Conflict multi-data source-based main insulation material life prediction method

Abstract

The application provides a main insulation material life prediction method based on conflict multi-data sources, and relates to the field of material life prediction. The method comprises the following steps: obtaining two data sets of the main insulation material, which are a residual breakdown voltage historical data set and a maximum local discharge amount field data set; determining two variable points based on the two data sets respectively; further constructing three evidences based on the sample capacity, data type of the two data sets and the variable point estimation model of the two variable points; determining the probability distribution numbers of the three evidences to the two variable points; taking one of the two variable points as a reference variable point based on the probability distribution numbers of the three evidences to the two variable points; and determining the cumulative distribution function of the main insulation material life with the reference variable point as a boundary. The method can effectively fuse historical data and field data, and improve the prediction accuracy of the main insulation material life.

Description

Technical Field

[0001] The present invention relates to the field of electrical equipment insulation life prediction, and in particular to a main insulation material life prediction method based on conflicting multiple data sources. Background Art

[0002] In modern industry, electrical equipment, such as motors, is critical, and its reliability is directly linked to the stability and safety of production systems. During operation, the primary insulation system of a motor gradually degrades due to long-term exposure to factors such as thermal stress, mechanical stress, and chemical corrosion. This degradation can lead to a decrease in the performance of the primary insulation material, ultimately causing electrical failures and significant economic losses.

[0003] Related art methods for predicting the lifespan of motor main insulation materials primarily rely on historical data obtained from laboratory accelerated aging tests, using only this data to establish a common model for predicting the lifespan of the main insulation materials. However, the main insulation materials of different electrical equipment inevitably vary, resulting in different lifespan values ​​for each device. Using this common characterization model to estimate lifespan has significant limitations, resulting in significant deviations from actual results. Summary of the Invention

[0004] In order to address the deficiencies of the prior art, the purpose of this application is to provide a main insulation material life prediction method based on conflicting multiple data sources, which can improve the accuracy and reliability of the main insulation material life prediction.

[0005] In a first aspect, the present application provides a method for predicting the life of a main insulating material based on conflicting multiple data sources, the method comprising:

[0006] Get the main insulation material dataset and datasets , dataset The remaining breakdown voltage history dataset, dataset This is the maximum partial discharge field data set;

[0007] Based on the dataset Determine the change point , based on the dataset Determine the change point , if the change point and change points Different, execute steps a, b and c; change point and Both represent the turning points of the degradation rate of the main insulation material performance;

[0008] Step a: Consider the dataset and datasets Different sample sizes, building sample size evidence and confirm Change Point The probability distribution number and change points The probability distribution number , Representation dataset Sample size versus change point The level of trust, Representation dataset Sample size versus change point level of trust;

[0009] Step b: Consider the dataset and datasets Different data types, build data type evidence and confirm Change Point The probability distribution number and change points The probability distribution number , Representation dataset Data type change point The level of trust, Representation dataset Data type change point level of trust;

[0010] Step c: Consider change points and change points Using different change point estimation models, constructing change point estimation model evidence and confirm Change Point The probability distribution number and change points The probability distribution number , Representation dataset Estimation model for change points The level of trust, Representation dataset Estimation model for change points level of trust;

[0011] Evidence-based 、 as well as Change Point and The probability distribution numbers of each change point and change points One of the reference change points , determine the benchmark change point is the cumulative distribution function of the main insulation material life of the boundary.

[0012] In one embodiment, based on evidence 、 as well as Change Point and The probability distribution numbers of each change point and change points One of the reference change points ,include:

[0013] Converging evidence and evidence , and obtain the first fusion evidence Change Point The probability distribution number and change points The probability distribution number ;

[0014] Converging evidence and evidence , and obtain the second fusion evidence Change Point The probability distribution number and change points The probability distribution number ;

[0015] like Greater than The change point As a benchmark change point ,like Less than The change point As a benchmark change point .

[0016] In one embodiment, the fusion evidence and evidence , and obtain the first fusion evidence Change Point The probability distribution number and change points The probability distribution number ,include:

[0017] Evidence-based and evidence Change Point and The probability distribution numbers of each, determine the evidence With evidence Conflict coefficient after fusion ;

[0018] Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the first fusion evidence Change Point The probability distribution number ;

[0019] Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the first fusion evidence Change Point The probability distribution number .

[0020] In one embodiment, the fusion evidence and evidence , and obtain the second fusion evidence Change Point The probability distribution number and change points The probability distribution number ,include:

[0021] Evidence-based and evidence Change Point and The probability distribution number of each is used to determine the conflict coefficient ;

[0022] Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the second fusion evidence Change Point The probability distribution number ;

[0023] Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the second fusion evidence Change Point The probability distribution number .

[0024] In one embodiment, determining Change Point The probability distribution number and change points The probability distribution number ,include:

[0025] Get the dataset The sample size N and the data set The sample size N t , where the sample size N t The size of N changes with the monitoring time. The longer the monitoring time, the larger the sample size N. t The larger the sample size N is, the t is a quantity that changes with monitoring time;

[0026] Determine the sample size N and sample size N t The logarithm of the product, where the ratio of the logarithm of the sample size N to the logarithm of the product is Change Point The probability distribution number , sample size N t The ratio of the logarithm of the product is Change Point The probability distribution number .

[0027] In one embodiment, determining Change Point The probability distribution number and change points The probability distribution number ,include:

[0028] Based on the dataset Determine historical data change points based on the data set Determine the change point of field data; among them, the change point It is determined based on the change point of historical data. It is determined based on the change points of field data;

[0029] The dependency relationship between historical data change points and field data change points is established through the objective function; r is the relevant parameter in the objective function, which is used to express the degree of linear correlation between the change points of historical data and the change points of field data;

[0030] Estimate relevant parameters by maximum likelihood estimation r , get the relevant parameters r Estimated value ;

[0031] Based on estimates Sure Change Point The probability distribution number and change points The probability distribution number .

[0032] In one embodiment, determining Change Point The probability distribution number and change points The probability distribution number ,include:

[0033] The location of the historical data change point follows the normal distribution, and the distribution interval is divided into k subintervals, calculate the location of the historical data change point within the interval c i The probability of ,in, i ={1,2,…, k};

[0034] The field data change point is the data set through the EWMA control chart The false alarm probability of the field data change point is determined by the central limit theorem. and the probability of missing data change points ; Among them, false positives include the EWMA control chart incorrectly judging that a change point has occurred when there is no change point, and false negatives include the EWMA control chart failing to detect a change point when a change point has occurred;

[0035] Based on probability and probability Determine the correct probability of the EWMA control chart judging the change point , based on probability and probability Sure Change Point The probability distribution number and change points The probability distribution number .

[0036] In one embodiment, based on the data set Determine the change point ,include:

[0037] Setting up the dataset , Q n ( t n ) means in t n The maximum partial discharge monitored at any time;

[0038] determining growth rates of multiple maximum partial discharge amounts based on the maximum partial discharge amounts monitored at each moment;

[0039] Calculate the EWMA value at each moment based on the growth rate, and determine whether a change point occurs at the corresponding moment based on the EWMA value;

[0040] Based on the determined multiple change point moments, determine the data set The corresponding change point .

[0041] In one embodiment, calculating the EWMA value at each moment based on the growth rate, and determining whether a change point occurs at the corresponding moment based on the EWMA value, includes:

[0042] For the growth rate at any moment, substitute the growth rate into the recursive formula to obtain the EWMA value; the recursive formula is: ;in, l represents the smoothing factor, 0< l ≤1, express The EWMA value at the moment, express The growth rate of time, express EWMA value at the moment;

[0043] Calculate the target growth rate and the standard deviation of the EWMA value, and determine the upper and lower control limits based on the target growth rate and standard deviation;

[0044] when When the value exceeds the upper control limit or falls below the lower control limit, a change point is determined.

[0045] In a second aspect, the present application also provides a computer device for predicting the life of main insulating materials, the computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the first aspect's method for predicting the life of main insulating materials based on conflicting multiple data sources when executing the computer program.

[0046] This method obtains a dataset of the main insulation materials and datasets , where the dataset The remaining breakdown voltage history dataset, dataset This is the maximum partial discharge field data set. Based on the data set Determine the change point , based on the dataset Determine the change point If the change point and change points If different, then execute step a, step b and step c, that is, construct the sample capacity evidence, sample data type evidence and change point estimation model evidence, and calculate the effect of these three kinds of evidence on the change point and change points The probability distribution numbers of each. Based on the above three kinds of evidence, the change point and change points Each probability distribution number, select one of the change points as the benchmark change point, use the benchmark change point as the boundary, and determine the benchmark change point. The cumulative distribution function of the main insulation material life is divided into two parts by the change point. This method can construct a cumulative distribution function by dividing the historical data and the field data into two parts with the change point as the boundary, thus resolving the conflict between the historical data and the field data and achieving a more accurate prediction of the life of the main insulation individual. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a method for predicting the life of a main insulating material based on conflicting multiple data sources in one embodiment;

[0048] Figure 2 In one embodiment, based on a data set Determine the change point Flowchart of

[0049] Figure 3 Computing evidence for an embodiment m 1. m 2 and m 3 Flowchart of probability distribution number;

[0050] Figure 4 In one embodiment, based on a data set Determine the change point ;

[0051] Figure 5 For an embodiment, Change Point The probability distribution number and change points The probability distribution number Flowchart of

[0052] Figure 6 A flowchart for determining a benchmark change point in one embodiment;

[0053] Figure 7 A flowchart of obtaining the probability distribution number of the first fusion evidence pair change point in one embodiment;

[0054] Figure 8 A flowchart of obtaining the probability distribution number of the second fusion evidence pair change point in one embodiment;

[0055] Figure 9 A flowchart of determining a regression function in one embodiment;

[0056] Figure 10 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this application more clear, the present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application. Unless otherwise defined, the technical terms or scientific terms involved in this application should have the general meaning understood by people with ordinary skills in the technical field to which this application belongs.

[0058] In one embodiment, Figure 1 As shown, a method for predicting the life of a main insulating material based on conflicting multiple data sources is provided, and the method includes the following steps:

[0059] Step 101: Get the main insulation material dataset and datasets , dataset The remaining breakdown voltage history dataset, dataset This is the field data set of maximum partial discharge.

[0060] Dataset The remaining breakdown voltage dataset, dataset May include residual breakdown voltage data; data sets The maximum partial discharge field data set, data set It can include discharge data. It should be noted that the data set This data set can be obtained by conducting accelerated degradation tests on main insulation materials (such as main insulation materials of motors) at different accelerated temperature stress levels; The data set can be obtained in the actual operating environment. It can reflect the degradation of the main insulation material under working conditions.

[0061] Step 102: Based on the dataset Determine the change point , based on the dataset Determine the change point , if the change point and change points Different, execute steps a, b and c; change point and Both indicate the turning points of the degradation rate of the main insulation material performance.

[0062] Specifically, such as Figure 2 As shown, based on the data set Determine the change point The steps include: selecting N main insulating materials as test samples to conduct a constant stress accelerated degradation test, setting n accelerated temperature stresses, dividing the N main insulating materials into n groups, and recording the corresponding n groups under the kth accelerated temperature stress. k Test samples are subjected to constant stress accelerated degradation tests on the corresponding test samples under each accelerated temperature stress, and N groups of accelerated degradation data are obtained. Each group of accelerated degradation data corresponds to the accelerated degradation data sequence of the main insulation material under the kth accelerated temperature stress and based on each measurement time, and k ranges from 1 to n. Historical data set It includes N groups of accelerated degradation data, i.e., N groups of residual breakdown voltage. For example, if 4 accelerated temperature stresses are set, N is divided into 4 groups, with n1 samples under the first accelerated temperature stress, n2 samples under the second accelerated temperature stress, n3 samples under the third accelerated temperature stress, and n4 samples under the fourth accelerated temperature stress. Under each accelerated temperature stress, the corresponding test sample is subjected to a constant stress accelerated degradation test, i.e., under the kth accelerated temperature stress, the n samples are subjected to a constant stress accelerated degradation test. k The stress accelerated degradation test is carried out on the test samples, and degradation data representing performance degradation is obtained at each measurement moment. The degradation data is based on the accelerated degradation data sequence at each measurement moment, and N groups of accelerated degradation data are obtained. u ijk represents the data obtained from the i-th measurement of the j-th test sample under the k-th stress, t ijk represents the measurement time of the i-th measurement of the j-th test sample under the k-th stress, Δ u ijk = u ijk – u (i-1)jk represents the degradation increment, Δ t ijk = t ijk – t (i-1)jk represents the time increment, where i=1,2,3…,n jk ,j=1,2,3…,n k , k=1,2,3…,n,n jk represents the kth accelerated temperature stress of the jth specimen, n k It represents the number of specimens under the kth accelerated temperature stress, and n is the number of accelerated temperature stresses.

[0063] Furthermore, based on the dataset Determine the change point The step S2 is also included: establishing a two-stage Wiener process degradation model based on each set of accelerated degradation data, determining the interval of the change point in the degradation process of each main insulation material through SIC, obtaining the estimated value of the change point based on the minimum residual square sum criterion, and then obtaining the k-th accelerated temperature stress under n k n of the main insulating material k The change point positions under each accelerated temperature stress are assumed to obey the two-parameter Weibull distribution. k The maximum likelihood estimation method is used to obtain the change point distribution parameters under the accelerated temperature stress, and then n groups of change point distribution parameters are obtained. For example, it can be seen from the above embodiment that under the first accelerated temperature stress, n1 samples have n1 change point positions, under the second accelerated temperature stress, n2 samples have n2 change point positions, under the third accelerated temperature stress, n3 samples have n3 change point positions, and under the fourth accelerated temperature stress, n4 samples have n4 change point positions. Therefore, the change point position of each main insulating material under each accelerated temperature stress can be obtained, and a total of n*n k There are n*n change point positions k A historical data change point.

[0064] Furthermore, based on the dataset Determine the change point The method further includes step S3: constructing an accelerated model of the change point distribution parameters according to the change law between the change point distribution parameters and the accelerated temperature stress, establishing a matrix equation between the change point distribution parameters and the accelerated temperature stress, obtaining a least squares estimate by solving the matrix equation, and obtaining the change point distribution parameter value in the two-stage degradation process under normal temperature stress according to the least squares estimate, and obtaining a probability density function of the change point under normal temperature stress, and determining the change point with the largest probability value under normal temperature stress as the change point based on the probability density function. , that is, the position corresponding to the change point with the largest probability value under normal temperature stress is the change point location.

[0065] Specifically, the probability density function of the change point under normal temperature stress is:

[0066] ; where μ τ is the mean of the historical data change points, σ τ is the standard deviation of the historical data change points;

[0067] , . Where T0 represents the extrapolated normal temperature.

[0068] The matrix equation is: ;

[0069] wherein, denotes μ of the change point under K temperature stresses following a normal distribution τ、 σ τ maximum likelihood estimate, T n denotes the accelerated temperature stress.

[0070] The least squares estimate is obtained by solving the matrix equation The least squares estimate is determined as the change point distribution parameter value in the two-stage degradation process under normal temperature stress, i.e. , , and .

[0071] Step a: Consider the data set and the data set The sample size evidence is constructed and the probability assignment number of the change point and the probability assignment number of the change point , denotes the degree of trust of the sample size of the data set to the change point , denotes the degree of trust of the sample size of the data set to the change point . Specifically, the flowcharts of steps a, b and c are shown in The purpose of constructing the sample size evidence

[0072] is to evaluate the degree of trust of the sample size of the data set Figure 3 and the data set to the change point position estimate. It should be noted that the data set with large sample size may dominate the analysis, thereby masking the characteristics of the data set with small sample size. The characteristics of the small sample data set may be ignored due to insufficient data. Therefore, the data set with large sample size usually provides more reliable statistical estimates, and thus the estimate of the change point position is more reliable. By converting the sample size of the data set into the degree of trust, the contribution of each data set to the estimate of the change point position can be quantified. The probability assignment number denotes the degree of trust of the sample size of the data set to the change point , and the probability assignment number denotes the degree of trust of the sample size of the data set to the change point .

[0073] ​​Step b: Consider the dataset and datasets Different data types, build data type evidence and confirm Change Point The probability distribution number and change points The probability distribution number , Representation dataset Data type change point The level of trust, Representation dataset Data type change point level of trust.

[0074] Specifically, the difference in sample data types can refer to the different physical quantities represented by the residual breakdown voltage and the maximum partial discharge. The change point location essentially reflects the turning point of the change in the degradation rate of the main insulation performance, while the sample data type can reflect the degradation pattern of the physical quantity in the sample set. Taking the field data of maximum partial discharge as an example, this sample data type reflects the change point location in the degradation process of the maximum partial discharge, but the extent to which the change point location in the maximum partial discharge degradation trajectory can reflect the change point location in the degradation process of the main insulation performance is uncertain. Therefore, the difference in sample type between the two data sets will have a direct impact on the estimation of the change point location.

[0075] Constructing evidence of sample data types m 2 and calculate its probability distribution number. m 2, we need to calculate the probability distribution number separately. and probability distribution number Among them, the probability distribution number Representation dataset Data type change point The degree of trust, probability distribution number Representation dataset Data type change point Among them, the maximum partial discharge data can also reflect the aging of the main insulation material, but its direct correlation with the inflection point position is not as strong as the residual breakdown voltage data, so its trust level is relatively low.

[0076] Step c: Consider change points and change points Using different change point estimation models, constructing change point estimation model evidence and confirm Change Point The probability distribution number and change points The probability distribution number , Representation dataset Estimation model for change points The level of trust, Representation dataset Estimation model for change points level of trust.

[0077] Specifically, model estimation can refer to an estimated value of the change point position obtained through statistical analysis of the sample data set, and the estimated value is related to the accuracy of the model estimation. Taking the residual breakdown voltage historical data as an example, the distribution function of the change point position is obtained by fitting the residual breakdown voltage data. The change point position estimate under a certain confidence level can be obtained through the change point distribution function. This estimate is a probabilistic estimate, and its uncertainty is introduced in the model building process. Therefore, the uncertainty brought by the model estimation of the two sample sets is another important source of uncertainty information of the change point position. It should be noted that for the data set , the change point can be estimated using the two-stage Wiener process model ; For the dataset , the EWMA control chart model can be used to estimate the change point .

[0078] Among them, the probability distribution number Representation dataset Estimation model for change points The degree of trust, probability distribution number Representation dataset Estimation model for change points level of trust. Change Point of Express evidence Based on the dataset The confidence level of the change point estimation model of the sample. It can reflect the degradation in the actual operating environment, but may be affected by more external interference and uncertainty factors, but it can provide information that is closer to the actual operating status.

[0079] It should be noted that the probability distribution number For example, It does not indicate evidence Based on the dataset The probability of the sample appearing is not the probability of the sample appearing, but the probability of the sample appearing. Estimation model for change points Therefore, the probability distribution number is essentially different from probability. This ability to process fuzzy information enables the probability distribution number to perform reasoning and analysis on uncertainty problems.

[0080] Step 103: Evidence-Based 、 as well as Change Point and The probability distribution numbers of each change point and change points One of the reference change points , determine the benchmark change point is the cumulative distribution function of the main insulation material life of the boundary.

[0081] In the process of predicting the life of main insulation materials, evidence needs to be integrated 、 as well as Change Point and The probability distribution numbers of each determine the benchmark change point . Evidence-based 、 as well as Change Point and Each probability distribution number evaluates the change point and The credibility of the evidence It can reflect the impact of sample size on change point estimation. The change point estimated by the data set with large sample size is more credible. It can indicate the degree of confidence in the sample data type. The residual breakdown voltage data usually reflects the insulation performance more directly and has higher credibility. It can involve the confidence level of the change point estimation model, the dataset S Model 1 is more credible due to the high quality of the data.

[0082] Combining three pieces of evidence 、 as well as , the change point can be determined and The overall credibility of the change point is selected as the benchmark change point. For example, if The credibility is higher, you can choose As a benchmark change point Otherwise, select As a benchmark change point . Based on the benchmark change point To demarcate, the cumulative distribution function of the lifetime of the first stage degradation model and the cumulative distribution function of the lifetime of the second stage degradation model are determined.

[0083] For example, the cumulative distribution function of the life of the two-stage degradation model can be updated separately using the Bayesian method (the cumulative distribution function of the life of the two-stage degradation model includes the cumulative distribution function of the life of the first-stage degradation model and the cumulative distribution function of the life of the second-stage degradation model). Used to determine the boundary between the cumulative distribution function of the lifetime of the first stage degradation model and the cumulative distribution function of the lifetime of the second stage degradation model. Specifically, the cumulative distribution function of the life of the two-stage degradation model can be obtained by the following method: constructing a two-stage Wiener process degradation model for N groups of accelerated degradation data, constructing an accelerated degradation model between the drift parameter and the diffusion parameter and the accelerated temperature stress in the first stage of the Wiener process, establishing the likelihood function of the first stage based on all accelerated degradation data before each main insulation material change point, and solving to obtain the maximum likelihood estimate of the first stage; in the second stage of the Wiener process, constructing an accelerated degradation model between the drift parameter and the diffusion parameter and the accelerated temperature stress, establishing the likelihood function of the second stage based on all accelerated degradation data after each main insulation material change point, and solving to obtain the maximum likelihood estimate of the second stage; based on the maximum likelihood estimate of the first stage and the maximum likelihood estimate of the second stage, the drift parameters and diffusion parameters of the two stages in the two-stage Wiener process degradation model under normal temperature stress are obtained, combined with the probability density function of the change point under normal temperature stress, and based on the preset performance parameter failure threshold, the cumulative distribution function of the product life under normal temperature stress (i.e., the cumulative distribution function of the life of the two-stage degradation model of the main insulation material) is obtained.

[0084] The Bayesian method is used to update the cumulative distribution function of the life of the two-stage degradation model respectively. Specifically, the Bayesian method can combine prior knowledge (such as the prior distribution of degradation model parameters obtained from historical data) and field data to calculate the posterior distribution of the two-stage degradation model parameters, and update the parameters of the cumulative distribution function of the life of the first-stage degradation model and the second-stage degradation model respectively. By updating the parameters, the updated life of the benchmark change point is obtained. is the cumulative distribution function of the product life under normal temperature stress, that is, the cumulative distribution function of the life of the first stage degradation model and the cumulative distribution function of the life of the second stage degradation model.

[0085] In this embodiment, the method obtains a data set of the main insulation material and datasets , where the dataset The remaining breakdown voltage history dataset, dataset This is the maximum partial discharge field data set. Based on the data set Determine the change point , based on the dataset Determine the change point If the change point and change points If different, then execute step a, step b and step c, that is, construct the sample capacity evidence, sample data type evidence and change point estimation model evidence, and calculate the effect of these three kinds of evidence on the change point and change points The probability distribution numbers of each. Based on the above three kinds of evidence, the change point and change points Each probability distribution number, select one of the change points as the benchmark change point, use the benchmark change point as the boundary, and determine the benchmark change point. The cumulative distribution function of the main insulation material life is divided into two parts by the change point. This method can construct a cumulative distribution function by dividing the historical data and the field data into two parts with the change point as the boundary, thus resolving the conflict between the historical data and the field data and achieving a more accurate prediction of the life of the main insulation individual.

[0086] In one embodiment, Figure 4 As shown, based on the data set Determine the change point , including the following steps:

[0087] Step 401: Set the data set , Q n ( t n ) means in t n The maximum partial discharge monitored at any time;

[0088] Step 402: Determine the growth rate of the maximum partial discharge , ;

[0089] Step 403: Based on the growth rate Calculating EWMA values Z t , and based on Z t Determine whether a change point occurs.

[0090] Step 404: Monitor multiple main insulation materials, determine multiple field data change points, and determine a data set based on the locations of the multiple determined change points. The corresponding change point (According to the normal distribution composed of multiple change point positions, the change point with the largest probability value can be taken as the change point ).

[0091] Among them, based on the growth rate Calculating EWMA values Z t , and based on Z t Determine whether a change point has occurred, including: Substituting into the recursive formula we get , the recursive formula is: ;in, l represents the smoothing factor, 0< l ≤1, express The EWMA value at the moment; determine the target growth rate and the standard deviation of the EWMA value, and determine the upper control limit and the lower control limit based on the target growth rate and the standard deviation; when When the value exceeds the upper control limit or falls below the lower control limit, a change point is determined.

[0092] Specifically, different from the dataset S 1. Dataset S The biggest feature of 2 is its real-time variability. From the moment of field data collection, the data set S 2 is updated over time. That is, the data set S 2 Unlike the dataset S 1 Generally, the initial inclusion is the change point The complete dataset, dataset S 2 change point is present over time. Therefore, for the dataset S 2 Change Points Detection requires real-time monitoring of abnormal changes in field data. The EWMA (Exponentially Weighted Moving Average) control chart offers significant advantages in monitoring process change points. Therefore, EWMA is used to monitor change points in field data.

[0093] Assume that the time series data of the maximum partial discharge monitored online is:

[0094] ;

[0095] in, Q n ( t n ) means in t n The maximum partial discharge value monitored at all times. Since the degradation rate of the main insulation before and after the change point is significantly different, the second change point is determined by monitoring the growth rate of the maximum partial discharge value. Location.

[0096] The data obtained from online monitoring can be used to obtain the growth rate of data within a unit sampling interval. G t for:

[0097] ;

[0098] The goal of the EWMA control chart is to monitor G t changes, identifying the changes from the first stage (assuming the average growth rate is m 1) Transition to the second stage (assuming an average growth rate of m 2) Change point .

[0099] The core of the EWMA control chart is to calculate the EWMA value Z t , which is defined as:

[0100] ;

[0101] in, l is the smoothing factor, 0< l ≤1, determines the weight of new data and historical data; Z 0 is usually set as the average growth rate in the first stage m 1.

[0102] EWMA monitors whether the entire process deviates from the target growth rate by setting control limits m 1. The control limits are calculated using the following formula:

[0103] ;

[0104] .

[0105] in, UCL represents the upper control limit, LCL represents the lower control limit; L is a constant, L The choice of will affect the confidence level of the EWMA control chart monitoring results; s Z is the standard deviation of the EWMA values:

[0106] ;

[0107] in s is the standard deviation of the data growth rate.

[0108] When the EWMA value Z t Exceed UCL or lower LCL WhenG t Significant changes may occur, indicating the presence of a change point The cumulative effect of EWMA on continuous small deviations is used to make the long-term accumulation of small deviations more effective. Z t Beyond the control limit, thus achieving the change point Detection.

[0109] In one embodiment, determining Change Point The probability distribution number and change points The probability distribution number ,include:

[0110] Get the dataset The sample size N and the data set The sample size N t , where the sample size N t The size of N changes with the monitoring time. The longer the monitoring time, the larger the sample size N. t The larger the sample size N is, the t is a quantity that changes with monitoring time;

[0111] Determine the sample size N and sample size N t The logarithm of the product, where the ratio of the logarithm of the sample size N to the logarithm of the product is Change Point The probability distribution number , sample size N t The ratio of the logarithm of the product is Change Point The probability distribution number .

[0112] Specifically, get the dataset The sample size is N, the data set The sample size is N t , N t The quantity that changes with monitoring time. The longer the monitoring time, the larger the sample capacity N. t The bigger.

[0113] Computational evidence Change Point and When the confidence level is calculated, the formula is used:

[0114] , .

[0115] Among them, ln( N ) and ln( Nt ) represent the data sets and The natural logarithm of the sample size, is the natural logarithm of the product of the two, which serves as a normalization factor to ensure and The sum of is 1.

[0116] In one embodiment, Figure 5 As shown, confirm Change Point The probability distribution number and change points The probability distribution number , including the following steps:

[0117] Step 501: Based on the data set Determine historical data change points based on the data set Determine the change point of field data; among them, the change point It is determined based on the change point of historical data. It is determined based on the change points of field data;

[0118] Step 502: Establishing the dependency relationship between historical data change points and field data change points through the objective function; wherein, r is the relevant parameter in the objective function, which is used to express the degree of linear correlation between the change points of historical data and the change points of field data;

[0119] Step 503: Estimate relevant parameters by maximum likelihood estimation r , get the relevant parameters r Estimated value ;

[0120] Step 504: Determine based on the estimated value Change Point The probability distribution number and change points The probability distribution number .

[0121] Specifically, due to the dataset Can reflect the degradation process of main insulation performance and give it credibility R 1; among them, Dataset The measured discharge capacity cannot fully reflect its degradation process. In order to determine the change point To determine the extent to which the position of the residual breakdown voltage change point can be reflected, the correlation between the two groups of change point positions can be analyzed using the Copula function. Copula can effectively model the correlation between variables with specific marginal distributions.

[0122] Assume that the random variables corresponding to the two sets of change point positions are X and Y , the former represents the change point Position, which represents the change point Position. The change point position is actually a random variable that obeys the normal distribution, and its cumulative distribution functions are:

[0123] ;

[0124] .

[0125] in, F X ( x ) is a random variable X The cumulative distribution function of X Less than or equal to x probability. F Y ( y ) is a random variable Y The cumulative distribution function of Y Less than or equal to y probability. m X and m Y They are random variables X and Y The shape parameter determines the shape of the distribution. or X and or Y They are random variables X and Y The scale parameter determines the scale of the distribution.

[0126] To separate marginal distributions from dependency structures, the original data is transformed into variables on a uniform distribution. Definition:

[0127] ;

[0128] .

[0129] According to the probability integral transformation theorem, U The cumulative distribution function of is:

[0130] .

[0131] in, u Is a probability value, representing a random variable X Cumulative distribution function of F X (X ) is less than or equal to u probability.

[0132] Can get U ~Uniform(0,1). Based on the same steps, we can get V ~Uniform(0,1).

[0133] Therefore, through and The change point location data corresponding to the two data sets can obtain a set of independent and identically distributed observation data , where each u i = F X ( x i ),v i = F Y ( y ) is a uniform distribution variable obtained by marginal distribution transformation.

[0134] The objective function (Copula function) is used to describe the dependency between the change points of historical data and the change points of field data. The Copula function is based on the multivariate normal distribution and is suitable for describing the linear correlation between the change points. The Copula function can be expressed as:

[0135] ;

[0136] Among them, Φ ρ Is a correlation coefficient r The cumulative distribution function of the two-dimensional standard normal distribution; Φ -1 is the inverse cumulative distribution function (i.e., quantile function) of the standard normal distribution; r is the relevant parameter of the Copula function, which indicates the degree of linear correlation between the change points; u and v are the values ​​of the cumulative distribution functions (CDFs) of the two marginal distributions.

[0137] For the Copula function, its probability density function is:

[0138] ;

[0139] in, z u =Φ -1 ( u ), z v =Φ -1 ( v ).

[0140] Furthermore, the parameters are estimated by maximum likelihood estimation. r , Copula's likelihood function L ( r ) is defined as the product of the joint density functions of all observation points, and taking the logarithm of the above formula can obtain the likelihood function L ( r ). Further using numerical optimization algorithms, the parameters can be obtained through iterative optimization. r Estimated value of .

[0141] It should be noted that the estimated value It can express the degree of linear correlation between the historical data change point and the field data change point. The larger the value, the stronger the correlation between the historical data change points and the field data change points. =1 indicates that there is a complete correlation between the historical data change points and the field data change points.

[0142] therefore As the credibility of its data type, that is, to what extent its change point location information can reflect the change point location of the main insulation degradation process. S The credibility of 2 is R 2, . R 1 and R 2 represent evidence right S 1 and S 2. The trust level of the two data sets. Since the probability needs to satisfy the trust level of each change point and be 1, R 1 and R 2 is normalized and we get Change Point The probability distribution number and change points The probability distribution number :

[0143] , .

[0144] In one embodiment, determining Change Point The probability distribution number and change points The probability distribution number ,include:

[0145] The location of the historical data change point follows the normal distribution, and the distribution interval is divided into k subintervals, calculate the location of the historical data change point within the interval ci The probability of ,in, i ={1,2,…, k};

[0146] The field data change point is the data set through the EWMA control chart The false alarm probability of the field data change point is determined by the central limit theorem. and the probability of missing reports of field data change points ; Among them, false positives include the EWMA control chart incorrectly judging that a change point has occurred when there is no change point, and false negatives include the EWMA control chart failing to detect a change point when a change point has occurred;

[0147] Based on probability and probability Determine the correct probability of the EWMA control chart judging the change point , based on probability and probability Sure Change Point The probability distribution number and change points The probability distribution number .

[0148] Specifically, the probability of the estimated value of a historical data change point is 0. However, in engineering practice, there is no truly continuous random variable with an infinite value space. For example, residual breakdown voltage degradation data, due to practical needs and measurement limitations, is equivalent to a finite discrete variable. Therefore, when calculating the probability of a historical data change point, the probability of the historical data change point falling within a certain time interval should be considered.

[0149] Since the historical data change point is theoretically an unbounded parameter, , according to the actual sample situation, the sample distribution interval is divided into k subintervals, denoted as c i =[ a i , b i ], i ={1,2,…, k}. For the historical data change point, it falls in the interval c i The probability of is:

[0150] .

[0151] Since the field data uses the EWMA control chart for change point detection, there are cases where the change point detection of the EWMA control chart is misjudged. Assuming that the probability of EWMA judging the change point correctly is P 2, then P 2 can represent the degree of confidence in the change point of the field data.

[0152] Whether the EWMA control chart correctly determines the location of the change point depends mainly on whether it has false positives or false negatives. The probability of a false positive is that when there is no change point, the EWMA control chart incorrectly determines that the system has a change point, that is, the EWMA value exceeds the control limit ( UCL or LCL ).set up Z t For the moment t The EWMA value of, according to the central limit theorem, Z t When there is no change point, it will approximately obey N ( m 1, s Z ),in s Z is the standard deviation of the EWMA. Then the probability of a false alarm in the EWMA control chart can be expressed as:

[0153] ;

[0154] in, Z t ~ N ( m 1, s Z 2 ).

[0155] The probability of EWMA control chart missing is the probability that the EWMA control chart fails to detect the change point when the change point occurs, that is, when the change point occurs, the EWMA value Z t Still falls in [ LCL , UCL ] interval. Assume that after the change point occurs, the mean of the data growth rate becomes m 2. EWMA value at this time Z t ~ N ( m 2, s Z ). Then the probability of missing values ​​in the EWMA control chart can be expressed as:

[0156] ;

[0157] in Z t ~N ( m 2, s Z 2 ).

[0158] In summary, the EWMA control chart determines the probability of the change point being correct. P 2 is:

[0159] .

[0160] right P 1 and P 2Do normalization processing and get Change Point The probability distribution number and change points The probability distribution number They are:

[0161] .

[0162] In one embodiment, Figure 6 Shown, based on evidence 、 as well as Change Point and The probability distribution numbers of each will change the point and change points One of the reference change points , including the following steps:

[0163] Step 601: Fusion of evidence and evidence , and obtain the first fusion evidence Change Point The probability distribution number and change points The probability distribution number .

[0164] Specifically, Indicates the fused evidence m ′ for the change point The confidence level, combined with the sample size evidence and sample data type evidence information. Indicates the fused evidence m ′ for the change point The confidence level, combined with the sample size evidence and sample data type evidence Through Dempster's synthesis rule, we can get the fused evidence m′, thereby more accurately evaluating the credibility of the change point position.

[0165] Step 602: Fusion of evidence and evidence , and obtain the second fusion evidence Change Point The probability distribution number and change points The probability distribution number .

[0166] Specific, second fusion evidence m The probability distribution number includes and . and Represent the second fusion evidence m Change Point and level of trust.

[0167] evidence m ′ is obtained by integrating evidence m 1 and m 2, the comprehensive sample size and sample data type have an impact on the change point estimation. m 3 is based on the evidence of the change point estimation model, reflecting the degree of trust in the change point estimation by different models. m ' and evidence m The probability distribution number of 3 is fused to obtain the fused evidence m The probability distribution number.

[0168] Step 603: If Greater than The change point As a benchmark change point ,like Less than The change point As a benchmark change point .

[0169] Specifically, the probability distribution number and probability distribution number Represent the fused evidence m Change Point and If Greater than , it means that after integrating all the evidence, the change point is more credible, so As a benchmark change point On the contrary, if Less than , indicating the change point is more credible, so As a benchmark change point .

[0170] In this embodiment, the method is based on the fusion of evidence m The probability distribution number, combined with the sample size evidence m 1. Evidence of sample data type m 2 and change point estimation model evidence m 3, the credibility of the change point position can be evaluated more comprehensively, thereby determining the benchmark change point position more accurately.

[0171] In one embodiment, Figure 7 As shown, the converging evidence and evidence , and obtain the first fusion evidence Change Point The probability distribution number and change points The probability distribution number , including the following steps:

[0172] Step 701: Evidence-Based and evidence Change Point and The probability distribution numbers of each, determine the evidence With evidence Conflict coefficient after fusion .

[0173] Specifically, the conflict coefficient The calculation formula is:

[0174] ;

[0175] Among them, the probability distribution number and probability distribution number Evidence m 1 pair of change points and level of trust. and Evidence m 2 pairs of change points t 1 and t 2 level of trust.

[0176] The formula takes into account the evidence and evidence At two different points and The sum of the products of the trust levels on . Conflict coefficient The larger the value, the more evidence and evidence The more significant the conflict is. =1, then the evidence and evidence There is a complete conflict and Dempster's synthesis rules cannot be used to merge them.

[0177] Step 702: Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the first fusion evidence Change Point The probability distribution number .

[0178] Specific, first fusion evidence The calculation formula is:

[0179] ;

[0180] in, It's evidence and evidence The conflict coefficient between the two pieces of evidence reflects the degree of conflict between them. Express evidence Change Point level of trust. Express evidence Change Point level of trust.

[0181] This formula takes into account the evidence and evidence The degree of conflict between the evidence and evidence The trust degree is integrated to obtain the first fusion probability after integration .

[0182] Step 703: Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the first fusion evidence Change Point The probability distribution number .

[0183] Specific, first fusion evidence Change Point The probability distribution number The calculation formula is:

[0184] .

[0185] in, k ' is evidence m 1 and m The conflict coefficient between 2; It's evidence m 1 pair of change points level of trust; It's evidence m 2 pairs of change points level of trust.

[0186] By considering the evidence m 1 and evidence m 2. The degree of conflict between the evidence m 1 and evidence m 2 to obtain the first fusion evidence. Change Point The probability distribution number .

[0187] In one embodiment, Figure 8 As shown, the converging evidence and evidence , and obtain the second fusion evidence Change Point The probability distribution number and change points The probability distribution number , including the following steps:

[0188] Step 801: Evidence-Based and evidence Change Point and The probability distribution number of each is used to determine the conflict coefficient .

[0189] Specifically, the conflict coefficient The calculation formula is:

[0190] ;

[0191] in, and First Fusion Evidence m ′ for the change point and level of trust. and Evidence m3 pairs of change points and level of trust.

[0192] This formula is used to measure the first fusion evidence m ' and evidence m 3. Conflict coefficient k The larger the value of , the more significant the conflict between the two pieces of evidence. k =1, it means that the two pieces of evidence are completely conflicting and cannot be fused using Dempster's synthesis rule.

[0193] Step 802: Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the second fusion evidence Change Point The probability distribution number .

[0194] Specific, second fusion evidence Change Point The probability distribution number The calculation formula is:

[0195] ;

[0196] in, k The first fusion evidence m ' and evidence m The conflict coefficient between 3 reflects the degree of conflict between the two pieces of evidence; The first fusion evidence m ′ for the change point level of trust; It's evidence m 3 pairs of change points level of trust.

[0197] This formula considers the conflict degree between the evidences and the first fusion evidence. m ' and evidence m 3 trust levels are integrated to obtain the second fusion evidence Change Point The probability distribution number .

[0198] Step 803: Based on the conflict coefficient ,evidence Change Point The probability distribution number and evidence Change Point The probability distribution number, calculate the second fusion evidence Change Point The probability distribution number .

[0199] Specific, second fusion evidence Change Point The probability distribution number The calculation formula is:

[0200] .

[0201] in, k The first fusion evidence m ' and evidence m The conflict coefficient between 3; The first fusion evidence m ′ for the change point level of trust; It's evidence m 3 pairs of change points level of trust.

[0202] This formula considers the conflict degree between the evidences and the first fusion evidence. m ' and evidence m 3 trust levels are integrated to obtain the second fusion evidence Change Point The probability distribution number .

[0203] In one embodiment, after determining the location of the benchmark change point, a Bayesian approach can be used to integrate field data with historical data in stages to accurately predict the remaining life of individual main insulation components. Bayesian statistical inference primarily involves two steps: solving the prior distribution of random parameters and updating the posterior distribution of random parameters.

[0204] Since the prior data is the residual breakdown voltage of the historical data, in order to determine the prior distribution of the random parameters under normal stress, the prior data is converted.

[0205] In historical data u ijk Indicates temperature T k Next j Sample (main insulation material) i The degradation data obtained during the measurement, t ijk Indicates the corresponding measurement time. u ij0 for u ijk Convert to T Degraded data under 0, tij0 for t ijk Convert to T The measurement time at 0. T 0 is normal working stress; i =1,2,…, I ; j =1,2,…, J ; k =1,2,…, K . I is the total number of measurements for each group of samples, J is the total number of test groups under each accelerated stress, K is the total accelerated stress.

[0206] According to the principle of constant acceleration factor, the accelerated degradation data is converted. On the premise of keeping the degradation measurement value unchanged, the measurement time under accelerated stress is converted to that under normal stress. The conversion relationship is:

[0207]

[0208] A main insulation performance degradation model based on a two-stage degradation process is established. The degradation processes of different motor main insulation materials often vary significantly. It is assumed that the drift parameter and diffusion parameter of the degradation process obey the following conjugate prior distribution:

[0209] .

[0210] Where Ga represents the Gamma distribution, N represents a normal distribution, a , b , c , d is a hyperparameter.

[0211] Using all ( u iz , t iz ) Establish the following likelihood function:

[0212] .

[0213] Among them, Δ u iz = u iz - u (i-1)z , Δ t iz =Δ t iz –Δ t (i-1)z .

[0214] Furthermore, it is possible to obtain The analytical expression of It can be obtained according to the following formula:

[0215] .

[0216] in, is the digamma distribution function.

[0217] Furthermore, The analytical expressions are:

[0218] ;

[0219] ;

[0220] .

[0221] Due to the parameters , is implicit data, It is impossible to solve directly, so we use the EM algorithm to solve it iteratively. Each round of the EM algorithm recursive iteration process consists of E step and M step.

[0222] After several recursive iterations, until all four hyperparameter values ​​converge to a given accuracy, the resulting hyperparameter value is the final prior estimate. The accuracy threshold is set to the relative error of the hyperparameter estimates of two adjacent iterations is no more than 10 -6 , expressed as:

[0223] .

[0224] In one embodiment, due to the need to use the data set A posteriori estimation of hyperparameters is performed, so the maximum partial discharge measured from field data can be converted into the same type of residual breakdown voltage as historical data through a support vector machine (SVR), including:

[0225] Construct a training data set with the maximum partial discharge as the input variable and the residual breakdown voltage as the output variable S : .

[0226] in, Q i is the maximum partial discharge, U i Residual breakdown voltage, n is the number of training samples.

[0227] Assumptions Q and U It is a linear relationship, such as Figure 9As shown, determining the regression function includes the following steps:

[0228] Step 901: Convert the original problem into a convex optimization problem.

[0229] SVR is essentially looking for the function f ( Q ) regresses the data, so the regression function (original problem) is:

[0230] ;

[0231] in, f The corresponding output variable U The regression function of w is the weight vector, and the input feature Q have the same dimensions; b is a parameter to be determined. The core is to obtain w and b The value of .

[0232] for f ( Q ), SVR requires the predicted value f ( Q i ) and the true value does not exceed e , so the slack variable is introduced and Measuring forecast deviation by more than e The original problem can be transformed into:

[0233] .

[0234] Step 902: Construct a Lagrangian function and optimize the dual of the original problem.

[0235] Furthermore, the Lagrange multiplier is introduced α i , , m i , transform the above formula into its dual optimization problem. First write the Lagrangian function L :

[0236] .

[0237] Among them, the Lagrange multiplier is: ; .

[0238] Step 903: Use KKT conditions to solve and obtain the regression function.

[0239] According to the KKT conditions, w , b, x i , find the partial derivative and set it to zero, and we can get the relationship between them and the Lagrange multiplier as follows:

[0240] ; ; ; .

[0241] Will w After eliminating the variables of the original problem and other variables, the original problem can be transformed into a dual optimization problem:

[0242] .

[0243] By solving the above formula, we can get α i and , we can get w and b .

[0244] In summary, the final regression function is:

[0245] .

[0246] when Q and U When it is a nonlinear relationship, the nonlinear problem can be transformed into a higher-dimensional linear problem by using a kernel function that satisfies the Mercer condition. Q i , Q ) is replaced by the kernel function K ( Q i , Q ), then the regression function is:

[0247] .

[0248] in, , β is the kernel function parameter.

[0249] Furthermore, the regression function can be obtained by transforming U =[ U 1( t 1), U 2( t 2),… U m ( t m )]. The posterior estimate of the hyperparameter can be derived from the Bayesian formula:

[0250] .

[0251] in, f ( m , oh |Δ U ) is the joint posterior density function, L (Δ U | m , oh ) is the likelihood function, f ( m , oh ) is the joint prior density function. Its likelihood function and joint prior density function expressions are:

[0252] ;

[0253] ;

[0254] Furthermore, we can obtain:

[0255] .

[0256] Since the conjugate prior distribution of a random parameter has the same form as its posterior distribution, the posterior estimate of the hyperparameter can be derived from the above formula:

[0257] ;

[0258] ;

[0259] ;

[0260] .

[0261] Furthermore, when a change point appears in the field data, the parameter values ​​of the two stages of the degradation process are updated, specifically including: using the EM algorithm to obtain the prior estimate of the hyperparameters of the first stage, and then using U =[ U 1( t 1), U 2( t 2),… ] Calculate the posterior estimate of the hyperparameters of the first stage; use the EM algorithm to obtain the prior estimate of the hyperparameters of the second stage, and then use U =[ , ,… U m ( t m)] Calculate the posterior estimate of the second stage hyperparameters. Further, the random parameters can be derived using the posterior estimate of the first stage hyperparameters and the posterior estimate of the second stage hyperparameters. , , , The posterior expected value of , , , .

[0262] Furthermore, the first stage (0 <t≤ ) Cumulative distribution function of the lifetime of the degradation model and the second stage (t> ) Cumulative distribution function of the lifetime of the degradation model for:

[0263] .

[0264] in, and are the mean and standard deviation of the historical data change points, It is also the probability density function of historical data, which is used to describe the change points (e.g. ) appears at time s The probability density of . , , , is a random parameter , , , The posterior expected value of the performance parameter is ΔU(t) = U(t+Δt) - U(t). The degradation increment ΔU(t) = U(t+Δt) - U(t) obeys the normal distribution. D is the failure threshold of the performance parameter. is the standard normal distribution function; exp is the natural exponential function.

[0265] Based on the same concept, the present application also provides a computer device for predicting the life of main insulating materials. The computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the above-mentioned main insulating material life prediction method based on conflicting multiple data sources.

[0266] The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 10As shown. The computer device includes a processor, memory, communication interface, display screen and input device connected via a system bus. The processor of the computer device is used 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 program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for predicting the life of a main insulating material based on conflicting multiple data sources is implemented.

[0267] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0268] The above embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting the life of main insulation materials based on conflicting multiple data sources, characterized in that: The method comprises: Get the main insulation material dataset and datasets , dataset The remaining breakdown voltage history dataset, dataset This is the maximum partial discharge field data set; Based on the dataset Determine the change point , based on the dataset Determine the change point , if the change point With the change point Different, perform steps a, b and c; the change point and Both represent the turning points of the degradation rate of the main insulation material performance; Step a: Consider the dataset and the dataset Different sample sizes, building sample size evidence and determine the For the change point The probability distribution number and the change point The probability distribution number , Representing a dataset The effect of sample size on the change point The level of trust Representing a dataset The effect of sample size on the change point level of trust; Step b: Consider the dataset and the dataset Different data types, build data type evidence and determine the For the change point The probability distribution number and the change point The probability distribution number , Representing a dataset Data type for the change point The level of trust Representing a dataset Data type for the change point level of trust; Step c: Consider the change point and the change point Using different change point estimation models, constructing change point estimation model evidence and determine the For the change point The probability distribution number and the change point The probability distribution number , Represents the dataset The strain point estimation model is used to estimate the change point The level of trust Represents the dataset The strain point estimation model is used to estimate the change point level of trust; Based on the evidence 、 as well as For the change point and The probability distribution number of each change point and the change point One of the reference change points , determine the reference change point is the cumulative distribution function of the main insulation material life of the boundary.

2. The main insulation material life prediction method based on conflicting multiple data sources according to claim 1 is characterized in that: Based on the evidence 、 as well as For the change point and The probability distribution number of each change point and the change point One of the reference change points ,include: Integrating the evidence and the evidence , and obtain the first fusion evidence For the change point The probability distribution number and the change point The probability distribution number ; Integrating the evidence and the evidence , and obtain the second fusion evidence For the change point The probability distribution number and change points The probability distribution number ; If the Greater than The change point As a benchmark change point , if the Less than The change point As a benchmark change point .

3. The main insulation material life prediction method based on conflicting multiple data sources according to claim 2 is characterized in that: Integrating the evidence and the evidence , and obtain the first fusion evidence For the change point The probability distribution number and the change point The probability distribution number ,include: Based on the evidence and the evidence For the change point and The probability distribution numbers of each, determine the evidence With the evidence Conflict coefficient after fusion ; Based on the conflict coefficient 、The evidence For the change point The probability distribution number and the evidence For the change point The probability distribution number, calculate the first fusion evidence For the change point The probability distribution number ; Based on the conflict coefficient 、The evidence For the change point The probability distribution number and the evidence For the change point The probability distribution number, calculate the first fusion evidence For the change point The probability distribution number .

4. The main insulation material life prediction method based on conflicting multiple data sources according to claim 2 is characterized in that: Integrating the evidence and the evidence , and obtain the second fusion evidence For the change point The probability distribution number and change points The probability distribution number ,include: Based on the evidence and the evidence For the change point and The probability distribution number of each is used to determine the conflict coefficient ; Based on the conflict coefficient 、The evidence For the change point The probability distribution number and the evidence For the change point The probability distribution number is used to calculate the second fusion evidence For the change point The probability distribution number ; Based on the conflict coefficient 、The evidence For the change point The probability distribution number and the evidence For the change point The probability distribution number is used to calculate the second fusion evidence For the change point The probability distribution number .

5. The main insulation material life prediction method based on conflicting multiple data sources according to claim 1 is characterized in that: Determine the For the change point The probability distribution number and the change point The probability distribution number ,include: Get the dataset The sample size N and the dataset The sample size N t , where the sample size N t The size of the sample capacity N changes with the monitoring time. The longer the monitoring time, the t The larger the sample size N t is a quantity that changes with monitoring time; Determine the sample size N and the sample size N t The logarithm of the product, wherein the ratio of the logarithm of the sample size N to the logarithm of the product is For the change point The probability distribution number , the sample size N t The ratio of the logarithm of the product is For the change point The probability distribution number .

6. The main insulation material life prediction method based on conflicting multiple data sources according to claim 1 is characterized in that: Determine the For the change point The probability distribution number and the change point The probability distribution number ,include: Based on the dataset Determine historical data change points based on the data set Determine the field data change point; wherein the change point is determined based on the historical data change point, the change point It is determined based on the change points of field data; The dependency relationship between historical data change points and field data change points is established through the objective function; ρ is the relevant parameter in the objective function, which is used to express the degree of linear correlation between the change points of historical data and the change points of field data; The relevant parameters are estimated by maximum likelihood estimation ρ , get the relevant parameters ρ Estimated value ; Based on the estimated Determine the For the change point The probability distribution number and the change point The probability distribution number .

7. The main insulation material life prediction method based on conflicting multiple data sources according to claim 6 is characterized in that: Determine the For the change point The probability distribution number and the change point The probability distribution number ,include: The location of the historical data change point follows the normal distribution, and the distribution interval is divided into k subintervals, calculate the position of the historical data change point within the interval c i The probability of ,in, i ={1,2,…, k }; The field data change point is obtained by EWMA control chart on the data set The false alarm probability of the field data change point is determined according to the central limit theorem. And the probability of missing the field data change point ; Wherein, a false positive includes the EWMA control chart mistakenly determining that a change point has occurred when there is no change point, and a false negative includes the EWMA control chart failing to detect a change point when a change point occurs; Based on the probability and the probability Determine the correct probability of the EWMA control chart judging the change point , based on the probability and the probability Determine the For the change point The probability distribution number and the change point The probability distribution number .

8. The main insulation material life prediction method based on conflicting multiple data sources according to claim 1 is characterized in that: Based on the dataset Determine the change point ,include: Setting up the dataset , Q n ( t n ) means in t n The maximum partial discharge monitored at all times; determining growth rates of a plurality of maximum partial discharge amounts based on the maximum partial discharge amounts monitored at each moment; Calculating an EWMA value at each moment based on the growth rate, and determining whether a change point occurs at the corresponding moment based on the EWMA value; Based on the determined multiple change point moments, the data set is determined The corresponding change point .

9. The main insulation material life prediction method based on conflicting multiple data sources according to claim 8, characterized in that: Calculating the EWMA value at each moment based on the growth rate, and determining whether a change point occurs at the corresponding moment based on the EWMA value, including: For the growth rate at any moment, the growth rate is substituted into the recursive formula to obtain the EWMA value; wherein the recursive formula is: ;in, λ Represents the smoothing factor, 0< λ ≤1, express The EWMA value at the moment, express The growth rate of time, express EWMA value at the moment; Calculating a target growth rate and a standard deviation of the EWMA value, and determining an upper control limit and a lower control limit based on the target growth rate and the standard deviation; When the When the value exceeds the upper control limit or falls below the lower control limit, a change point is determined to have occurred.

10. A computer device for predicting the life of a main insulating material, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method for predicting the life of a main insulating material based on conflicting multiple data sources according to any one of claims 1 to 9 are implemented.

Images