A reliability estimation method for secondary equipment in substations

Abstract

This invention discloses a method for estimating the reliability of substation secondary equipment, belonging to the field of product reliability analysis. Based on raw failure rate statistics and a failure rate prediction model for substation secondary equipment, the method obtains an expanded failure rate data sample within an operating time period. The method then uses the least squares method to estimate failure rate model parameters based on the expanded sample. Finally, the method converts this data into a time expression for the reliability of the substation secondary equipment, and determines the reliable operating life of the device based on a preset failure threshold. This method can analyze and predict the operating failure rate and failure life of substation secondary equipment with high precision, high reliability, high speed, and low power consumption.

Description

Technical Field

[0001] The present invention relates to the technical field of secondary equipment reliability analysis, in particular to a method for estimating the reliability of secondary equipment in a transformer substation. Background Art

[0002] Currently, power systems are expanding in scale, becoming increasingly complex, and developing towards ultra-high voltage (UHV). To meet these development needs and the public's demand for high-quality electricity, it is crucial to ensure that relay protection devices, the first line of defense for safe and stable substation operation, are accurately and quickly evaluated for their reliability.

[0003] Under actual operating conditions, environmental factors such as temperature, humidity, vibration, and dust can affect the reliability of secondary equipment. As the device ages, the failure rate of numerous electronic components gradually increases, reducing the device's remaining lifespan. For highly complex substation systems, accurate and effective reliability assessments greatly facilitate the maintenance and repair of secondary equipment, significantly contributing to the safe and stable operation of the substation. This also helps improve device utilization and reduce costs.

[0004] Mature reliability models are currently available for lifespan assessment, but for high-reliability relay protection systems, the scarcity of operational failure data presents significant challenges for model parameter estimation. In recent years, researchers have applied prediction methods from machine learning to parameter prediction for secondary equipment, effectively improving assessment effectiveness and accuracy. However, when processing average failure rate data for secondary equipment, which is time-related and represents time series data, BP neural networks and artificial neural networks have drawbacks, such as slow learning speeds and a tendency to fall into local minima, which significantly impact their prediction accuracy. Furthermore, in traditional neural network models, the hidden layers are disconnected from the input layer to the hidden layer and then to the output layer, whereas the average failure rate time series data are not independent of each other.

[0005] In a recurrent neural network, the current output of a sequence is correlated with the previous output, making it possible to predict average failure rate data in time series. However, the calculations of traditional recurrent neural networks are dependent on all the previous n calculations, which leads to exponential accumulation of computational effort and long training times. This is clearly unsuitable for relay protection devices with extremely long rated lifespans. The improved long short-term memory network (LSTM) based on recurrent neural networks solves the long-term memory problem and avoids the problem of all the previous n calculation results being included in the calculation. Summary of the Invention

[0006] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is to provide a high-precision and robust method for estimating the reliability of substation secondary equipment.

[0007] The technical solution adopted by the present invention to achieve the above-mentioned object is: a method for estimating the reliability of secondary equipment in a substation, comprising the following steps:

[0008] Obtaining original time series statistical data; the original time series statistical data is the operating indicators of the secondary equipment at multiple time points during actual operation, which are counted in chronological order. The operating indicator is the average failure rate of the secondary equipment under the operating limit; the average failure rate is the average number of failures of the substation secondary equipment per year;

[0009] Inputting the data of multiple consecutive operating years at time points in the original time series statistical data into the substation secondary equipment failure rate prediction model to obtain the average failure rate corresponding to the multiple consecutive operating years at time points;

[0010] Based on the actual statistical data of the average failure rate of the secondary equipment corresponding to multiple continuous operating years, an expanded average failure rate data sample of the substation secondary equipment that conforms to the failure distribution model law is obtained. The failure rate distribution function is parameterized by the average failure rate data sample to obtain the reliability model parameters of the substation secondary equipment. The life of the substation secondary equipment is obtained from the reliability model of the substation secondary equipment.

[0011] The construction method of the substation secondary equipment failure rate prediction model is as follows:

[0012] Obtaining original time series statistical data, where the original time series statistical data is an operating indicator of multiple time points during the actual operation of the secondary equipment, which is counted in chronological order. The operating indicator is an average failure rate of the secondary equipment under the operating limit, and the average failure rate is the average number of failures of the secondary equipment in the substation per year.

[0013] A long short-term memory network improved by a recurrent neural network is trained based on the original time series statistical data to obtain a failure rate prediction model for secondary equipment in the substation;

[0014] Among them, the input of the long short-term memory network improved based on the recurrent neural network is a plurality of continuous operating time points in the original time series statistical data, and the output of the long short-term memory network improved based on the recurrent neural network is the operating indicators corresponding to the plurality of continuous operating time points in the original time series statistical data.

[0015] The improved long short-term memory network based on the recurrent neural network includes:

[0016] Input layer, used to obtain the test index x at the current time point t ;

[0017] Hidden layer, used to test the indicator x according to the current time point t , short-term memory at the previous time point h t-1 and the long-term memory c at the previous time point t-1 Get the short-term memory h at the current time point t and the current time point long-term memory c t ;

[0018] The output layer is used to store the current time point short-term memory h t Output y as the current time point t Output, the current time point output y t That is the test index x at the next time point t+1 .

[0019] The hidden layer includes multiple hidden modules, each of which includes four fully connected layers that interact with each other. The four fully connected layers that interact with each other are:

[0020] g t Layer, used to test the indicator x at the current time point t and the short-term memory h at the previous time point t-1 Conduct analysis;

[0021] Forget Gate, by f t Control, the forget gate is used to control whether to discard the long-term memory c of the previous time point t-1 Part of the content;

[0022] Input gate, composed of i t Control, the input gate is used to filter the g t The analysis results of the layer are combined with the forgetting results of the forget gate to form the long-term memory c at the current time point. t ;

[0023] Output gate, by o t Control, the output gate is used to control whether to read and output the current time point long-term memory c t Part of the content.

[0024] The substation secondary equipment failure rate prediction model is implemented by the following formula:

[0025]

[0026]

[0027]

[0028] Among them, σ and tanh correspond to the sigmod and tanh nonlinear activation functions in the neural network respectively; W corresponds to the weight coefficient matrix, and the four fully connected layers are related to the test index x at the current time point. t The weight matrix is ​​{W xi ,W xf ,W xo ,W xg}, the four fully connected layers are about the short-term memory h of the previous time point t-1 The weight matrix is ​​{W hi ,W hf ,W ho ,W hg}, c t-1 is the long-term memory of the previous time point, h t-1 Short-term memory of the previous time point.

[0029] The method further includes performing error loss calculation on the output value and theoretical value of the substation secondary equipment failure rate prediction model, and adjusting the weight coefficient matrix W according to the error loss calculation result.

[0030] The original time series test data includes training data and test data, the training data is used to train the long short-term memory network improved based on the recurrent neural network, and the test data is used to test the accuracy of the substation secondary equipment failure rate prediction model;

[0031] The test data for testing the accuracy of the substation secondary equipment failure rate prediction model includes:

[0032] Inputting the test data into the substation secondary equipment failure rate prediction model;

[0033] The obtained output of the substation secondary equipment failure rate prediction model is compared with the original time series statistical test data to obtain the accuracy of the substation secondary equipment failure rate prediction model.

[0034] After obtaining the accuracy of the substation secondary equipment failure rate prediction model, the method further includes:

[0035] Comparing the obtained accuracy of the substation secondary equipment failure rate prediction model with a preset accuracy;

[0036] When the obtained accuracy of the substation secondary equipment failure rate prediction model is lower than the preset accuracy, the long short-term memory network improved based on the recurrent neural network is retrained.

[0037] The failure rate distribution function is expressed by the following formula:

[0038]

[0039] Where λ(t) is the failure rate distribution function, t is time, α is the scale parameter, and β is the shape parameter.

[0040] Before performing parameter estimation on the failure rate distribution function using the average failure rate data sample, the average failure rate data sample is pre-processed by taking the logarithm, and the parameter estimation of the failure rate distribution function is converted into a linear form:

[0041] ln[λ(t)]=m+nlnt (5)

[0042] Where t is time, λ(t) is the failure rate distribution function, and m and n are introduced simplified parameters.

[0043] The parameter estimation of the failure rate distribution function is achieved by the following formula:

[0044]

[0045] Among them, α is the scale parameter of the failure rate distribution function, β is the shape parameter of the failure rate distribution function, and m and n are the simplified parameters introduced above.

[0046] The reliability model of the substation secondary equipment is expressed by the following formula:

[0047]

[0048] Among them, the reliability model parameters α and β are the failure rate distribution function parameters, t is time, and R(t) is the reliability of the substation secondary equipment.

[0049] The calculation method for the life of the substation secondary equipment is:

[0050] Taking the reliability of the substation secondary equipment R(t0) = 0.9, the reliable service life prediction result of the secondary equipment is t0, which is used as the recommended decommissioning time of the substation secondary equipment.

[0051] The present invention has the following advantages and beneficial effects:

[0052] 1. This method uses a recurrent neural network (RNN)-modified long short-term memory (LSTM) network for data processing and prediction. It constructs a multidimensional input variable using the operating time of a set of secondary devices, predicts a multidimensional average failure rate matrix, and uses this expanded failure rate data to estimate the parameters of the failure rate model. Finally, it predicts the reliable service life of the secondary devices. Compared to traditional reliability assessment methods, this method effectively expands small data samples, achieving higher parameter estimation accuracy. The algorithm also exhibits excellent robustness, maintaining a relatively small prediction error with changes in the amount of training data.

[0053] 2. The method of the present invention can effectively estimate the life of secondary equipment, saving time and cost of operation and maintenance. At the same time, it has high prediction accuracy and good applicability, and can provide fast and reliable life estimation for secondary equipment products, shorten the R&D cycle, accelerate industry development iteration, and provide guarantees for the rapid and widespread application of secondary equipment.

[0054] 3. Due to the structural characteristics of the long short-term memory network itself, the reliability of the algorithm can also be guaranteed in the real-time life prediction of long-term secondary equipment aging tests, and the calculation and training time will not be greatly increased due to the accumulation of data, which will affect the system prediction timeliness and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Schematic diagram of the long short-term memory network architecture of the present invention.

[0056] Figure 2 Schematic diagram of a hidden layer cell module in the long short-term memory network of the present invention;

[0057] Figure 3 1 is a schematic diagram of failure rate model parameter estimation of a preferred embodiment of the present invention;

[0058] Figure 4 It is a reliability model and life prediction diagram of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0060] The present invention relates to a method for estimating the reliability of secondary equipment in a substation. The reliability function can be obtained from a failure rate distribution function with determined parameters, and the reliable service life of the secondary equipment can be predicted. The method specifically includes the following steps:

[0061] In step S1, taking a 110kV substation as an example, based on the operation and maintenance records of the protection devices, the average failure rate data of the substation secondary equipment (including relay protection devices, measurement and control devices, integrated intelligent devices, etc.) at each operating time point are sorted out as shown in Table 1.

[0062] Table 1 Statistics of secondary equipment failure rate

[0063]

[0064] Step S2: Standardize and preprocess the original time series test data; and divide the processed average failure rate data into training data and test data. The life distribution of secondary equipment is a continuous random variable. The reliability model adopts the two-parameter Weibull distribution. The reliability function can be expressed as follows:

[0065]

[0066] Where t is time, α is the scale parameter, and β is the shape parameter. Accordingly, the failure rate distribution function can be expressed as:

[0067]

[0068] make:

[0069] λ(t)=e m ·t n (8)

[0070] The present invention performs logarithmic processing on the average failure rate data and converts formula (8) into a linear form:

[0071] ln[λ(t)]=m+nlnt (5)

[0072] Both training data and test data contain input matrices and output matrices, where the input matrix X t Composed of lnt, output matrix Y t The preferred embodiment of the present invention inputs the matrix X t And the output matrix Y t The dimensions of are all chosen to be 4, that is, four runtime matrices and corresponding average failure rate matrices. The specific training data and test data are shown in Table 2.

[0073] Table 2 Long short-term memory network training set and test set

[0074]

[0075]

[0076] Step S3, designing a long short-term memory network framework based on the improved recurrent neural network. Figure 1 and Figure 2 shown.

[0077] like Figure 1 , the hidden layer modules of the long short-term memory network can be regarded as cells (C t ), its state is divided into two vectors: h t and c t .h t is the short-term state (with the output Y at this moment t t equal), c t represents long-term memory. t The core of the operation is: what information to read, what information to keep, and what information to discard in the long-term state of network learning. t Running horizontally above the hidden layer cells, there is less interaction and information is better preserved.t and the short-term memory h of the previous moment t-1 Input cell C t Then enter the four interactive fully connected layers, g t The layer performs basic analysis on the two inputs. In the standard RNN, the output is generated at this point and the module terminates. However, in the cells of the long short-term memory network, another three layers of "gate" structures are designed to select and process information. t control) determines which long-term memories should be forgotten, and the input gate (i t Control) determines g t Which contents should be added to the long-term memory, the output gate (o t Control) determines which long-term memories should be read and output. t 、i t 、o t The three items are three fully connected layers with sigmoid activation function, represented by function, and the input is the test index x at the current time point t and the short-term memory h at the previous time point t-1 , used to calculate the values ​​of the input gate, forget gate, and output gate.

[0078] like Figure 2 , when long-term memory c t-1 When passing through the neural network from left to right, it first passes through a forget gate, which is composed of f t Control to discard some memory, and then filter the memory with the input gate (i t and g t Control) to merge some current information and produce long-term memory results c t is directly output. In addition, long-term memory c t is copied and the tanh function is applied, and the current input X t and the short-term state h at the previous moment t-1 After filtering by the output gate, and tanh(c t ) together generate the current short-term state h t , which is the output Y at the current moment t . Figure 2 The output process can be expressed by the following formula:

[0079]

[0080]

[0081]

[0082] Among them, σ and tanh correspond to the sigmod and tanh nonlinear activation functions in the neural network respectively; W corresponds to the weight coefficient matrix. Specifically, {Wxi ,W xf ,W xo ,W xg} are four fully connected layers about the input vector X t The weight matrix, {W hi ,W hf ,W ho ,W hg} are four fully connected layers on the short-term memory h t-1 The weight matrix W is calculated by calculating the error loss between the output value (average failure rate at a certain time point) of the substation secondary equipment failure rate prediction model and the theoretical value, and adjusting the weight coefficient matrix W according to the error loss calculation result.

[0083] Step S4: Determine the number of input layers, hidden layers, and output layers, and train the LSTM network using the training set. In this preferred embodiment, the LSTM network has 7 input layers and 3 output layers. After repeated calculations, the number of hidden layers is set to 14.

[0084] In step S5, after training, the LSTM network is applied to the test set to verify the prediction accuracy. The predicted output of the test set is [0.0541, 0.0633, 0.1097, 0.1726], which has an average error of 4.56% compared to the actual values ​​in Table 2, indicating good accuracy.

[0085] In step S6, the long short-term memory network determined by the above steps is input with several groups of random operating time limit matrices arranged in ascending order within a limited time range to generate the prediction results of the average failure rate of secondary equipment at that operating time; the predicted average failure rate data of secondary equipment is combined with the actual statistical data in step S1 to obtain the expanded average failure rate data sample of secondary equipment that conforms to the failure distribution model law, as shown in Table 3.

[0086] Table 3 Expanded statistics of secondary equipment failure rate

[0087]

[0088] Step S7: Estimating the parameters of the failure rate distribution function based on the expanded average failure rate data; and obtaining the life prediction value of the secondary equipment based on the reliability experience threshold and the reliability function. Figure 3 The secondary equipment expansion failure rate data in Table 3 are processed by taking the logarithm and are represented by * in the figure. The least squares method is used to estimate the parameters m and n in formula (5), and the results are m = -8.3319 and n = 1.8119; then, according to formula (6), α = 27.9600 and β = 2.8119.

[0089]

[0090] Thus, the specific expression of the reliability function is obtained:

[0091]

[0092] like Figure 4 Based on expert experience, the reliability of relay protection devices is 0.9, and R(t0) = 0.9. The predicted reliable service life of the secondary equipment is t0 = 12.559 years. Based on this lifespan, the recommended decommissioning time for the relay protection device is 12.559 years. This is consistent with the operating time of the secondary equipment relay protection devices in actual substations. This method can effectively estimate the lifespan of secondary equipment, saving time and costs for operation and maintenance. It also has high prediction accuracy and good applicability, providing fast and reliable lifespan estimates for secondary equipment products, shortening R&D cycles and accelerating industry development iterations, ensuring the rapid and widespread application of secondary equipment. In particular, even with only early average failure rate data, this method can still expand later failure rate data to achieve effective predictions, pre-determining the lifespan of relay protection devices and achieving the expected results of predictive behavior.

[0093] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for estimating reliability of secondary equipment in a substation, characterized in that: The following steps are involved: Obtaining original time series statistical data; the original time series statistical data is the operating indicators of the secondary equipment at multiple time points during actual operation, which are counted in chronological order. The operating indicator is the average failure rate of the secondary equipment under the operating limit; the average failure rate is the average number of failures of the substation secondary equipment per year; Inputting the data of multiple consecutive operating years at time points in the original time series statistical data into the substation secondary equipment failure rate prediction model to obtain the average failure rate corresponding to the multiple consecutive operating years at time points; Based on actual statistical data of average failure rates of the secondary equipment corresponding to multiple continuous operating years, an expanded average failure rate data sample of the substation secondary equipment that conforms to the failure distribution model is obtained, and parameters of the failure rate distribution function are estimated based on the average failure rate data sample to obtain reliability model parameters of the substation secondary equipment. The life of the substation secondary equipment is obtained based on the reliability model of the substation secondary equipment; The construction method of the substation secondary equipment failure rate prediction model is as follows: Obtaining original time series statistical data, where the original time series statistical data is an operating indicator of multiple time points during the actual operation of the secondary equipment, which is counted in chronological order. The operating indicator is an average failure rate of the secondary equipment under the operating limit, and the average failure rate is the average number of failures of the secondary equipment in the substation per year. A long short-term memory network improved by a recurrent neural network is trained based on the original time series statistical data to obtain a failure rate prediction model for secondary equipment in the substation; The input of the long short-term memory network improved by the recurrent neural network is a plurality of continuous operating time points in the original time series statistical data, and the output of the long short-term memory network improved by the recurrent neural network is the operating indicators corresponding to the plurality of continuous operating time points in the original time series statistical data; The improved long short-term memory network based on the recurrent neural network includes: Input layer, used to obtain test indicators at the current time point x t ; Hidden layer, used to test the indicator according to the current time point x t , short-term memory at the previous time point h t-1 and long-term memory at the previous time point c t-1 Get the short-term memory at the current time point h t and long-term memory at the current time point c t ; The output layer is used to store the current time point in short-term memory h t Output as the current time point y t Output, the current time point output y t That is the test index at the next time point x t+1 .

2. A method for estimating reliability of secondary equipment in a substation according to claim 1, characterized in that: The hidden layer includes multiple hidden modules, each of which includes four fully connected layers that interact with each other. The four fully connected layers that interact with each other are: g t Layer, used to test the indicator at the current time point x t and the short-term memory at the previous time point h t-1 Conduct analysis; The Forget Gate, by f t Control, the forget gate is used to control whether to discard the long-term memory of the previous time point c t-1 Part of the content; Input gate, composed of i t control, the input gate is used to filter the g t The analysis results of the layer are combined with the forgetting results of the forget gate to form the long-term memory of the current time point c t ; Output gate, by o t Control, the output gate is used to control whether to read and output the long-term memory of the current time point c t Part of the content.

3. A method for estimating reliability of secondary equipment in a substation according to claim 2, characterized in that: The substation secondary equipment failure rate prediction model is implemented by the following formula: ； in, σ and tanh correspond to the sigmod and tanh nonlinear activation functions in neural networks respectively; W Corresponding to the weight coefficient matrix, the four fully connected layers test indicators at the current time point x t The weight matrix is ​​{ W xi , W xf , W xo , W xg }, the four fully connected layers are about the short-term memory of the previous time point h t-1 The weight matrix is ​​{ W hi , W hf , W ho , W hg }, is the long-term memory of the previous time point, Short-term memory of the previous time point.

4. A method for estimating reliability of secondary equipment in a substation according to claim 3, characterized in that: It also includes calculating the error loss between the output value and the theoretical value of the substation secondary equipment failure rate prediction model, and adjusting the weight coefficient matrix according to the error loss calculation result. W .

5. A method for estimating reliability of secondary equipment in a substation according to claim 1, characterized in that: The original time series test data includes training data and test data, the training data is used to train the long short-term memory network improved based on the recurrent neural network, and the test data is used to test the accuracy of the substation secondary equipment failure rate prediction model; The test data for testing the accuracy of the substation secondary equipment failure rate prediction model includes: Inputting the test data into the substation secondary equipment failure rate prediction model; The obtained output of the substation secondary equipment failure rate prediction model is compared with the original time series statistical test data to obtain the accuracy of the substation secondary equipment failure rate prediction model.

6. A method for estimating reliability of secondary equipment in a substation according to claim 5, characterized in that: After obtaining the accuracy of the substation secondary equipment failure rate prediction model, the method further includes: Comparing the obtained accuracy of the substation secondary equipment failure rate prediction model with a preset accuracy; When the obtained accuracy of the substation secondary equipment failure rate prediction model is lower than the preset accuracy, the long short-term memory network improved based on the recurrent neural network is retrained.

7. A method for estimating reliability of secondary equipment in a substation according to claim 1, characterized in that: The failure rate distribution function is expressed by the following formula: ； in, is the failure rate distribution function, t For time, α is the scale parameter, β is the shape parameter.

8. A method for estimating reliability of secondary equipment in a substation according to claim 7, characterized in that: Before performing parameter estimation on the failure rate distribution function using the average failure rate data sample, the average failure rate data sample is pre-processed by taking a logarithm, and the parameter estimation of the failure rate distribution function is converted into a linear form: ； in, t For time, is the failure rate distribution function, m 、 n is the simplified parameter introduced; The parameter estimation of the failure rate distribution function is achieved by the following formula: ； in, α is the scale parameter of the failure rate distribution function, β is the shape parameter of the failure rate distribution function, m 、 n is the simplified parameter introduced above.

9. A method for estimating reliability of secondary equipment in a substation according to claim 1, characterized in that: The reliability model of the substation secondary equipment is expressed by the following formula: ； Among them, the reliability model parameters α 、 β is the failure rate distribution function parameter, t For time, Reliability of substation secondary equipment.

10. A method for estimating reliability of secondary equipment in a substation according to claim 9, characterized in that: The calculation method for the life of the substation secondary equipment is: Reliability of secondary equipment in substations , the reliable service life prediction result of secondary equipment is t 0 , and serves as the recommended shutdown time for substation secondary equipment.

Images