Transformer time-varying reliability evaluation method combining degradation process and real-time state

Abstract

The present application relates to the technical field of power plant equipment state evaluation, and discloses a transformer time-varying reliability evaluation method combining degradation process and real-time state, comprising the following steps: step 1, collecting basic data sets; step 2, when there is missing data, combining a universal data restoration method, filling in the missing data; step 3, constructing a degradation equation of the historical degradation trajectory of the power transformer; based on the degradation equation, calculating a time-varying aging index HA i ; and then calculating the internal aging failure probability of the power transformer at time t i ; step 4, constructing and improving a logistic regression model; and determining the parameter value of the logistic regression model; and then establishing an external random failure probability evaluation model of the meteorological region where the power transformer is located; step 5, calculating the time-varying reliability of the power transformer combining the degradation process and the real-time state. The present application can effectively evaluate the time-varying reliability of the power transformer, has high precision and strong practicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power plant equipment state evaluation, and particularly relates to a transformer time-varying reliability evaluation method combining degradation process and real-time state. BACKGROUND

[0002] With the rapid development of economy and the continuous progress of society, electric power energy has become an indispensable basic material support in various industries. As the most important power generation method in the world at present, thermal power generation occupies an important position in power production, and it is particularly important to ensure the reliable and continuous delivery of thermal power plant electric energy. As the key hub equipment for realizing the delivery of thermal power plant electric energy, the operation reliability of power transformers not only directly affects the economic benefits and service quality of power plants, but also has an important influence on the power supply reliability of the entire power system. However, with the continuous extension of the service life, the power transformers of the thermal power plant in operation are directly exposed to the outside world for a long time, and after long-term bearing of internal electric, thermal and mechanical stress impact and external high temperature, rainfall and other complex weather coupling, they will gradually enter the aging stage. This will significantly increase the frequency of generator unit trip outage accidents caused by power transformer faults, seriously threaten the reliable and continuous delivery capability of electric energy, and even cause certain safety risks. Under this background, the time-varying reliability level evaluation of power transformers is particularly important.

[0003] However, the existing time-varying reliability level evaluation scheme for power transformers mostly performs poorly (low accuracy and lack of practicality), and the specific reasons are as follows: (1) Power transformer multi-source data processing: Due to the limitations of geographical environment, economic factors and other objective conditions, most power transformer monitoring devices currently use low-cost power distribution carrier and industrial wiring transmission methods. The above communication channels often have data missing in the collection, transmission and storage process when experiencing strong magnetic interference such as overvoltage and large current impact, limiting the quality of multi-source data and its subsequent application value. Existing methods usually focus on directly using raw data to develop reliability modeling and evaluation, without fully considering the objectively existing problem of data missing in the field, making it difficult to effectively improve the accuracy and practicality of modeling. (2) Power transformer internal failure probability modeling: Existing methods usually determine the current operating state of the power transformer based on its current monitoring data, and then evaluate the internal aging failure probability, without fully utilizing the past historical monitoring data of the power transformer. (3) Power transformer external random failure probability modeling: Current methods mainly give the random failure probability of power transformers under different weather conditions (such as light rain, moderate rain, etc.) based on weather state classification (such as three states). However, these methods usually only consider the influence of a single weather condition, ignoring the coupling effect of multiple weather factors such as precipitation, environmental temperature, air humidity and wind speed. (4) In addition, current methods have not provided a complete module architecture design, making it difficult to effectively realize the practical application of time-varying reliability remote monitoring of power transformers in thermal power plants. SUMMARY

[0004] The present application aims to provide a transformer time-varying reliability evaluation method combining degradation process and real-time state, which can effectively evaluate the time-varying reliability of power transformers with high accuracy and strong practicality.

[0005] The basic scheme provided by the present application is a transformer time-varying reliability evaluation method combining degradation process and real-time state, which includes the following steps:

[0006] Step 1: Collect historical oil chromatographic monitoring data, sensor data and weather monitoring data recorded by the meteorological platform of the power transformer as the basic data set;

[0007] Step 2: When there is missing data in the basic data set, combine the experience classification with the improved K-nearest neighbor algorithm to fill in the missing data in the oil chromatographic monitoring data or weather monitoring data;

[0008] Step 3: Based on the oil chromatographic monitoring data, combine the Wiener degradation process and the recursive filtering algorithm to construct the degradation equation of the historical degradation trajectory of the power transformer, i.e. i s i-1 (t i-1 )i -t i-1 )+σε i ; wherein, s i represents the degradation observation value of the power transformer at t i time evaluated according to the oil chromatographic monitoring data; μ is a drift coefficient; σ is a diffusion coefficient and σ>0, ε i obeys a normal distribution with mean value 0 and variance t i -t i-1 ;

[0009] Based on the degradation equation, the probability density distribution function of the residual life estimation value of the power transformer at t i time is calculated, and the time-varying aging index HA i of the power transformer at t i time is obtained therefrom; based on the time-varying aging index HA i , the internal aging failure probability of the power transformer at t i time is calculated;

[0010] Step 4, constructing a logistic regression model; based on the weather monitoring data and the historical failure account data of the power transformer, a principal component analysis method is used to improve the logistic regression model, the coupling effect of weather factors on the logistic regression model is analyzed, and the parameter value of the logistic regression model is determined; based on the parameter value of the logistic regression model, an external random failure probability evaluation model of the meteorological region where the power transformer is located is established; the external random failure probability evaluation model is used to calculate the external random failure probability of the power transformer;

[0011] Step 5, combining the internal aging failure probability output in step 3 and the external random failure probability output in step 4, the time-varying reliability of the power transformer combined with the degradation process and the real-time state is calculated.

[0012] The working principle and advantages of the present application are:

[0013] Firstly, the oil chromatographic information and weather data and other multi-source data of the power transformer are collected, and the data is cleaned through the general data restoration method combining experience classification and improved K nearest neighbor algorithm; secondly, based on the oil chromatographic monitoring data, a degradation process, a recursive filtering algorithm and the like are comprehensively used to construct an evaluation method of the internal aging failure probability of the power transformer; at the same time, based on various weather data, an improved logistic regression model is used to construct an evaluation method of the external random failure probability of the power transformer. Finally, on the basis of the present method, the online application of the present method can be realized through system module integration, and based on real-time oil chromatographic monitoring data and weather data, the time-varying reliability level of the power transformer can be evaluated online, and remote monitoring is supported.

[0014] The transformer time-varying reliability evaluation method combining the degradation process and the real-time state can effectively evaluate the time-varying reliability of the power transformer, has high precision and strong practicability.

[0015] Firstly, the evaluation is reasonable, and the evaluation result has high reference value. The power transformer data used in the scheme can be obtained from the existing database, which can ensure the rationality of the evaluation result from the root, and significantly improve the feasibility in actual engineering application. In addition, with the continuous increase of monitoring data, the related model can be continuously updated, thereby improving the accuracy of the time-varying reliability evaluation result of the power transformer.

[0016] Secondly, the evaluation precision is high. The scheme combines multi-source data, comprehensively considers the influence of internal performance degradation and external random weather factors of the power transformer, and performs targeted probability calculation. Combined with the two probabilities, the time-varying reliability of the power transformer combining the degradation process and the real-time state is calculated, which can reasonably and effectively evaluate the time-varying reliability level of the power transformer in the thermal power plant, has important significance for guaranteeing the reliable operation of the thermal power plant, and can provide more scientific decision basis for formulating the power transformer inspection cycle and maintenance strategy on site.

[0017] Particularly, the scheme fully considers the objective problem of data missing in the field, and designs a universal data restoration method combining experience classification and improved K nearest neighbor algorithm to fill in the missing data, which can accurately restore the data and help to improve the subsequent evaluation precision. Moreover, the scheme selects the overall monitoring data sequence of the power transformer for mining analysis, which can completely analyze the entire performance degradation process involved in the power transformer, and further can fully consider the inconsistency of the performance degradation process of individual power transformers, so that the evaluation result has higher accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 It is a method flowchart of the transformer time-varying reliability evaluation method combining the degradation process and the real-time state of the embodiment of the present application.

[0019] Figure 2 It is an input-output relationship diagram between the modules constructed in the method of the transformer time-varying reliability evaluation method combining the degradation process and the real-time state of the embodiment of the present application.

[0020] Figure 3 It is a missing data restoration value and error analysis diagram obtained by different methods in the example of the transformer time-varying reliability evaluation method combining the degradation process and the real-time state of the embodiment of the present application.

[0021] Figure 4The figure shows the comprehensive failure probability values of two power transformers under the same operation time according to the embodiment of the transformer time-varying reliability evaluation method combining the degradation process and real-time state. DETAILED DESCRIPTION

[0022] The following is further described in detail through specific embodiments:

[0023] The embodiment is basically as shown in the accompanying drawings: Figure 1 and Figure 2 The transformer time-varying reliability evaluation method combining the degradation process and real-time state comprises the following steps:

[0024] Step 1, collect multi-source heterogeneous data such as historical oil chromatographic monitoring data, sensor data and weather monitoring data recorded by a meteorological platform of the power transformer as a basic data set.

[0025] Specifically, the sensor data includes daily ambient temperature collected based on a temperature sensor, and air relative humidity data collected based on a humidity sensor. The oil chromatographic monitoring data includes historical oil chromatographic monitoring data of the power transformer collected by an oil dissolved gas analyzer derived from a device account. The weather monitoring data includes micro-meteorological data of the area where the thermal power plant is located derived from a meteorological platform or a meteorological station.

[0026] In this embodiment, two specific data storage forms are used to store the oil chromatographic monitoring data and the weather monitoring data, respectively. For example, for the power transformer n, it is specifically shown in G n and M n

[0027]

[0028] In the formula, G n is the oil chromatographic historical monitoring data matrix corresponding to the power transformer n; M n is the historical weather data matrix corresponding to the power transformer n. T is the total length of the historical sampling period, and the unit is day.

[0029] In turn, the daily average values of the seven kinds of oil dissolved gas contents of the power transformer obtained at the historical t-th time sequence collection point are represented, which correspond to methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), hydrogen (H2), carbon monoxide (CO) and carbon dioxide (CO2), respectively.

[0030] ​The six weather factors data suffered by the power transformer obtained at the historical t-th time sequence collection point are sequentially represented; the first five are numerical data, corresponding to the environmental temperature, air relative humidity, precipitation, wind speed, and sunshine duration, respectively; the last one is a text attribute label data, used to describe the overall weather condition of the day. In addition, if there is data missing, a special symbol is marked for distinction. For example, if the air relative humidity of the third collection point is missing, it is marked with * to facilitate subsequent missing data restoration.

[0031] In this step, a total historical database is also constructed in the data station of the thermal power plant; the total historical database is used to store the basic data set in a classified manner. Specifically, the construction of the total historical database can be completed by classifying and storing the two data matrices of the N power transformers respectively (N matrices).

[0032] Step 2. When there is missing data in the basic data set, a universal data restoration method combining empirical classification and improved K-nearest neighbor algorithm is used to fill in the missing data in the oil chromatogram monitoring data or weather monitoring data.

[0033] Here, the daily environmental temperature and air relative humidity data in the sensor data are incorporated into the weather monitoring data, and subsequent processing is based on the weather data matrix. In this way, the missing data in the basic data set can be properly handled.

[0034] In this step, the universal data restoration method includes:

[0035] Step2.1, based on the empirical classification standard, the oil chromatogram monitoring data or weather monitoring data is classified into different historical sample sets.

[0036] As shown in Table 1, the empirical classification standard corresponding to the oil chromatogram monitoring data includes classification according to the alarm information displayed by the oil chromatogram data on the same day, specifically including: category 1, normal; category 2, partial discharge; category 3, internal overheating; category 4, winding circulating current; category 5, electric flashover; and category 6, continuous spark.

[0037] Table 1: Example of empirical classification standard corresponding to oil chromatogram monitoring data

[0038]

[0039] As shown in Table 2, the empirical classification standard corresponding to the weather monitoring data includes: category 1, thunderstorm; category 2, strong wind; category 3, heavy rain; category 4, ice and snow; category 5, high temperature; and category 6, overcast.

[0040] Table 2: Example of empirical classification standard corresponding to weather monitoring data

[0041]

[0042]

[0043] Specifically, taking weather monitoring data as an example, each piece of weather data information is classified in combination with the overall weather condition of the day (i.e., the 6th column of feature data of the weather data matrix), and each type of historical sample set is denoted as S mi (i = 1, …, 6).

[0044] Step 2.2, when the oil chromatogram monitoring data or weather monitoring data of the a-th day is missing, the neighbor sample with the highest similarity to the data of the a-th day is selected from the historical sample set by improving the K-neighbor algorithm, and the missing data is restored according to the weighted value of the neighbor sample.

[0045] Specifically, taking weather monitoring data as an example, considering that under the same weather condition, environmental temperature, air relative humidity, precipitation, wind speed, and sunshine duration usually show certain correlation, which reflects the comprehensive influence of different weather conditions on the overall weather condition. Therefore, if the weather condition recorded by the historical weather data of the a-th day is r, the weather data missing in the a-th day is restored by using the complete weather record in the historical sample set S mr .

[0046] In this way, the deviation introduced by blind interpolation or time series prediction is avoided, and the potential correlation in the historical data is fully utilized, improving the rationality of the missing data restoration process.

[0047] The specific operation is: first, select num mr complete weather data samples with high similarity to the weather data of the a-th day from the historical sample set S k by improved K-neighbor algorithm, and then delete the last column (attribute label data) of these weather data, which is specifically represented as:

[0048]

[0049] In the formula: m n,a is the weather data sample of the a-th day, for example, if the relative humidity value of the second data is *, which means that the collected value is missing and needs to be restored. Top-num k Neighbors of m n,a represents selecting num n,a neighbor samples with the highest similarity to the weather data m k .

[0050] represents the data sample m n,a and m base,i(i = 1, …, num k The similarity between the weather data of the a-th day and the weather data of the i-th neighbor sample can be calculated by the following formula.

[0051]

[0052] In the formula, j represents the dimension corresponding to the missing data in the weather data of the a-th day. For example, if the second data, the relative humidity value, is missing, j takes the value of 2.

[0053] ξ h The weight of each dimension data can be calculated by the following formula:

[0054]

[0055] In the formula, num(Sm r ) is the total number of data samples without missing data in the historical sample set S mr .

[0056] p hb is the proportion of the h-th meteorological index in the j-th data sample, which can be calculated by the following formula:

[0057]

[0058] In the formula, is the normalized value of the h-th meteorological index in the j-th data sample.

[0059] According to the above steps, the num k neighbor samples with the highest similarity to the weather data sample of the a-th day are selected, and the missing j-th dimension data is restored according to the weighted value of the j-th dimension data of the num k neighbor samples. The specific restoration formula is as follows:

[0060]

[0061] In the formula, is the missing value of the missing j-th dimension data in the weather data sample of the a-th day.

[0062] According to the foregoing steps, the missing data in the historical weather data of any power transformer n is restored, thereby perfecting the historical weather database of the power transformer. The same method is also applicable to repairing the missing values in the historical oil chromatogram data.

[0063] In this step, it also includes: building a data cleaning module in the data station of the thermal power plant; the data cleaning module is used to execute step 2.

[0064] Step 3, based on the oil chromatogram monitoring data, combining the Wiener degradation process and the recursive filtering algorithm, a degradation equation of the historical degradation trajectory of the power transformer is constructed.

[0065] Specifically, a degradation equation based on Wiener process in physics (Wiener degradation process) is adopted to represent the performance aging evolution of the power transformer in its entire life cycle, and the specific performance aging process can be represented as:

[0066] S(t) = s i + μ(t - t i ) + σ(B(t) - B(t i )) = s i + μ(t - t i ) + σB(t - t i ), t > t i ;

[0067] In the formula, S(t) represents the performance degradation process function of the power transformer; s i represents the degradation observation estimate value of the power transformer at time t i ; μ is the drift coefficient; σ > 0 is the diffusion coefficient, B(t) represents the standard Brownian motion, which is a normally distributed random variable, and represents the randomness of the degradation observation.

[0068] The reason for using the Wiener process to describe the degradation process of the power transformer is that the small change behavior of the power transformer degradation in a small time interval is similar to the random fluctuation of small particles in fluid and air. By setting in this way, the degradation state of the power transformer can be effectively characterized and quantified.

[0069] Further, the degradation observation estimate value s i of the power transformer at time t i in the above formula can be calculated by using the oil chromatogram data value of the power transformer at time t i reduced in step 2:

[0070]

[0071] In the formula, β(t) is the oil dissolved gas ratio vector based on the three-ratio method in the IEC standard, which is C2H2 / C2H4 CH4 / H2 and C2H4 / C2H6 respectively, δ is the weight matrix, the sum of the matrix elements is 1, and in actual application, it can be determined by the experience knowledge of the field operation personnel.

[0072] Where g(·) is a normalization function used to correct the deviation caused by different dimensions of characteristic gas ratios (for example, C2H2 / C2H4 is usually less than 0.1, while CH4 / H2 is usually between 0.5 and 1). Specifically as follows:

[0073]

[0074] wherein, and are the lower and upper bounds of the feature gas ratio β i (t) respectively; the higher the value of U(β i (t)) is, the more obvious the internal aging degree of the power transformer is.

[0075] In the above steps, in order to consider the historical degradation trajectory of the power transformer in the power transformer degradation observation evaluation, a random walk strategy is also used to dynamically correct the drift coefficient μ corresponding to each time interval.

[0076] Specifically represented as: μ i = μ i-1 + η;

[0077] In the formula, η is a noise term, which is subject to a normal distribution with a mean of 0 and a variance of Q, and the change of the drift coefficient μ in a short monitoring interval can be simulated through the noise term, so that μ i is expected to be near μ i-1 adjusted by the noise term η. According to the condition definition of the random walk strategy, the initial drift coefficient μ0 is subject to a normal distribution with a mean of A0 and a variance of P0. Thus, the drift coefficient μ i is a random variable that changes over time, and its value is conditioned on μ i-1 .

[0078] Further, the degradation equation of the historical degradation trajectory of the power transformer is constructed: s i = s i-1 + μ i-1 (t i -t i-1 ) + σε i ;

[0079] Wherein, s i represents the degradation observation estimate of the power transformer at t i time evaluated according to the oil chromatographic monitoring data; μ is the drift coefficient; σ is the diffusion coefficient and σ>0, and ε i is subject to a normal distribution with a mean of 0 and a variance of t i -t i-1 .

[0080] The initial time when the power transformer is put into operation is t0=0; g0=0. ε i is subject to a normal distribution with a mean of 0 and a variance of t i -t i-1 , and t i -t i-1 is used as ε iThe variance of the Brownian motion B(t) is required by the properties of the Brownian motion.

[0081] Further, based on the degradation equation, the probability density distribution function of the residual life estimation value of the power transformer at time t i is calculated, and the time-varying aging index HAt i of the power transformer at time t i is obtained therefrom.

[0082] Specifically, the time sequence variation of the drift coefficient μ i reflects the state evolution process of the power transformer, and can be estimated by the observed power transformer oil chromatographic degradation data S i ={s0, s1, s2, …, s 0:i} from t0 to t i . In this step, the probability density function p(μ i | S i ) of μ 0:i is estimated by using a recursive method:

[0083]

[0084] In the formula, p(μ i | μ i-1 ) is the conditional probability of the drift coefficient μ i-1 at the current time under the condition that the drift coefficient μ i at the previous time is given; p(g i | μ i-1 , S 0:i-1 ) is the conditional probability of observing the current degradation observation g i-1 under the condition that the drift coefficient μ 0:i-1 at the previous time and all previous observation data S i are given; p(μ i-1 | S 0:i-1 ) is the probability distribution of the drift coefficient μ 0:i-1 under the condition that the observation data S i-1 is given; and p(g i | S 0:i-1 ) is a normalization constant. Through this recursive process, the estimation of the drift coefficient can be updated at each time step. Since the model depends on the estimation at the previous time and the current observation data, the model allows its prediction to be gradually improved as new data arrives.

[0085] In the process of recursively evaluating the drift coefficient μ i , a strong tracking filter is introduced to handle the occasional jumps or mutations that may exist in the drift coefficient and the degradation observation, and a multiple suboptimal fading factor is used to adaptively adjust the state prediction error:

[0086]

[0087] In the formula, Λ i Let F be the multiple fading factor matrix of the i-th recursive evaluation; k Let be the Jacobian matrix of the degraded observations of the power transformer.

[0088] Based on the aforementioned steps, the probability density distribution of the drift coefficient μi can be obtained, which can be specifically expressed as:

[0089]

[0090] The above formula represents the drift coefficient μ i Satisfy the mean The variance is P i∣i =var(μ i |G 0:i Gaussian distribution of μ; i With G 0:i The dependencies between them are contained in μ i The mean of the probability density distribution and variance P i∣i .

[0091] Based on the predefined aging threshold w for transformer aging failure (defined as 1 in this embodiment), the current t is calculated. i Remaining life of the power transformer at any given time: R i =inf{r i :S(n+t i )≥w∣S 0:i};

[0092] Then, based on the degradation equation, the power transformer at t is calculated. i probability density distribution function of the remaining lifetime estimate at time t

[0093]

[0094] In the formula, Φ(·) is the cumulative probability density function of the standard normal distribution; μ of the variable i Historical observation data and variance are updated recursively. The above formula includes four unknown parameters: the mean and variance a0 and P0 corresponding to the initial drift coefficient μ0; and the mean and variance Q and σ corresponding to the normal distribution of the parameter η. 2 The estimate can be dynamically updated using the expectation-maximization algorithm.

[0095] Furthermore, the time-varying aging index HA i Calculate using the following formula:

[0096]

[0097] In the formula, Lmean E(R) represents the average aging failure life of transformers of the same model; i ) is the power transformer at t i Remaining lifetime estimate R at time 1 i Expectations; HA i The larger the value, the shorter the remaining lifespan of the power transformer, indicating more severe aging.

[0098] Furthermore, based on the time-varying aging index HA i The power transformer at time t is calculated based on the two-parameter Weibull distribution model. i Internal aging failure probability at any given time:

[0099]

[0100] In the formula, P aging (t i ) is the power transformer at t i The probability of internal aging failure at any given time; α mean and κ mean These are the shape and scale parameters of the Weibull distribution, respectively. In practical applications, they can be obtained by fitting and evaluating statistical data on aging failures of similar power transformers.

[0101] This step also includes: setting up an internal modeling and evaluation module in the data station of the thermal power plant; the internal modeling and evaluation module is used to execute step 3.

[0102] Step 4: Construct a logistic regression model; Based on weather monitoring data and historical fault log data of power transformers, principal component analysis is used to improve the logistic regression model, analyze the coupling effect of weather factors on the logistic regression model, and determine the parameter values ​​of the logistic regression model by maximizing the likelihood function; Based on the parameter values ​​of the logistic regression model, an external random fault probability assessment model for the meteorological area where the power transformer is located is established; The external random fault probability assessment model is used to calculate the external random fault probability of the power transformer.

[0103] Specifically, it includes the following sub-steps:

[0104] Step 4.1: Based on weather monitoring data and historical fault log data of power transformers, construct the model fitting sample data required for this step. The data format and construction rules for each sample are as follows:

[0105]

[0106] Assignment rules:

[0107] In the formula, is the sample constructed according to the weather data and the operation records of the power transformer n in the t-th time period region; DY n,t is the output corresponding to the weather sample, determined by the historical record data of the fault, and is a binary variable.

[0108] Step 4.2, after obtaining the fitted sample data, the input data in the fitted sample data is standardized to eliminate the influence of the order of magnitude difference of each weather index:

[0109]

[0110] Step 4.3, the weight of each factor is determined by principal component analysis, and the variables with low weight or redundancy are screened out, and num cri key weather factors are retained, so as to improve the accuracy of the subsequent logistic regression model. Since the principal component analysis method is mature, this step will not be described in detail.

[0111] Step 4.4, the input data of num cri weather factors after standardization is used as the dependent variable, and DY is a binary output variable, and a sigmoid function is used to construct a logistic regression model:

[0112]

[0113] In the formula, l i (j=0,1,…,num cri ) is the estimated coefficient in the logistic regression model; V(l) is the output of the logistic regression model. It should be noted that the advantage of logistic regression is that in the model fitting stage, the output variable of the model is a binary variable, DY=1 represents that an external random fault has occurred, and DY=0 represents that the power transformer has not occurred. External random fault; in the application stage after fitting, the output probability of external random fault can take a continuous value between 0 and 1.

[0114] Step 4.5, based on the extracted N all groups of fitted sample data, the maximum likelihood estimation method is used to estimate the parameters of the logistic regression model. The specific value of num cri model parameters is determined by maximizing the likelihood function estimation function:

[0115]

[0116] In the formula, L(l) is the likelihood function,

[0117] Step 4.6, the specific value of the logistic regression model parameter is determined by maximizing the logarithmic value of the likelihood function, which is as follows:

[0118]

[0119] Step4.7, after the above steps, based on the parameter values of the logistic regression model, the external random failure probability evaluation model of the power transformer at t i is established:

[0120]

[0121] wherein, is the weather monitoring data of the power transformer at ti. It should be noted that with the increase of external weather data and failure account data of the power transformer, the logistic regression model parameters can be dynamically estimated to improve the accuracy of the evaluation.

[0122] In this step, it also includes: in the data station of the thermal power plant, an external modeling evaluation module is built; the external modeling evaluation module is used to execute step 4;

[0123] Step 5, combining the internal aging failure probability output by step 3 and the external random failure probability output by step 4, the time-varying reliability of the power transformer combined with the degradation process and the real-time state is calculated, and the calculation formula is as follows:

[0124]

[0125] wherein, P com (t i ) is the comprehensive time-varying failure probability of the power transformer, and the third negative term in the formula is used to consider that the power transformer generally does not occur external random failure and internal aging failure at the same time.

[0126] In this step, it also includes: in the data station of the thermal power plant, a result output module is built; the result output module is used to execute step 5. Combined with the modules constructed in steps 1-4, the system module architecture suitable for the time-varying reliability remote monitoring of the power transformer of the thermal power plant can be constructed, and the dynamic reliability evaluation of the power transformer based on online monitoring data can be realized.

[0127] In addition, in order to intuitively show the superiority of the present scheme, an example is combined to explain the application effect of the present scheme.

[0128] Two in-service transformers T1 and T2 in the power transformer database of a power enterprise are selected for example analysis. Based on the historical oil chromatographic monitoring data and regional weather data of the two transformers, the time-varying comprehensive failure probability at different state monitoring points is evaluated.

[0129] Firstly, taking the restoration of missing oil chromatographic data as an example, the effectiveness of the present scheme in data restoration is verified, and the traditional KNN method and ARIMA method are compared, and the specific analysis is shown in Table 3,Figure 3 and Figure 4 .

[0130] Table 3 Oil chromatogram sample construction with missing data

[0131]

[0132] Note: "*" indicates that the gas content monitoring data is missing and needs to be restored.

[0133] As can be seen from Figure 3 , when restoring the missing samples of the four oil chromatogram monitoring data, the relative errors of the present scheme are all controlled within 15%, while the relative errors of the traditional KNN method and the ARIMA method may exceed 30% in some cases, leading to data restoration failure. This is because the ARIMA method relies on time series prediction, but the monitoring data of transformer oil chromatogram may have sudden changes due to internal defects (such as short-time overheating, internal discharge); and the traditional KNN method finds similar oil chromatogram data samples in all unclassified data samples, which is also prone to large errors. The above results show that compared with the traditional KNN method and the ARIMA method, the missing data restoration method set in the present scheme is more suitable for handling the missing problem of transformer monitoring data and has higher accuracy.

[0134] Figure 4 Further shows that under the same service time, the comprehensive failure probability values calculated at different monitoring points (same service time) of the two power transformers. As can be seen from Figure 4 , the comprehensive failure probabilities of the two transformers gradually increase with the increase of service time. However, due to the difference in historical performance degradation process and weather conditions on the same day, there are differences in the comprehensive failure probability evaluation values of the two power transformers under the same service time. At the second and third monitoring points, the comprehensive failure probability of transformer T1 at each state monitoring point is significantly higher than that of T2 at the same period, mainly because the internal aging failure probability of transformer T1 is higher than that of T2. This phenomenon is specifically manifested in that the daily load rate of transformer T1 in field operation is significantly higher than that of T2, thereby causing its performance degradation speed to be also obviously faster than that of T2. At the first monitoring point, the comprehensive failure probability of T2 is higher than that of T1, because at this time, the internal aging failure probability of the two transformers is very small, at this time, the random failure probability caused by external weather dominates. The above results verify the evaluation effectiveness of the present scheme, prove that the present scheme can reasonably evaluate the comprehensive failure probability of the transformer, is a promising and applicable method to engineering practice, and has high popularization and application value.

[0135] The transformer time-varying reliability evaluation method combining degradation process and real-time state provided in the present embodiment can effectively evaluate the time-varying reliability of the power transformer, has high precision and strong practicability.

[0136] The above are only embodiments of the present application, and the common knowledge of the specific structure and characteristics in the scheme is not described in detail here. The ordinary skilled person in the art knows all the ordinary technical knowledge in the field of the application before the application date or the priority date, can know all the prior art in the field, and has the ability to apply conventional experimental means before that date. The ordinary skilled person in the art can perfect and implement the present scheme based on the disclosure given in the present application and in combination with their own ability. Some typical known structures or known methods should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be noted that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application. These will not affect the effect and practicality of the patent implementation.

Claims

1. A transformer time-varying reliability evaluation method combining degradation process and real-time state, characterized in that, The method comprises the following steps: Step 1, collecting oil chromatographic monitoring data, sensor data and weather monitoring data recorded by a weather platform of a power transformer history as a basic data set; Step 2, when there is missing data in the basic data set, combining the experience classification and the improved K nearest neighbor algorithm general data restoration method, filling the missing data in the oil chromatographic monitoring data or the weather monitoring data; Step 3, based on the oil chromatography monitoring data, combined with the Wiener degradation process and the recursive filtering algorithm, the degradation equation of the historical degradation trajectory of the power transformer is constructed, that is, s i = s i-1 + μ i-1 (t i -t i-1 )+σε i ; wherein s i represents the degradation observation estimate of the power transformer at time t i evaluated according to the oil chromatographic monitoring data; μ is the drift coefficient; σ is the diffusion coefficient and σ > 0, ε i obeys a normal distribution with mean 0 and variance t i -t i-1 . Based on the degradation equation, the probability density distribution function of the residual life estimation value of the power transformer at time t i is calculated, and the time-varying aging index HA i of the power transformer at time t i is obtained therefrom; based on the time-varying aging index HA i , the internal aging failure probability of the power transformer at time t i is calculated. Step 4, constructing a logistic regression model; based on the weather monitoring data and the historical fault account data of the power transformer, using the principal component analysis method to improve the logistic regression model, analyzing the coupling effect of weather factors on the logistic regression model, and determining the parameter value of the logistic regression model; based on the parameter value of the logistic regression model, establishing an external random failure probability evaluation model of the weather region where the power transformer is located; the external random failure probability evaluation model is used to calculate the external random failure probability of the power transformer; Step 5, combining the internal aging failure probability output in step 3 and the external random failure probability output in step 4, to calculate the time-varying reliability of the power transformer combined with the degradation process and the real-time state.

2. The transformer time-varying reliability evaluation method combining degradation process and real-time state according to claim 1, characterized in that, The sensor data includes daily ambient temperature collected based on a temperature sensor, and air relative humidity data collected based on a humidity sensor.

3. The method of claim 1, wherein the method is characterized by, In step 2, the general data restoration method comprises: first, based on the experience classification standard, classifying the oil chromatographic monitoring data or the weather monitoring data into different historical sample sets; when the oil chromatographic monitoring data or the weather monitoring data of the a-th day is missing, selecting the neighbor sample with the highest similarity to the data of the a-th day from the historical sample set by using the improved K nearest neighbor algorithm, and restoring the missing data according to the weighted value of the neighbor sample.

4. The transformer time-varying reliability evaluation method combining degradation process and real-time state according to claim 3, characterized in that, The experience classification standard corresponding to the oil chromatographic monitoring data comprises: category 1, normal; category 2, partial discharge; category 3, internal overheating; category 4, winding circulating current; category 5, electric flashover; and category 6, continuous spark.

5. The method for transformer time-varying reliability assessment combining degradation process and real-time state according to claim 3, characterized in that, The experience classification standard corresponding to the weather monitoring data comprises: category 1, thunderstorm; category 2, strong wind; category 3, heavy rain; category 4, ice and snow; category 5, high temperature; and category 6, overcast.

6. The method of claim 1, wherein the method is characterized by, In step 3, the random walk strategy is also used to dynamically correct the drift coefficient μ corresponding to each time interval.

7. The method of claim 1, wherein the method is characterized by, the time-varying aging index HA i The calculation is made with reference to the following formula: wherein L mean is the average aging failure life of the same type of transformer; E(R i ) is the expectation of the residual life estimate R i of the power transformer at time t i ; HA i is the larger, the more serious the aging condition of the power transformer, i.e., the shorter the residual life.

8. The method for transformer time-varying reliability assessment combining degradation process and real-time state according to claim 1, characterized in that, In step 3, the internal aging failure probability of the power transformer at time t is calculated based on the two-parameter Weibull distribution model: i P(t) = 1 - exp[-(t / η)β] where P aging (t i ) is the internal aging failure probability of the power transformer at time t i ; and α mean and κ mean are the shape parameter and scale parameter of the Weibull distribution, respectively.

9. The method for transformer time-varying reliability assessment combining degradation process and real-time state according to claim 1, characterized in that, In step 4, the parameter value of the logistic regression model is determined by maximizing the likelihood function.

10. The method of claim 1, wherein the method is characterized by, In step 1, it also includes: in the data station of the thermal power plant, constructing a general historical database; the general historical database is used for classified storage of the basic data set; In step 2, it also includes: in the data station of the thermal power plant, building a data cleaning module; the data cleaning module is used for executing step 2; In step 3, it also includes: in the data station of the thermal power plant, building an internal modeling evaluation module; the internal modeling evaluation module is used for executing step 3; In step 4, it also includes: in the data station of the thermal power plant, building an external modeling evaluation module; the external modeling evaluation module is used for executing step 4; In step 5, it also includes: in the data station of the thermal power plant, building a result output module; the result output module is used for executing step 5.

Images