, and is a method for searching the optimal solution by simulating the natural evolution process; the Partial Least Squares (PLS) is a multivariate calibration method, in raman spectroscopy, a certain band is selected to participate in modeling through PLS selecting a characteristic variable, so that a better calibration model is obtained, and the terminal 102 can predict the aging stage of the oilpaper insulation by combining the genetic algorithm and the partial least squares.

Step S206, determining a predicted aging stage corresponding to the insulation of the oiled paper to be predicted according to the predicted aging parameters.

The predicted aging parameter may be a predicted raman spectrum obtained by predicting the sample aging characteristic parameter by the terminal 102 using the target aging prediction model. The terminal 102 may determine a predicted aging stage corresponding to the insulation of the oiled paper to be predicted according to the predicted aging parameter obtained by the prediction.

Wherein, since the predicted aging parameter may be a predicted raman spectrum corresponding to the aging characteristic parameter to be predicted, the terminal 102 determines a predicted aging stage corresponding to the insulation of the oiled paper to be predicted according to the predicted raman spectrum. For example, in one embodiment, determining a predicted aging stage corresponding to the insulation of the oiled paper to be predicted based on the predicted aging parameters includes: inquiring an aging stage table according to the predicted Raman spectrogram, and determining a predicted aging stage corresponding to the insulation of the oil paper to be predicted; the aging stage table comprises a plurality of correspondence relations between the oil paper insulation aging stages and the Raman spectrogram. In this embodiment, the terminal 102 may query the aging stage table according to the predicted raman spectrum obtained by the prediction, so that the terminal 102 may query the aging stage table for the oiled paper insulation corresponding to the predicted raman spectrum. The aging stage table may include a correspondence relationship between a plurality of aging stages of oiled paper insulation and a raman spectrum. The terminal 102 may predict the aging stage of the oiled paper insulation by the above method, and the specific predicted result and the actual aging result may be compared as shown in table 1:

TABLE 1 Raman Spectroscopy diagnostic results

As can be seen from table 1, by the method provided by the above embodiment, the accuracy of the aging stage prediction for the oiled paper insulation can be improved.

According to the oil paper insulation aging prediction method, the to-be-predicted aging characteristic parameters corresponding to the to-be-predicted oil paper insulation are obtained, the to-be-predicted aging characteristic parameters are input into the target aging prediction model obtained based on the genetic bias least square method, the sample aging characteristic parameters of the plurality of sample oil paper insulation and the sample real aging parameters, and the predicted aging parameters output by the target aging prediction model are obtained, so that the predicted aging stage corresponding to the to-be-predicted oil paper insulation is determined according to the predicted aging parameters. Compared with the traditional method for observing the aging degree of the oilpaper insulation by manpower. According to the scheme, the target aging prediction model obtained based on the genetic bias least square method training is utilized, and the aging stage of the oil paper insulation is predicted according to the aging characteristics of the oil paper insulation, so that the accuracy of prediction is improved.

In one embodiment, further comprising: obtaining a plurality of sample aging characteristic parameters corresponding to the insulation of the plurality of sample oilpapers and obtaining a plurality of sample real aging parameters corresponding to the insulation of the plurality of sample oilpapers; obtaining an aging prediction model to be trained; the aging prediction model to be trained comprises an input layer, a hidden layer and an output layer; each layer in the aging prediction model to be trained comprises a corresponding threshold value, and each layer comprises a corresponding weight; inputting a plurality of sample aging characteristic parameters into an input layer, and acquiring a plurality of sample aging prediction parameters output by a to-be-trained aging prediction model based on a hidden layer and an output layer; acquiring error values of the predicted aging parameters of each sample and the real aging parameters of the corresponding sample, and judging whether the error values are smaller than a preset error threshold value or not; if not, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained according to the error value, and returning to the step of inputting the aging characteristic parameters of a plurality of samples into the input layer; if yes, ending the circulation, and obtaining the target aging prediction model according to the weight among layers in the aging prediction model to be trained and the threshold value of each layer when the current training is ended.

In this embodiment, the terminal 102 may utilize a plurality of sample oiled paper insulations to extract a plurality of sample aging characteristic parameters therefrom, and obtain a plurality of sample real aging parameters corresponding to the plurality of sample oiled paper insulations, and train the aging prediction model to be trained through the sample aging characteristic parameters and the sample real aging parameters. The aging prediction model may be a BP neural network, and the structure of the BP neural network may be shown in fig. 3, and fig. 3 is a schematic structural diagram of the aging prediction model in one embodiment. The neural network includes an input layer, a hidden layer, and an output layer, and the terminal 102 may train the target aging prediction model based on the neural network. The obtaining of the aged sample may be performed in advance before obtaining a plurality of sample aging characteristic parameters corresponding to the plurality of sample oilpaper insulation and a plurality of sample real aging parameters. For example, (1) fresh insulating oil and insulating paper which have been dried in advance are placed in an aging tank, the aging tank is set to a heating temperature (for example, 130 ℃), and then the aging samples of the transformer oil are periodically sampled, and finally, 14 groups of aging samples of 1d, 2d, 3d, 4d, 6d, 8d, 10, d12, d14, d16, d20, d24, d26 and d30 are obtained in total, and each group of 15 samples comprises 210 aging samples of the transformer oil, and the aging samples comprise four stages from good insulation, early aging stage, middle aging stage and final aging stage of the oil paper insulation. The terminal 102 may perform raman spectrum detection on each of the aging samples, and perform corresponding pretreatment on the raman spectrum obtained by the detection, so that the pretreated raman spectrum is used as a sample real aging parameter corresponding to each aging sample. In addition, it should be noted that, in the actual prediction process, the raman spectrum obtained by performing raman spectrum detection on the insulation of the oil paper to be predicted may be subjected to extraction of the aging characteristic parameters to be predicted after the pretreatment. The terminal 102 may also divide the various aging samples into a training set and a validation set. For example, the terminal 102 may divide 210 samples by a Cross-validation method on the preprocessed samples according to a ratio of 7:3, to obtain 147 training sets and 63 validation sets. The terminal 102 can thus extract the sample aging characteristic parameters from the oiled paper insulation in the training set as an aging sample. For example, the terminal 102 may select and model the data sample features by the GA-PLS method, so as to complete the selection and dimension reduction of the raman spectrum feature sample information.

The terminal 102 may input the extracted multiple sample aging characteristic parameters into an aging prediction model to be trained, where the aging prediction model to be trained may be a BP neural network. The terminal 102 may obtain a plurality of sample predicted aging parameters based on the hidden layer and output layer output by the aging prediction model to be trained. The sample aging parameter may be a predicted raman spectrum corresponding to insulation of the sample oilpaper obtained after the sample aging characteristic parameter is predicted by the aging prediction model to be trained. Specifically, the terminal 102 may use the effective features extracted from the 147 training set data as the input of the BP neural network, and establish a GA-PLS-BP neural network diagnostic model and train.

After obtaining the sample prediction aging parameter output by the aging prediction model to be predicted, the terminal 102 may compare the sample prediction aging parameter with the error value of the corresponding sample real aging parameter, and determine whether the error value is smaller than the preset error threshold. If not, the terminal 102 may adjust the weights between layers and the thresholds of the layers in the aging prediction model to be trained according to the error values, and return to the step of inputting the plurality of sample aging characteristic parameters into the input layer, so that the terminal 102 may train the sample aging characteristic parameters according to the aging prediction model after adjusting the weights and the thresholds. If so, the terminal 102 may end the cycle and obtain the target aging prediction model according to the weights between the layers and the thresholds of the layers in the aging prediction model to be trained when the current training is ended.

The training process of the aging prediction model to be trained can be training based on a GA-PLS algorithm. As shown in fig. 4, fig. 4 is a schematic flow diagram of the genetic partial least squares method in one embodiment. The implementation of the Genetic Algorithm (GA) mainly includes several basic elements as shown in fig. 4, after the genetic iteration is terminated, all variables are rearranged according to the selected frequency, and then the selected variable number is plotted with the correlation coefficient (r) to select the optimal variable number, so as to obtain the selected variable. The BP neural network consists of an input layer, an implicit layer and an output layer. The learning process of the BP neural network consists of two processes of forward propagation of signals and backward propagation of errors. In forward propagation, an input sample is transmitted from an input layer, is processed layer by a hidden layer, and is transmitted to an output layer.

Through the above embodiment, the terminal 102 may train the aging prediction model by using the genetic bias least square method, and may predict the aging degree of the oiled paper insulation by using the target aging prediction model after training, thereby improving the prediction accuracy.

In one embodiment, obtaining a plurality of sample real aging parameters corresponding to a plurality of sample oiled paper insulation includes: aiming at each sample oilpaper insulation, acquiring a Raman spectrum of the sample oilpaper insulation; and carrying out the treatments of removing the spectrum base line, removing the peak and reducing the noise on the Raman spectrogram to obtain a target Raman spectrogram corresponding to the sample oil paper insulation, and taking the target Raman spectrogram as a sample real aging parameter.

In this embodiment, the terminal 102 may perform preprocessing on the raman spectrum detected from the oiled paper insulation, so as to obtain the corresponding real aging parameters of the sample. The terminal 102 may pre-process raman spectra of both the oilpaper insulation to be predicted and the sample oilpaper insulation. For example, for sample oilpaper insulation, the terminal 102 may obtain a raman spectrum of each sample oilpaper insulation, and perform a spectrum baseline removal, peak removal and noise reduction treatment on the raman spectrum to obtain a target raman spectrum corresponding to each sample oilpaper insulation as a sample real aging parameter.

Wherein each of the above-described pretreatments may be performed in a different manner. For example, for the baseline, the terminal 102 may use polynomial iterative fitting to subtract, adjust the original spectrum data by continuously comparing, and directly compare the adjusted spectrum data with points on the fitting curve, so as to make the calculated baseline function more approximate to the actual baseline, and finally subtract the obtained baseline from the original spectrum; for noise reduction, the terminal 102 may use a cyclic three-point zero-order Savitzky-Golay filtering method, and perform polynomial least square fitting on elements in the window through a moving window to obtain a value after element smoothing at the central position of the window. The method can achieve good denoising effect and well retain important information components in the spectrum. For spikes, the terminal 102 may perform cyclic filtering using a three-point sliding window averaging method, and stop the cyclic filtering when the standard deviation of the filtered noise is greater than or equal to the estimated standard deviation of the noise in the original spectrum. By using this method, the interference caused by high intensity spikes such as cosmic rays is eliminated.

Through the embodiment, the terminal 102 can remove impurities from the acquired raman spectrum in various manners, so that the effectiveness of the raman spectrum can be improved, and the accuracy of ageing degree prediction can be improved.

In one embodiment, inputting a plurality of sample aging characteristic parameters into an input layer, obtaining a plurality of sample prediction aging parameters output by a to-be-trained aging prediction model based on a hidden layer and an output layer, includes: acquiring a first weight corresponding to a first node in an input layer and a second node in a hidden layer, and a first threshold of the second node; acquiring a second weight corresponding to a second node in the hidden layer and a third node in the output layer, and a second threshold of the third node; inputting the sample aging characteristic parameters into an input layer aiming at each sample aging characteristic parameter, and acquiring a first output result of the sample aging characteristic parameters in a hidden layer according to a first weight, a first threshold and a preset activation function; and obtaining a second output result of the sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, and obtaining a sample prediction aging parameter corresponding to the sample aging characteristic parameter.

In this embodiment, the aging prediction model may be a BP neural network, including an input layer, a hidden layer, and an output layer, and each layer includes a plurality of nodes. The first weight is contained between each node of the input layer and each node in the hidden layer, and each node in the hidden layer contains a first threshold value; a second threshold is included between each node in the hidden layer and each node in the output layer, and each node in the output layer includes the second threshold. In the training process, when obtaining the sample predicted aging parameter, the terminal 102 may obtain the first weight, the first threshold, the second weight, and the second threshold. For each sample aging characteristic parameter, the terminal 102 may input the sample aging characteristic parameter to the input layer, and obtain a first output result of each sample aging characteristic parameter in the hidden layer according to the first weight, the first threshold, and a preset activation function. The first output result also needs to be input to the output layer, and the terminal 102 may obtain the second output result of each sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, so that the terminal 102 may obtain the sample predicted aging parameter corresponding to each sample aging characteristic parameter according to the output result of the output layer.

Wherein the first output result can be obtained according to the output of each node in the hidden layer, and the second output result can be obtained according to the output of each node in the output layerObtaining the product. For example, the terminal 102 may set the weight between node i and node j to be w _ij Node j has a threshold b _j The output value of each node is x _j The output value of each node is realized according to the output value of all nodes at the upper layer, the weights of the current node and all nodes at the upper layer, the threshold value of the current node and the activation function. The specific calculation formula can be as follows:

x _j ＝f(S _j )；

the node i may be a node of a previous layer, the node j may be a node of a next layer corresponding to the previous layer, and f is an activation function.

Through the embodiment, the terminal 102 can predict the aging parameters by using the threshold value of each layer and the weight value between each layer in the neural network, and predict the aging stage of the oilpaper insulation by using the trained neural network, thereby improving the prediction accuracy.

In one embodiment, obtaining error values for each sample predicted aging parameter and a corresponding sample true aging parameter includes: obtaining root mean square error values of sample prediction aging parameters and corresponding real aging parameters; obtaining a corresponding prediction set correlation coefficient according to the square of the difference between the sample prediction aging parameter and the sample real aging parameter and the square of the difference between the sample real aging parameter and the average value of the plurality of sample real aging parameters; the root mean square error value and the prediction set correlation coefficient are taken as error values.

In this embodiment, the terminal 102 may also determine the accuracy of training in the training process. For example by determining the error value between the sample predicted aging parameter and the corresponding sample true aging parameter. The terminal 102 may obtain root mean square error values corresponding to the sample aging prediction parameters and the corresponding real aging parameters; the terminal 102 may further obtain a corresponding prediction set correlation coefficient according to the square of the difference between the sample predicted aging parameter and the sample real aging parameter and the square of the difference between the sample real aging parameter and the average value of the plurality of sample real aging parameters; the terminal 102 may thus take the root mean square error value and the prediction set correlation coefficient as the error values.

Specifically, to verify the accuracy of the GA-PLS-BP neural network model predictions, the terminal 102 may, during the training process, verify the model's cross-validation root mean square error value (MSE) and the prediction set correlation coefficient (R) ² ) As an evaluation index of the model. R is R ² Representing the degree of correlation of the true value and the predicted value. The degree of variation in the data may be evaluated by the MSE, with smaller values of MSE indicating better accuracy of the predictive model. The specific calculation formula is as follows:

wherein,,

for the spectral prediction value, y is the actual value, +. >

And n is the number of samples, namely the number of sample aging characteristic parameters of the input aging prediction model, which is the average value of the actual values. In the actual training process, when the terminal 102 judges the error value and the preset error threshold, whether the root mean square error value is smaller than the root mean square error threshold or not can be respectively judged, whether the correlation coefficient of the prediction set is larger than the similarity threshold or not is judged, and if the judgment of the terminal 102 is yes, the error value is determined to meet the requirement; otherwise, the method is not satisfactory.

With this embodiment, the terminal 102 may determine an error value between the predicted aging parameter and the actual aging parameter based on the root mean square error value and the correlation coefficient of the prediction set, thereby improving the prediction accuracy of the aging prediction model.

In one embodiment, adjusting the weights between layers and the thresholds of the layers in the aging prediction model to be trained according to the error value includes: obtaining a corresponding error function according to the sample prediction aging parameter and the sample real aging parameter; and according to the error function and the preset activation function, adjusting a first weight, a first threshold, a second weight and a second threshold based on a Widry-Hoff learning rule and a momentum gradient descent algorithm.

In this embodiment, during the training of the aging prediction model, the terminal 102 may adjust the weight between each layer and the threshold value of each layer in the aging prediction model when the error value does not meet the requirement. The terminal 102 may obtain a corresponding error function according to the sample predicted aging parameter and the sample actual aging parameter. The terminal 102 may thus adjust the first weight, the first threshold, the second weight, and the second threshold described above based on the widow-Hoff learning rule and the momentum gradient descent algorithm according to the error function and the preset activation function.

Wherein, the hidden layer can also be called as hidden layer; the adjustment process may be a back propagation process in BP neural network training. For example, if the terminal 102 detects that the actual output of the output layer does not match the desired output, the counter-propagation phase of the error may be reversed. The back propagation of the error is to reversely propagate the output error layer by layer to the input layer through the hidden layer in a certain form, and to distribute the error to all units of each layer, so as to obtain the error signal of each layer unit, which is used as the basis for correcting the weight of each unit. In BP neural network, the error signal reverse transfer sub-process is based on the Widry-Hoff learning rule. Assuming that all results of the output layer are d _j The error function is as follows:

wherein d _j The sample can be the predicted aging parameter, y _j May be the corresponding sample true aging parameter.

The main purpose of the BP neural network is to repeatedly correct the weights and thresholds so that the error function value is minimized. The widry-Hoff learning rule is to continuously adjust the weight and the threshold of the network along the steepest descent direction of the sum of squares of relative errors, and according to the gradient descent method, the correction of the weight vector is in direct proportion to the gradient of E (w, b) at the current position, and for the jth output node:

the terminal 102 may cause the activation function to be:

The terminal 102 may derive the activation function to obtain:

for the weight w between the node i and the node j _ij The method comprises the following steps: />

Wherein,,

threshold b for node j _j Has the following components

The above is the process of calculating the adjustment amount for the weight between the hidden layer and the output layer and the threshold value of the output layer. For the weight adjustment from the input layer to the hidden layer and the calculation of the threshold adjustment amount of the hidden layer, the terminal 102 may assume that the weight between the kth node of the input layer and the ith node of the hidden layer is obtained, and the terminal 102 may obtain:

wherein,,

according to the gradient descent algorithm, the terminal 102 may adjust the weights between the hidden layer and the output layer and the threshold of the hidden layer according to the following formula:

The terminal 102 may adjust the weights between the input layer and the hidden layer and the threshold of the input layer according to the following formula:

the main idea of the gradient descent method is to search the optimal solution along the negative gradient direction, wherein the negative gradient direction is the direction in which the function value descends fastest, if the gradient iterated to a certain place is 0, the local minimum is reached, and the parameter updating is stopped. During the actual training process, the terminal 102 may use some improved method to get as close to the global minimum as possible.

In addition, in some embodiments, the terminal 102 may also add a motion term to the weight adjustment formula. For example, the BP algorithm is modified by using the momentum gradient descent method, and the BP network before modification is only adjusted according to the gradient direction of the error at the time t when the weight between neurons is adjusted, and the gradient direction before the time t is not considered, so that the training process is easy to developThe method comprises the steps of generating oscillation, converging slowly, adding a motion term to a weight adjustment formula in order to improve the training speed of a network, and if W represents a certain layer of weight matrix and x represents a certain layer of input vector, then a weight adjustment vector expression comprising the motion term is as follows: w (w) _ij ＝w _ij -η ₁ ·δ _ij ·x _i +αΔw _ij The method comprises the steps of carrying out a first treatment on the surface of the As can be seen from the above, adding the momentum term, that is, taking out a part from the previous weight adjustment and adding it to the current weight adjustment, alpha is called momentum coefficient, generally 0<α<1. The physical meaning represented by the motion term reflects the experience accumulated before, plays a damping role at the moment t, and can reduce the oscillation trend and improve the training process when the weight curved surface suddenly descends.

Through the above embodiment, the terminal 102 first screens out the characteristic wave number points representing the aging product of the oilpaper by the GA-PLS method, then uses the characteristic wave number points as the input vector of the neural network, continuously trains the constructed neural network, adjusts the parameters, and finally obtains the optimal parameters to achieve a smaller error value. The training and construction of the whole model are performed in a Matlab neural network tool box, and the terminal 102 can observe the final optimization performance through the correlation coefficient R2 and the MSE index. Thereby, the accuracy of predicting the degree of aging of the oiled paper insulation can be improved.

It should be understood that, although the steps in the flowcharts of fig. 2 and 4 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least a portion of the steps of fig. 2 and 4 may include a plurality of steps or stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily sequential, but may be performed in rotation or alternately with at least a portion of the steps or stages of other steps or steps.

In one embodiment, as shown in fig. 5, there is provided an oiled paper insulation aging prediction apparatus, comprising: an acquisition module 500, an input module 502, and a prediction module 504, wherein:

the obtaining module 500 is configured to obtain an aging characteristic parameter to be predicted corresponding to the insulation of the oiled paper to be predicted.

The input module 502 is configured to input the aging characteristic parameter to be predicted into a target aging prediction model, and obtain a predicted aging parameter output by the target aging prediction model; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters.

And the prediction module 504 is configured to determine a predicted aging stage corresponding to the oiled paper insulation to be predicted according to the predicted aging parameter.

In one embodiment, the apparatus further comprises: the training module is used for acquiring a plurality of sample aging characteristic parameters corresponding to the plurality of sample oil paper insulations and acquiring a plurality of sample real aging parameters corresponding to the plurality of sample oil paper insulations; obtaining an aging prediction model to be trained; the aging prediction model to be trained comprises an input layer, a hidden layer and an output layer; each layer in the aging prediction model to be trained comprises a corresponding threshold value, and each layer comprises a corresponding weight; inputting a plurality of sample aging characteristic parameters into an input layer, and acquiring a plurality of sample aging prediction parameters output by a to-be-trained aging prediction model based on a hidden layer and an output layer; acquiring error values of the predicted aging parameters of each sample and the real aging parameters of the corresponding sample, and judging whether the error values are smaller than a preset error threshold value or not; if not, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained according to the error value, and returning to the step of inputting the aging characteristic parameters of a plurality of samples into the input layer; if yes, ending the circulation, and obtaining the target aging prediction model according to the weight among layers in the aging prediction model to be trained and the threshold value of each layer when the current training is ended.

In one embodiment, the training module is specifically configured to obtain, for each sample oiled paper insulation, a raman spectrum of the sample oiled paper insulation; and carrying out the treatments of removing the spectrum base line, removing the peak and reducing the noise on the Raman spectrogram to obtain a target Raman spectrogram corresponding to the sample oil paper insulation, and taking the target Raman spectrogram as a sample real aging parameter.

In one embodiment, the training module is specifically configured to obtain a first weight corresponding to a first node in the input layer and a second node in the hidden layer, and a first threshold of the second node;

acquiring a second weight corresponding to a second node in the hidden layer and a third node in the output layer, and a second threshold of the third node; inputting the sample aging characteristic parameters into an input layer aiming at each sample aging characteristic parameter, and acquiring a first output result of the sample aging characteristic parameters in a hidden layer according to a first weight, a first threshold and a preset activation function; and obtaining a second output result of the sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, and obtaining a sample prediction aging parameter corresponding to the sample aging characteristic parameter.

In one embodiment, the training module is specifically configured to obtain a root mean square error value of the sample predicted aging parameter and the corresponding real aging parameter; obtaining a corresponding prediction set correlation coefficient according to the square of the difference between the sample prediction aging parameter and the sample real aging parameter and the square of the difference between the sample real aging parameter and the average value of the plurality of sample real aging parameters; the root mean square error value and the prediction set correlation coefficient are taken as error values.

In one embodiment, the training module is specifically configured to obtain a corresponding error function according to a sample predicted aging parameter and a sample actual aging parameter; and according to the error function and the preset activation function, adjusting a first weight, a first threshold, a second weight and a second threshold based on a Widry-Hoff learning rule and a momentum gradient descent algorithm.

In one embodiment, the prediction module is specifically configured to query an aging stage table according to a predicted raman spectrum, and determine a predicted aging stage corresponding to insulation of the oilpaper to be predicted; the aging stage table comprises a plurality of correspondence relations between the oil paper insulation aging stages and the Raman spectrogram.

The specific limitation of the oil paper insulation aging prediction apparatus may be referred to the limitation of the oil paper insulation aging prediction method hereinabove, and will not be described herein. The above-mentioned various modules in the oiled paper insulation aging prediction device may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a terminal, and the internal structure of which may be as shown in fig. 6. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The computer program when executed by a processor implements a oiled paper insulation degradation prediction method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structure shown in fig. 6 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor implementing the oiled paper insulation degradation prediction method described above when executing the computer program.

In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor, implements the oiled paper insulation degradation prediction method described above.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, or the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.

Claims

1. A method for predicting oiled paper insulation aging, the method comprising:

obtaining aging characteristic parameters to be predicted corresponding to the insulation of the oil paper to be predicted; the aging characteristic parameters to be predicted comprise effective characteristics in a Raman spectrum corresponding to the oil paper insulation to be predicted;

inputting the aging characteristic parameters to be predicted into a target aging prediction model, and obtaining predicted aging parameters output by the target aging prediction model; the predicted aging parameters comprise predicted Raman spectrograms corresponding to the aging characteristic parameters to be predicted; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters;

According to the predicted aging parameters, determining a predicted aging stage corresponding to the oiled paper insulation to be predicted, including: inquiring an aging stage table according to the predicted Raman spectrogram, and determining a predicted aging stage corresponding to the insulation of the oil paper to be predicted; the aging stage table comprises a plurality of correspondence relations between oil paper insulation aging stages and a Raman spectrogram;

the target aging prediction model is obtained through the following steps:

inputting the plurality of sample aging characteristic parameters into the input layer, and obtaining the plurality of sample aging parameter to be trained based on the output of the hidden layer and the output layer by the aging prediction model to be trained comprises the following steps: acquiring a first weight corresponding to a first node in the input layer and a second node in the hidden layer, and a first threshold of the second node; acquiring a second weight corresponding to a second node in the hidden layer and a third node in the output layer, and a second threshold of the third node; inputting the sample aging characteristic parameters into the input layer aiming at each sample aging characteristic parameter, and acquiring a first output result of the sample aging characteristic parameters in the hidden layer according to the first weight, the first threshold and a preset activation function; obtaining a second output result of the sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, and obtaining a sample prediction aging parameter corresponding to the sample aging characteristic parameter; the hidden layer and the output layer each comprise a plurality of nodes; the plurality of sample prediction aging parameters are obtained according to a first output result and the output of each node in the output layer; the first output result is obtained according to the output of each node in the hidden layer;

if not, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained according to the error value, and returning to the step of inputting the aging characteristic parameters of the plurality of samples into the input layer, wherein the step comprises the following steps: reversely transmitting the error value to the hidden layer and the input layer, spreading the error value to each node in the hidden layer and the input layer to obtain error signals of each node, respectively adjusting weights among layers and thresholds of each layer in the aging prediction model to be trained based on the error signals, and returning to the step of inputting the aging characteristic parameters of the samples into the input layer; the step of respectively adjusting the weights of all layers in the aging prediction model to be trained based on the error signals comprises the following steps: determining a corresponding error function based on the error signal between the output layer and the hidden layer, and determining a second weight adjustment parameter between each node of the output layer and each node of the hidden layer based on the partial derivative of the current second weight between each node of the output layer and each node of the hidden layer and the partial derivative of the error function, wherein the second weight adjustment parameter comprises the product of a second output error determined based on the error signal and a preset activation function and a second input vector of each node of the current layer; determining the adjusted second weight corresponding to each node of the current output layer and each node of the hidden layer according to the current second weight, the second output error, the second input vector of each node of the current layer, a preset momentum coefficient and a previous second weight adjustment parameter of a preset value; the preset value is smaller than or equal to the value of the previous second weight adjustment parameter;

Determining first weight adjustment parameters between each node of the input layer and each node of the hidden layer according to the partial derivative of the current first weight between each node of the input layer and each node of the hidden layer and the partial derivative of the error function, wherein the first weight adjustment parameters comprise products of first output errors determined based on first output errors, the current first weight and a preset activation function and first input vectors of each node of the current layer; determining the adjusted first weight corresponding to each node of the current input layer and each node of the hidden layer according to the current first weight, the first output error, the first input vector of each node of the current layer, a preset momentum coefficient and a first weight adjustment parameter of a preset value; the preset value is smaller than or equal to the value of the first weight adjustment parameter of the previous time;

2. The method of claim 1, wherein the obtaining the plurality of sample real aging parameters corresponding to the plurality of sample oiled paper insulation comprises:

3. The method of claim 1, wherein obtaining error values for each sample predicted aging parameter and the corresponding sample true aging parameter comprises:

4. The method according to claim 1, wherein the adjusting weights between layers and thresholds of layers in the aging prediction model to be trained according to the error values includes:

5. An oiled paper insulation degradation prediction apparatus, the apparatus comprising:

the acquisition module is used for acquiring aging characteristic parameters to be predicted corresponding to the insulation of the oil paper to be predicted; the aging characteristic parameters to be predicted comprise effective characteristics in a Raman spectrum corresponding to the oil paper insulation to be predicted;

the input module is used for inputting the aging characteristic parameters to be predicted into a target aging prediction model and obtaining predicted aging parameters output by the target aging prediction model; the predicted aging parameters comprise predicted Raman spectrograms corresponding to the aging characteristic parameters to be predicted; the target aging prediction model is obtained by training based on a genetic deviation least square method, a plurality of sample oil paper insulation sample aging characteristic parameters and sample real aging parameters;

the prediction module is used for determining a predicted aging stage corresponding to the oil paper insulation to be predicted according to the predicted aging parameter, and particularly used for querying an aging stage table according to the predicted Raman spectrogram and determining the predicted aging stage corresponding to the oil paper insulation to be predicted; the aging stage table comprises a plurality of correspondence relations between oil paper insulation aging stages and a Raman spectrogram;

Further comprises: the training module is used for acquiring a plurality of sample aging characteristic parameters corresponding to the plurality of sample oilpaper insulations and acquiring a plurality of sample real aging parameters corresponding to the plurality of sample oilpaper insulations;

inputting the plurality of sample aging characteristic parameters into the input layer, and acquiring a plurality of sample aging prediction parameters output by the aging prediction model to be trained based on the hidden layer and the output layer, wherein the sample aging prediction parameters are specifically used for acquiring a first weight corresponding to a first node in the input layer and a second node in the hidden layer and a first threshold of the second node; acquiring a second weight corresponding to a second node in the hidden layer and a third node in the output layer, and a second threshold of the third node; inputting the sample aging characteristic parameters into the input layer aiming at each sample aging characteristic parameter, and acquiring a first output result of the sample aging characteristic parameters in the hidden layer according to the first weight, the first threshold and a preset activation function; obtaining a second output result of the sample aging characteristic parameter at the output layer according to the second weight, the second threshold, the first output result and the preset activation function, and obtaining a sample prediction aging parameter corresponding to the sample aging characteristic parameter; the hidden layer and the output layer each comprise a plurality of nodes; the plurality of sample prediction aging parameters are obtained according to a first output result and the output of each node in the output layer; the first output result is obtained according to the output of each node in the hidden layer;

if not, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained according to the error value, returning to the step of inputting the plurality of sample aging characteristic parameters into the input layer, specifically, reversely transmitting the error value to the hidden layer and the input layer, and distributing the error value to each node in the hidden layer and the input layer to obtain an error signal of each node, respectively adjusting the weight between layers and the threshold value of each layer in the aging prediction model to be trained based on the error signal, and returning to the step of inputting the plurality of sample aging characteristic parameters into the input layer; the method specifically comprises the steps of determining a corresponding error function based on the error signal between an output layer and a hidden layer, and determining a second weight adjustment parameter between each node of the output layer and each node of the hidden layer based on the partial derivative of a current second weight between each node of the output layer and each node of the hidden layer and the partial derivative of the error function, wherein the second weight adjustment parameter comprises the product of a second output error determined based on the error signal and a preset activation function and a second input vector of each node of the current layer; determining the adjusted second weight corresponding to each node of the current output layer and each node of the hidden layer according to the current second weight, the second output error, the second input vector of each node of the current layer, a preset momentum coefficient and a previous second weight adjustment parameter of a preset value; the preset value is smaller than or equal to the value of the previous second weight adjustment parameter;

6. The device according to claim 5, wherein the training module is specifically configured to:

7. The device according to claim 5, wherein the training module is specifically configured to:

8. The device according to claim 5, wherein the training module is specifically configured to:

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 4 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 4.