Power distribution network equipment aging life prediction method and system and storage medium

Abstract

The application belongs to the field of power distribution network equipment management, and discloses a power distribution network equipment aging life prediction method and system and a storage medium. The method comprises the following steps: S1, monitoring the power distribution network equipment, periodically acquiring various types of monitoring data of the power distribution network equipment, and collecting the various types of monitoring data at a consistent period; S2, filtering each type of monitoring data of the power distribution network equipment in a preset time window to obtain various types of filtered monitoring data; S3, acquiring the data characteristics of each type of filtered monitoring data; and S4, inputting the data characteristics of all types of monitoring data into a pre-trained deep learning model for prediction to obtain the remaining life of the power distribution network equipment. The application can effectively improve the accuracy of the filtering result and thus improve the accuracy of the remaining life prediction result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power distribution network equipment management, and in particular to a power distribution network equipment aging life prediction method and system and a storage medium. BACKGROUND

[0002] Because different power distribution network equipment has different use environments, the aging speed of the power distribution network equipment is also different, and the actual use length of the power equipment may be less than or equal to the designed use life of the power equipment. Therefore, by predicting the residual life of the power distribution network equipment such as a transformer, the estimated residual use life of the power distribution network equipment can be obtained, so that the power operation and maintenance personnel can more timely replace the power equipment with a too small residual life and perform other operation and maintenance operations.

[0003] The process of predicting the residual life of the power equipment generally includes data preprocessing, data feature extraction, and inputting the extracted features into a trained prediction model for prediction. In the process of data preprocessing, the monitoring data of the power distribution network usually needs to be filtered. The existing filtering methods generally use only one filtering algorithm to obtain the filtering result of the monitoring data of the power distribution network or use two or more filtering algorithms to respectively calculate the filtering result of the monitoring data of the power distribution network, and then perform weighted fusion on the filtering results of all kinds of filtering algorithms to obtain the final filtering result. However, these filtering algorithms can only filter based on the monitoring data itself, so if the state of the power distribution network equipment suddenly changes, causing the monitoring data to change greatly, the monitoring data currently being filtered may be considered as noise and thus be incorrectly over-filtered, resulting in a large deviation between the result after filtering and the actual situation. This will result in inaccurate features obtained in the subsequent feature extraction process, affecting the accuracy of the prediction result obtained by predicting the residual life of the power distribution network equipment. SUMMARY

[0004] The present application aims to disclose a power distribution network equipment aging life prediction method and system and a storage medium to solve the technical problems proposed in the background.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a power distribution network equipment aging life prediction method, comprising:

[0007] S1, monitoring the power distribution network equipment, periodically obtaining multiple types of monitoring data of the power distribution network equipment, and the collection periods of the various types of monitoring data are consistent;

[0008] S2, filtering each type of monitoring data of the power distribution network equipment in a preset time window respectively to obtain multiple types of filtered monitoring data;

[0009] S3, obtaining data features of each type of filtered monitoring data respectively;

[0010] S4, inputting data features of all types of monitoring data into a deep learning model trained in advance for prediction to obtain the remaining life of the power distribution network equipment;

[0011] S2 includes:

[0012] S21, for monitoring data of type B, storing the monitoring data of type B obtained in the preset time window into a sequence U B ;

[0013] S22, using TU to represent a sequence of all types of monitoring data except type B, and obtaining a sequence of monitoring data of each type in TU obtained in the preset time window respectively;

[0014] S23, calculating a correlation value between monitoring data of U B and U i ; U i represents a sequence of monitoring data of type i obtained in the preset time window, i∈TU;

[0015] S24, using relv max to represent a maximum value of the obtained correlation value, judging whether relv max is greater than a set correlation value threshold, if yes, filtering the monitoring data in U B using a non-correlation filtering algorithm, if not, filtering the monitoring data in U B using a correlation filtering algorithm.

[0016] Preferably, the types of monitoring data include output voltage, output current, vibration frequency, temperature, operating noise, voltage of partial discharge and current of partial discharge.

[0017] Preferably, the preset time window is [t star -Tw, t star ], t star represents a time point at which the prediction of the remaining life of the power distribution network equipment starts, and Tw represents a length of the preset time window.

[0018] Preferably, the data features include time domain features and frequency domain features;

[0019] The time domain features include mean, variance, peak-to-peak value and zero-crossing rate.

[0020] The frequency domain features include a spectral center, a spectral peak, and a spectral width.

[0021] Preferably, the pre-trained deep learning model comprises an LSTM model or a GRU model.

[0022] Preferably, the correlation value between the monitoring data of U B and U i is calculated, comprising:

[0023] The correlation value is calculated using the following formula:

[0024]

[0025] relv B,i represents the correlation value between the monitoring data of U B and U i , N represents the total number of monitoring data of type B obtained within a preset time window; value B,n represents the nth monitoring data of type B obtained within the preset time window, value i,n represents the nth monitoring data of type i obtained within the preset time window.

[0026] Preferably, the monitoring data in U B is filtered using a non-correlation filtering algorithm, comprising:

[0027] The monitoring data in U B is filtered using an adaptive filtering algorithm.

[0028] The adaptive filtering algorithm comprises:

[0029] For the monitoring data d B in U B , the formula for filtering d B is:

[0030]

[0031] sd B represents the filtered monitoring data corresponding to d B , M represents the total number of monitoring data in the sequence seq, d j represents the jth monitoring data in seq, η1 represents a first control value; η2 represents a second control value.

[0032] The acquisition process of seq is as follows:

[0033] In U B , the acquisition time is before d B , and the acquisition time is before d BThe sequence formed by the first M monitoring data with the minimum absolute value of the difference between the acquisition time of the monitoring data and the acquisition time of d in the order from early to late is taken as seq, and M is a range control value.

[0034] Preferably, the data in U B is filtered by using a correlation filtering algorithm, including:

[0035] For the monitoring data d B in U B , the monitoring data with the acquisition time before d B and the minimum absolute value of the difference between the acquisition time of the monitoring data and the acquisition time of d B is denoted as zd B ; Bthr represents the upper limit value of the value range of the monitoring data of type B;

[0036] If zd B , thrs is a preset proportion, then d is filtered by using an adaptive filtering algorithm;

[0037] If zd C , then C represents the type of the monitoring data corresponding to the maximum correlation value; U C represents the sequence of the monitoring data of type C obtained within a preset time window;

[0038] The judgment value is calculated as:

[0039]

[0040] d C and zd C represent the monitoring data with the same acquisition time as d B and zd C in U B , and jdval is the judgment value;

[0041] If jdval is less than or equal to a preset judgment value threshold, then d B is filtered by using an adaptive filtering algorithm;

[0042] If jdval is greater than the preset judgment value threshold, then d B is filtered by using the following method:

[0043]

[0044] β represents the filtering weight, and mid B represents the average value of the monitoring data in seq.

[0045] In a second aspect, the application provides a power distribution network equipment aging life prediction system, including

[0046] The acquisition module, the filtering module, the feature extraction module and the prediction module;

[0047] The acquisition module is configured to monitor the power distribution network equipment, and periodically acquire various types of monitoring data of the power distribution network equipment, the acquisition periods of the various types of monitoring data being consistent;

[0048] The filtering module is configured to filter each type of monitoring data of the power distribution network equipment within a preset time window respectively, and obtain various types of filtered monitoring data;

[0049] The feature extraction module is configured to acquire data features of each type of filtered monitoring data respectively;

[0050] The prediction module is configured to input the data features of all types of monitoring data into a deep learning model trained in advance to perform prediction, and obtain the remaining life of the power distribution network equipment;

[0051] The filtering of each type of monitoring data of the power distribution network equipment within a preset time window respectively to obtain various types of filtered monitoring data includes:

[0052] For the monitoring data of type B, the monitoring data of type B obtained within the preset time window is stored in sequence U B ;

[0053] TU represents the sequence of all types of monitoring data except type B, and the sequence of monitoring data of each type within the preset time window in TU is acquired respectively;

[0054] The correlation value between the monitoring data of U B and U i is calculated; U i represents the sequence of monitoring data of type i obtained within the preset time window, i∈TU;

[0055] relv max represents the maximum value of the obtained correlation value, and it is judged whether relv max is greater than the set correlation value threshold, if yes, the monitoring data in U B is filtered by using the non-correlation filtering algorithm, and if not, the monitoring data in U B is filtered by using the correlation filtering algorithm.

[0056] In a third aspect, the present application provides a computer storage medium, which stores a computer program, and the computer program is executed by a processor to implement the power distribution network equipment aging life prediction method.

[0057] Advantages:

[0058] The present application improves the existing method for predicting the aging life of power distribution network equipment. When the correlation value meets the requirements, the monitoring data can be filtered based on the correlation value when there is a large change in the filtered monitoring data. In this way, the state of the power distribution network equipment can be effectively reduced when the monitoring data changes greatly, and the monitoring data being filtered is considered as noise, thereby reducing the probability of the occurrence of the event of false filtering, improving the accuracy of the filtering result, and improving the accuracy of the remaining life prediction result. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0060] Figure 1 A schematic diagram of the power distribution network equipment aging life prediction method of the present application.

[0061] Figure 2 A schematic diagram of the power distribution network equipment aging life prediction system of the present application. DETAILED DESCRIPTION

[0062] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0063] As shown in one embodiment, the present application provides a power distribution network equipment aging life prediction method, comprising: Figure 1

[0064] S1, monitoring the power distribution network equipment, periodically acquiring multiple types of monitoring data of the power distribution network equipment, and the collection period of various types of monitoring data is consistent;

[0065] S2, filtering each type of monitoring data of the power distribution network equipment in a preset time window respectively, and obtaining multiple types of filtered monitoring data; ​

[0066] S3, respectively acquiring data features of the filtered monitoring data of each type;

[0067] S4, inputting the data features of the monitoring data of all types into the deep learning model trained in advance for prediction to obtain the remaining life of the power distribution network equipment;

[0068] The S2 comprises:

[0069] S21, for the monitoring data of type B, storing the monitoring data of type B obtained within the preset time window into a sequence U B ;

[0070] S22, using TU to represent the sequence of all types of monitoring data except B, respectively acquiring the sequence of the monitoring data of each type obtained within the preset time window in TU;

[0071] S23, calculating the correlation value between the monitoring data of U B and U i ; U i represents the sequence of monitoring data of type i obtained within the preset time window, i∈TU;

[0072] S24, using relv max to represent the maximum value of the obtained correlation value, judging whether relv max is greater than the set correlation value threshold, if yes, using the non-correlation filtering algorithm to filter the monitoring data in U B , if not, using the correlation filtering algorithm to filter the monitoring data in U B .

[0073] The above scheme can filter the monitoring data based on the correlation value when the correlation value meets the requirements, when the filtered monitoring data changes greatly, so that the occurrence probability of the event that the monitoring data being filtered is considered as noise and is incorrectly over-filtered can be effectively reduced, the accuracy of the filtering result can be effectively improved, and the accuracy of the remaining life prediction result is improved.

[0074] Further, the set correlation value threshold can be 0.9. The collection period can be 1 hour.

[0075] Preferably, the types of monitoring data include output voltage, output current, vibration frequency, temperature, running noise, voltage of partial discharge, and current of partial discharge.

[0076] Further, the output voltage and the output current can be obtained by setting sensors on the power output lines of the power distribution network equipment. The vibration frequency can be obtained by setting vibration sensors on the surface of the power distribution network equipment. The temperature can be the temperature of the surface of the power distribution network equipment. The operating noise can be obtained by setting sound monitoring sensors near the power distribution network equipment. The partial discharge voltage and the partial discharge current can be obtained by setting corresponding sensors on the surface of the power distribution network equipment.

[0077] Preferably, the preset time window is [t star -Tw,t star ], Tw represents the length of the preset time window. star represents the time when the prediction of the remaining life of the power distribution network equipment starts, and Tw represents the length of the preset time window.

[0078] Further, the length of the preset time window can be one week.

[0079] Preferably, the data features include time domain features and frequency domain features.

[0080] The time domain features include mean, variance, peak-to-peak value and zero-crossing rate.

[0081] The frequency domain features include spectral center, spectral peak and spectral width.

[0082] The peak-to-peak value is the difference between the maximum value and the minimum value. The zero-crossing rate represents the ratio of the number of times the monitoring data crosses the zero axis to the total number of monitoring times. The spectral center is the weighted average frequency of the spectral energy, which can be used to describe the overall frequency range of the monitoring data. The spectral peak refers to the peak frequency component in the spectrum. The spectral width refers to the width of the peak in the spectrum.

[0083] Preferably, the pre-trained deep learning model includes an LSTM model or a GRU model.

[0084] Specifically, the training process of the deep learning model is as follows:

[0085] First, obtain the monitoring data that affects the life of the power distribution network equipment.

[0086] Second, pre-process the monitoring data.

[0087] Third, extract the data features of the pre-processed monitoring data.

[0088] Fourth, divide the pre-processed monitoring data into a training set, a validation set and a test set.

[0089] Fifth, use the training set to optimize the key parameters of the model (such as learning rate, hidden layer size, etc.). Grid search, random search or Bayesian optimization can be used.

[0090] Step 6, the model effect is verified by the verification set to avoid overfitting.

[0091] Step 7, the prediction effect of the model is evaluated by the verification set.

[0092] Preferably, the correlation value between the monitoring data of U B and U i is calculated, including:

[0093] The correlation value is calculated using the following formula:

[0094]

[0095] relv B,i represents the correlation value between the monitoring data of U B and U i , N represents the total number of monitoring data of type B obtained within a preset time window; value B,n represents the nth monitoring data of type B obtained within a preset time window, value i,n represents the nth monitoring data of type i obtained within a preset time window.

[0096] Specifically, in the present application, the upper limit value of the correlation value is 1, so when the correlation between the monitoring data of U B and U i is higher, the correlation value is larger, so another type of data can be introduced in the subsequent filtering process to select the filtering algorithm, and the accuracy of the filtering effect is improved.

[0097] Preferably, the monitoring data in U B is filtered by a non-correlation filtering algorithm, including:

[0098] The monitoring data in U B is filtered by an adaptive filtering algorithm;

[0099] The adaptive filtering algorithm includes:

[0100] For the monitoring data d B in U B , the formula for filtering d B is:

[0101]

[0102] sd B represents the corresponding filtered monitoring data of d B , M represents the total number of monitoring data in the sequence seq, d j represents the jth monitoring data in seq, η1 represents the first control value; η2 represents the second control value;

[0103] The process of obtaining seq is as follows:

[0104] Will U B In the middle, the acquisition time is in d B Previously, and the acquisition time was the same as d. B The sequence formed by sorting the M monitoring data with the smallest absolute value of the difference between the acquisition times in order from earliest to latest is called seq, where M is the range control value.

[0105] This invention employs an adaptive filtering algorithm when filtering monitoring data. This algorithm uses d B and d j The difference in numerical value and the difference in position within the sequence are used to determine d. j The corresponding weight, when d B and d j The greater the numerical difference, the more d j The later the position in the seq, the more d... j The larger the weight, the greater the impact on the filtering result. Therefore, while preserving the original numerical trend, the weight of d can be adjusted. B Effective filtering helps retain more useful information for subsequent data feature extraction.

[0106] Furthermore, the first control value is Bthr represents the upper limit of the value range for monitoring data of type B.

[0107] For example, for a given model of power distribution equipment, there is a corresponding range of output current, which generally includes an upper limit and a lower limit.

[0108] Furthermore, the second control value is

[0109] Furthermore, for U B Monitoring data d B If the acquisition time is in d B The amount of other monitoring data previously available was less than M, so mean filtering was used for d. B Perform filtering.

[0110] During filtering, the time d is obtained by calculation. B The average of other previous monitoring data is used as d B The corresponding filtered monitoring data.

[0111] Since the adaptive filtering algorithm of this invention requires reference to a large amount of prior monitoring data, when the amount of prior monitoring data is insufficient, mean filtering is used. Mean filtering does not require a specific setting of the number of data points involved in the filtering process. Subsequent data can be filtered using a small amount of prior data.

[0112] Preferably, the formula for calculating the range control value is:

[0113]

[0114] K represents the number of monitoring data in the sequence keq, where keq is U. B The acquisition time in d B Previously and with the acquisition time d B The sequence consisting of the K monitoring data points with the smallest absolute value of the difference in acquisition time, d k Let dm represent the k-th monitoring data in keq, and zd represent the maximum value of the monitoring data in keq. B Indicates the acquisition time in d B Previously and the time obtained is related to d B The monitoring data with the smallest absolute value of the difference in acquisition time, xd B This indicates the acquisition time is in zd B Previously and the time obtained is related to d B The monitoring data with the smallest absolute value of the difference between the acquisition time is denoted by leth, where leth represents the preset quantity and λ represents the range control weight.

[0115] When calculating the range control value, this invention calculates the fluctuation of the monitoring data in the sequence keq within a larger range and the drastic change of the monitoring data within a smaller range. This results in a larger range control value when the fluctuation and drastic change of the monitoring data in keq are greater, thus increasing the filtering strength and achieving more effective filtering. Conversely, a smaller range control value when the fluctuation and drastic change of the monitoring data in keq are smaller reduces the number of monitoring data points involved in filtering, thereby improving filtering efficiency. Therefore, this invention comprehensively considers both filtering efficiency and filtering effectiveness, maximizing filtering efficiency while ensuring the effectiveness of the filtering results.

[0116] Furthermore, the value of K is 5.

[0117] Furthermore, the preset quantity is 20.

[0118] Furthermore, the range control weight is 0.6.

[0119] Preferably, a correlation filtering algorithm is used for U B The data in the data is filtered, including:

[0120] For U B , the monitoring data d B with the acquisition time before d B and the absolute value of the difference between the acquisition time and the acquisition time of d B is expressed as zd B ; Bthr represents the upper limit value of the value range of the monitoring data of type B;

[0121] If , filter d B using an adaptive filtering algorithm, and thrs is a preset proportion;

[0122] If , C represents the type of monitoring data corresponding to the maximum correlation value; and U C represents the sequence of monitoring data of type C obtained within a preset time window;

[0123] Calculate the judgment value:

[0124]

[0125] d C and zd C represent the monitoring data in U C with the same acquisition time as d B and zd C , and jdval is the judgment value;

[0126] If jdval is less than or equal to a preset judgment value threshold, filter d B using an adaptive filtering algorithm;

[0127] If jdval is greater than the preset judgment value threshold, filter d B using the following method:

[0128]

[0129] β represents the filtering weight, and mid B represents the average value of the monitoring data in seq.

[0130] When the other type of monitoring data is qualified, the application does not directly filter based on the other type of monitoring data, because if the change range of the monitoring data is small, such filtering method has limited effect on the filtering result, and instead, it will affect the overall filtering efficiency due to the increase of calculation items. Therefore, the application adaptively determines the algorithm for filtering by calculating the judgment value only when the change range of the monitoring data is large, which can further ensure the overall filtering efficiency. When the judgment value is less than or equal to the judgment value threshold, the application still uses the adaptive filtering algorithm to filter, because the smaller the judgment value is, the closer the fluctuation ratio of the two types of monitoring data (i.e., the absolute value of the difference between the two adjacent monitoring data in the acquisition time and the upper limit value of the monitoring data) is, d B the probability of belonging to noise is greater. When the judgment value is large, the fluctuation ratio of the two types of monitoring data is greater, d B the probability of not belonging to noise is greater. At this time, the application improves the adaptive filtering by introducing the average value of the monitoring data. Since the average value will be affected by the monitoring data with large change range, the value of the filtered monitoring data can be made larger, so that the filtering formula with large difference in parameter type and formula structure can be effectively reduced without increasing the filtering formula. B the probability of being over-filtered, so that the filtering result can more truly reflect the change trend of the type B data, and the application difficulty of the application can be reduced because too many types of parameters are not added for filtering.

[0131] Further, the value of thrs is

[0132] Further, the judgment value threshold is

[0133] Further, the filtering weight is 0.5.

[0134] For example, Figure 2 The application provides a power distribution network equipment aging life prediction system, which comprises a collection module, a filtering module, a feature extraction module and a prediction module.

[0135] The collection module is used for monitoring the power distribution network equipment and periodically acquiring various types of monitoring data of the power distribution network equipment.

[0136] The filtering module is used for filtering each type of monitoring data of the power distribution network equipment in a preset time window to obtain various types of filtered monitoring data.

[0137] The feature extraction module is used for acquiring the data features of each type of filtered monitoring data.

[0138] The prediction module is used for inputting data features of all types of monitoring data into a deep learning model trained in advance to make a prediction, so as to obtain the remaining life of the power distribution network equipment;

[0139] The monitoring data of each type of the power distribution network equipment in the preset time window is filtered respectively, and a plurality of types of filtered monitoring data are obtained, including:

[0140] For the monitoring data of type B, the monitoring data of type B obtained in the preset time window is stored in sequence U B ;

[0141] TU represents the sequence of all types of monitoring data except type B, and the sequence of the monitoring data of each type obtained in the preset time window is obtained in TU respectively;

[0142] The correlation value between the monitoring data of U B and U i is calculated; U i represents the sequence of the monitoring data of type i obtained in the preset time window, i∈TU;

[0143] relv max represents the maximum value of the obtained correlation value, and whether relv max is greater than the set correlation value threshold is judged, if yes, the monitoring data in U B is filtered by using the non-correlation filtering algorithm, and if not, the monitoring data in U B is filtered by using the correlation filtering algorithm.

[0144] The application further provides a computer storage medium, which stores a computer program, and the computer program is executed by a processor to realize the power distribution network equipment aging life prediction method.

[0145] The above is only a specific embodiment of the application, but the protection scope of the application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A method for predicting the aging life of power distribution network equipment, characterized in that, include: S1 monitors the power distribution network equipment and periodically acquires various types of monitoring data from the power distribution network equipment, with the acquisition period for each type of monitoring data being consistent. S2, filter each type of monitoring data of the power distribution network equipment within the preset time window to obtain multiple types of filtered monitoring data; S3, acquire the data characteristics of each type of filtered monitoring data; S4. Input the data features of all types of monitoring data into a pre-trained deep learning model for prediction to obtain the remaining lifespan of the power distribution network equipment. S2 includes: S21, For monitoring data of type B, the monitoring data of type B obtained within the preset time window will be stored in the sequence. ; S22, TU represents the sequence of all types of monitoring data except B, and the sequence of monitoring data obtained for each type in TU within a preset time window is obtained respectively; S23, Calculation and The correlation values ​​between the monitoring data; This represents a sequence of monitoring data of type i obtained within a preset time window, where i∈TU; S24, using This represents the maximum value of the relevant values ​​obtained, and the judgment is made accordingly. If the value exceeds the set correlation threshold, then an uncorrelated filtering algorithm is used. If the monitoring data is not filtered, then a correlation filtering algorithm is used. The monitoring data is filtered.

2. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, The types of monitoring data include output voltage, output current, vibration frequency, temperature, operating noise, partial discharge voltage, and partial discharge current.

3. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, Within the preset time window , This indicates the point at which the prediction of the remaining lifespan of the distribution network equipment begins. Indicates the preset time window length.

4. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, Data characteristics include time-domain characteristics and frequency-domain characteristics; Time-domain characteristics include mean, variance, peak-to-peak value, and zero-crossing rate; Frequency domain characteristics include the spectral center, spectral peaks, and spectral width.

5. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, Pre-trained deep learning models include LSTM or GRU models.

6. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, calculate and The correlation values ​​between the monitoring data include: Calculate the relevant values ​​using the following formula: express and The correlation value between the monitoring data, where N represents the total number of monitoring data of type B obtained within the preset time window; This represents the nth monitoring data of type B obtained within a preset time window. This represents the nth monitoring data of type i obtained within a preset time window.

7. The method for predicting the aging life of power distribution network equipment according to claim 1, characterized in that, Using an uncorrelated filtering algorithm The monitoring data in the data is filtered, including: Adaptive filtering algorithm is used for Filter the monitoring data in the process; Adaptive filtering algorithms include: for Monitoring data ,right The formula for filtering is: express The corresponding filtered monitoring data, where M represents the total number of monitoring data in the sequence seq. This represents the j-th monitoring data in seq. Indicates the first control value; Indicates the second control value; The process of obtaining seq is as follows: Will In the middle, the acquisition time is Before the acquisition time, and the acquisition time is the same as The sequence formed by sorting the M monitoring data with the smallest absolute value of the difference between the acquisition times in order of acquisition time from earliest to latest is called seq.

8. The method for predicting the aging life of power distribution network equipment according to claim 7, characterized in that, Using correlation filtering algorithm to The data in the data is filtered, including: for Monitoring data The time will be acquired The acquisition time before and the acquisition time with The monitoring data with the smallest absolute value of the difference in acquisition time is represented as ;use This represents the upper limit of the value range for monitoring data of type B; like Then an adaptive filtering algorithm is used to... Perform filtering, where thrs is a preset ratio; like Let C represent the type of monitoring data corresponding to the maximum value of the correlation value; This represents a sequence of monitoring data of type C obtained within a preset time window; Calculate the judgment value: and express Obtaining time and and The same monitoring data, For judgment value; like If the value is less than or equal to a preset threshold, an adaptive filtering algorithm is used. Perform filtering; like If the value exceeds the preset threshold, the following method will be used to determine the value. Perform filtering: Indicates the filter weights. This represents the average value of the monitoring data in the seq.

9. A power distribution network equipment aging life prediction system, characterized in that, It includes an acquisition module, a filtering module, a feature extraction module, and a prediction module; The data acquisition module is used to monitor the power distribution network equipment and periodically acquire various types of monitoring data from the equipment, with the acquisition period for each type of monitoring data being consistent. The filtering module is used to filter each type of monitoring data of the power distribution network equipment within a preset time window to obtain multiple types of filtered monitoring data. The feature extraction module is used to obtain the data features of each type of filtered monitoring data; The prediction module is used to input the data features of all types of monitoring data into a pre-trained deep learning model for prediction, in order to obtain the remaining lifespan of the power distribution network equipment. Specifically, monitoring data for each type of distribution network equipment within a preset time window is filtered to obtain multiple types of filtered monitoring data, including: For monitoring data of type B, the monitoring data of type B obtained within the preset time window will be stored in the sequence. ; Let TU represent the sequence of all types of monitoring data except B, and obtain the sequence of monitoring data obtained for each type in TU within a preset time window; calculate and The correlation values ​​between the monitoring data; This represents a sequence of monitoring data of type i obtained within a preset time window, where i∈TU; use This represents the maximum value of the relevant values ​​obtained, and the judgment is made accordingly. If the value exceeds the set correlation threshold, then an uncorrelated filtering algorithm is used. If the monitoring data is not filtered, then a correlation filtering algorithm is used. The monitoring data is filtered.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for predicting the aging life of power distribution equipment according to any one of claims 1 to 8.

Images