Overhead ground wire stress corrosion fracture prediction method and device based on machine learning, terminal equipment and storage medium

Abstract

The invention discloses an overhead ground wire stress corrosion fracture prediction method and device based on machine learning, terminal equipment and a storage medium, and belongs to the technical field of overhead ground wire stress corrosion fracture prediction.The method comprises the steps that historical working condition time sequence data of an overhead ground wire sample is obtained, and a corresponding historical SCC value is determined; then mutual information, importance scores and principal component analysis are calculated, and a second feature and a third feature are determined; obtaining working condition time sequence data, a second characteristic value and a third characteristic value of the overhead ground wire to be predicted; calculating an SCC prediction value of each corrosion prediction model according to the working condition time sequence data, the second characteristic value and the third characteristic value; and finally, the prediction probability of corrosion fracture of the overhead ground wire to be predicted is calculated in combination with a preset probability mapping function. By implementing the stress corrosion fracture prediction method and device, the problems that in the prior art, due to the fact that stress corrosion fracture prediction is conducted through manual inspection and experience judgment, the accuracy of the prediction result is low, and efficiency is low can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of overhead ground wire stress corrosion cracking prediction, and particularly relates to an overhead ground wire stress corrosion cracking prediction method and device based on machine learning, a terminal equipment and a storage medium. BACKGROUND

[0002] As an indispensable part of the power transmission line, the overhead ground wire bears multiple key functions such as lightning protection, electromagnetic interference shielding and communication optical cable bearing. Due to long-term erection in the open field environment, the surface of the metal conductor is continuously eroded by electrochemical corrosion medium. At the same time, under the action of long-term mechanical stress such as self-weight, wind vibration and ice load, the metal lattice structure of the overhead ground wire is prone to produce micro cracks. When the electrochemical corrosion and mechanical stress are coupled, the stress corrosion cracking (SCC) phenomenon is easily induced - this failure form is characterized by the formation of macro cracks along the grain boundary or through the grain under the joint action of tensile stress and corrosion medium, and eventually leads to sudden rupture of the conductor.

[0003] At present, the detection of overhead ground wire stress corrosion cracking mainly adopts traditional detection methods, that is, professional technicians regularly inspect the line, and identify potential stress corrosion cracks through experience judgment and simple detection equipment. Therefore, the existing technology has the problem that stress corrosion cracking prediction is performed through manual inspection and experience judgment, resulting in low accuracy of the prediction result and low efficiency. SUMMARY

[0004] The present application provides an overhead ground wire stress corrosion cracking prediction method and device based on machine learning, a terminal equipment and a storage medium, which can solve the problem in the prior art that stress corrosion cracking prediction is performed through manual inspection and experience judgment, resulting in low accuracy of the prediction result and low efficiency.

[0005] An embodiment of the present application provides an overhead ground wire stress corrosion cracking prediction method based on machine learning, comprising:

[0006] Obtain historical working condition time series data of an overhead ground wire sample, determine a historical SCC value corresponding to the overhead ground wire sample according to the historical working condition time series data, determine a plurality of material service performance characteristics corresponding to the overhead ground wire sample according to the historical working condition time series data, calculate mutual information of each material service performance characteristic and the corresponding historical SCC value, and take the first N material service performance characteristics with the maximum mutual information as a plurality of first characteristics;

[0007] Take the first M characteristics with the maximum importance score in the first characteristics as a plurality of second characteristics, perform principal component analysis on the plurality of second characteristics, retain principal components with a cumulative contribution rate greater than a preset contribution rate threshold, and obtain a plurality of third characteristics; wherein N and M are positive integers.

[0008] obtain the working condition time series data of the overhead ground wire to be predicted in the current period, and determine a plurality of second characteristic values and third characteristic values corresponding to the time series data;

[0009] Based on a plurality of corrosion prediction models constructed by different algorithms, the SCC prediction values corresponding to each corrosion prediction model are calculated according to the working condition time series data, the plurality of second characteristic values and the third characteristic values;

[0010] According to the preset probability mapping function and the SCC prediction values corresponding to different corrosion prediction models, the prediction probability of corrosion fracture of the overhead ground wire to be predicted is calculated.

[0011] Further, the material service performance characteristics include temperature-humidity interaction characteristics, stress-environment interaction characteristics, stress corrosion sensitivity index, stress intensity factor, corrosion rate, historical statistical characteristics and historical time-frequency characteristics

[0012] The plurality of material service performance characteristics corresponding to the overhead ground wire sample are determined according to the historical working condition time series data, including:

[0013] The historical working condition time series data is standardized and preprocessed to obtain preprocessed historical working condition time series data;

[0014] The temperature-humidity interaction characteristics are calculated according to the historical environmental temperature and the historical relative humidity in the preprocessed historical working condition time series data;

[0015] The stress-environment interaction characteristics are calculated according to the historical stress, the historical chloride ion concentration and the historical PH value in the preprocessed historical working condition time series data;

[0016] The stress corrosion sensitivity index is calculated according to the historical environmental temperature, the historical stress, the historical chloride ion concentration and the historical PH value in the preprocessed historical working condition time series data;

[0017] The stress intensity factor is calculated according to the historical stress in the preprocessed historical working condition time series data;

[0018] The corrosion rate is calculated according to the historical relative humidity, the historical environmental temperature, the historical chloride ion concentration and the historical PH value in the preprocessed historical working condition time series data;

[0019] The historical mean, the historical standard deviation, the skewness, the kurtosis and the coefficient of variation corresponding to each preprocessed historical working condition time series data are calculated according to the preprocessed historical working condition time series data, and the historical mean, the historical standard deviation, the skewness, the kurtosis and the coefficient of variation are taken as the historical statistical characteristics;

[0020] According to the pretreated historical working condition time series data, the trend feature, the periodic feature and the autocorrelation feature corresponding to each pretreated historical working condition time series data are calculated, and the trend feature, the periodic feature and the autocorrelation feature are taken as the historical time-frequency features of the historical working condition time series data.

[0021] Further, the historical working condition time series data is standardized and pretreated to obtain the pretreated historical working condition time series data.

[0022] According to the historical working condition time series data, the average value corresponding to each historical working condition time series data is calculated.

[0023] According to the average value and the corresponding historical working condition time series data, the standard deviation corresponding to each historical working condition time series data is calculated.

[0024] According to the average value and the standard deviation, the historical working condition time series data is subjected to outlier correction to obtain first corrected historical sample data.

[0025] The first corrected historical sample data is denoised to obtain second corrected historical sample data.

[0026] For each historical working condition time series data, the trend term and the seasonal term corresponding to the missing value are calculated according to the historical working condition time series data.

[0027] According to the trend term and the seasonal term, the corresponding missing value is subjected to interpolation processing to obtain second corrected historical sample data.

[0028] For each historical working condition time series data, the median and the absolute median deviation of the historical working condition time series data are determined according to the historical working condition time series data.

[0029] According to the median and the absolute median deviation, the corresponding historical working condition time series data is standardized to obtain standardized historical working condition time series data, thereby obtaining pretreated historical working condition time series data.

[0030] Further, the historical working condition time series data is standardized and pretreated to obtain the pretreated historical working condition time series data.

[0031] For each historical working condition time series data, the first quartile, the third quartile and the interquartile range are extracted from the historical working condition time series data.

[0032] According to the first quartile and the interquartile range, the first threshold value is calculated.

[0033] According to the third quartile and the interquartile range, the second threshold value is calculated.

[0034] When the above-mentioned historical operating condition time series data is less than the above-mentioned first threshold value, the above-mentioned historical operating condition time series data is greater than the above-mentioned second threshold value; or the absolute value of the difference between the above-mentioned historical operating condition time series data and the corresponding average value is greater than the preset multiple of the standard deviation, the above-mentioned historical operating condition time series data is regarded as an abnormal value, and the above-mentioned abnormal value is corrected to obtain the first corrected historical sample data.

[0035] Furthermore, the above-mentioned corrosion prediction models constructed by different algorithms are used to calculate the SCC prediction value corresponding to each corrosion prediction model according to the working condition time series data, the second eigenvalues ​​and the third eigenvalues, including:

[0036] Inputting the operating condition time series data into a first preset corrosion prediction model, so that the first preset corrosion prediction model predicts a first SCC prediction value based on the operating condition time series data;

[0037] Inputting the plurality of second characteristic values ​​into a second preset corrosion prediction model and a third preset corrosion prediction model, respectively, so that the second preset corrosion prediction model and the third preset corrosion prediction model predict a second SCC prediction value and a third SCC prediction value based on the plurality of second characteristic values;

[0038] inputting the third eigenvalues ​​into a fourth preset corrosion prediction model, so that the fourth preset corrosion prediction model predicts a fourth SCC prediction value based on the third eigenvalues;

[0039] Among them, the above-mentioned first preset corrosion prediction model is constructed based on a deep neural network, the above-mentioned second preset corrosion prediction model is constructed based on a support vector machine model, the above-mentioned third preset corrosion prediction model is constructed based on a random forest model, and the above-mentioned fourth preset corrosion prediction model is constructed based on a gradient boosting decision tree model.

[0040] Furthermore, the training of the first preset corrosion prediction model includes:

[0041] Inputting the preprocessed historical operating condition time series data and the corresponding historical SCC values ​​into the first preset corrosion prediction model to be trained for iterative training until the loss function converges, thereby generating the first preset corrosion prediction model;

[0042] Among them, during each iterative training, the current first historical SCC prediction value is predicted based on the current preprocessed historical operating condition time series data; the current loss function is calculated based on the current first historical SCC prediction value and the corresponding historical SCC value, and it is judged whether the current loss function converges; if the current loss function converges, the current first preset corrosion prediction model is used as the trained first preset corrosion prediction model; otherwise, the parameters in the current first preset corrosion prediction model are adjusted and training continues.

[0043] Furthermore, the above-mentioned prediction probability of the overhead ground wire to be predicted to be corroded and broken is calculated based on the preset probability mapping function and the SCC prediction values ​​corresponding to different corrosion prediction models, including:

[0044] Obtain the model weight corresponding to each corrosion prediction model;

[0045] Calculate the final SCC value based on the model weight, the first SCC prediction value, the second SCC prediction value, the third SCC prediction value, and the fourth SCC prediction value;

[0046] The final SCC value is probability mapped according to a preset probability mapping function to obtain the predicted probability that the overhead ground wire will corrode and break.

[0047] Based on the above method embodiment, the present invention provides a corresponding device embodiment;

[0048] The present invention provides a device for predicting stress corrosion cracking of overhead ground wires based on machine learning, comprising:

[0049] Feature extraction module, principal component analysis module, data acquisition module, SCC prediction module and probability mapping module;

[0050] The feature extraction module is configured to obtain historical operating condition time series data of overhead ground wire samples, determine the historical SCC values ​​corresponding to the overhead ground wire samples based on the historical operating condition time series data, determine a number of material service performance characteristics corresponding to the overhead ground wire samples based on the historical operating condition time series data, calculate the mutual information between each material service performance characteristic and the corresponding historical SCC value, and select the top N material service performance characteristics with the largest mutual information as the first features.

[0051] The principal component analysis module is configured to select the first M features with the largest importance scores among the first features as a plurality of second features; perform principal component analysis on the plurality of second features, and retain principal components whose cumulative contribution rates are greater than a preset contribution rate threshold, to obtain a plurality of third features; wherein N and M are positive integers;

[0052] The data acquisition module is configured to acquire working condition time series data of the overhead ground wire to be predicted in a current period, and determine a plurality of second characteristic values and third characteristic values corresponding to the time series data.

[0053] The SCC prediction module is configured to calculate SCC prediction values corresponding to a plurality of corrosion prediction models constructed by different algorithms based on the working condition time series data, the plurality of second characteristic values and the third characteristic values.

[0054] The probability mapping module is configured to calculate a prediction probability of corrosion fracture of the overhead ground wire to be predicted based on a preset probability mapping function and the SCC prediction values corresponding to different corrosion prediction models.

[0055] On the basis of the method embodiment, the application correspondingly provides a terminal device embodiment.

[0056] The application provides a terminal device, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the method for predicting stress corrosion cracking of an overhead ground wire based on machine learning according to any one of the embodiments of the application when executing the computer program.

[0057] On the basis of the method embodiment, the application correspondingly provides a storage medium embodiment.

[0058] The application provides a storage medium, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the method for predicting stress corrosion cracking of an overhead ground wire based on machine learning according to any one of the embodiments of the application when executing the computer program.

[0059] The embodiments of the application have the following beneficial effects:

[0060] The application provides a machine learning-based overhead ground wire stress corrosion cracking prediction method and device, terminal equipment and a storage medium. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0062] Figure 1 is a flowchart of a machine learning-based overhead ground wire stress corrosion cracking prediction method provided by an embodiment of the present application.

[0063] Figure 2 is an environmental condition risk assessment schematic diagram provided by an embodiment of the present application.

[0064] Figure 3 is a mechanical condition assessment schematic diagram provided by an embodiment of the present application.

[0065] Figure 4 is a risk factor contribution degree analysis schematic diagram provided by an embodiment of the present application.

[0066] Figure 5 is an evaluation result schematic diagram provided by an embodiment of the present application.

[0067] Figure 6 is a structural schematic diagram of a machine learning-based overhead ground wire stress corrosion cracking prediction device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0068] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of the present application.

[0069] Reference Figure 1 To solve the problem in the prior art that stress corrosion cracking prediction is performed by manual inspection and experience judgment, resulting in low prediction result accuracy and low efficiency. An embodiment of the present application provides a machine learning-based overhead ground wire stress corrosion cracking prediction method, comprising:

[0070] Step S101: Obtain historical working condition time series data of an overhead ground wire sample, determine a historical SCC (stress corrosion cracking) value corresponding to the overhead ground wire sample according to the historical working condition time series data, determine a plurality of material service performance characteristics corresponding to the overhead ground wire sample according to the historical working condition time series data, calculate mutual information of each material service performance characteristic and the corresponding historical SCC value, and take the first N material service performance characteristics with the maximum mutual information as a plurality of first characteristics.

[0071] Specifically, the historical working condition time series data includes environmental parameters, mechanical parameters, material parameters, and historical SCC values of the overhead ground wire sample. The environmental parameters include: environmental temperature (unit: ℃, range: -40-60 ℃), relative humidity (unit: %, range: 0-100 %), rainfall (unit: mm / h), wind speed (unit: m / s), atmospheric pressure (unit: kPa), ultraviolet intensity (unit: W / m 2 ), atmospheric pH value (range: 3-9), chloride ion concentration [Cl-] (unit: mg / L), sulfide concentration [S 2 -] (unit: mg / L), and sulfur dioxide concentration [SO2] (unit: mg / m 3 ). These parameters directly affect the speed of electrochemical corrosion process and the corrosion mechanism, and are important environmental driving factors of stress corrosion cracking.

[0072] The mechanical parameters include: conductor tension (unit: N), stress (unit: MPa), strain (dimensionless), and vibration acceleration (unit: m / s 2), the wind vibration frequency (unit: Hz), the conductor sag (unit: m), and the torsion angle (unit: °). The mechanical parameters determine the stress state inside the material, which is the key driving factor of stress corrosion cracking, and together with the environmental parameters determine the crack initiation and propagation behavior.

[0073] The material parameters include: the conductor material type, the conductor diameter (unit: mm), the cross-sectional area (unit: mm 2 ), the surface roughness (unit: μm), the running years (unit: years), the surface coating thickness (unit: μm), and the surface damage degree (grade: 1-5, grade 1: smooth surface, no obvious damage; grade 2: slight oxidation or wear; grade 3: moderate corrosion or scratch; grade 4: obvious corrosion pits or cracks; grade 5: severe corrosion, broken strands or obvious defects). The material parameters reflect the intrinsic characteristics and aging state of the conductor, and together with the environmental and mechanical parameters determine the stress corrosion sensitivity.

[0074] Specifically, after discretizing the material service performance characteristics and historical working condition time series data, the mutual information is calculated by the following formula:

[0075]

[0076] In the formula, I(X;Y) represents the mutual information value between the feature X and the target variable Y, Y is the SCC index, which is a continuous value ranging from 0 to 1, representing the risk degree of stress corrosion cracking, when Y=0, there is no SCC risk, when Y=1, there is a very high SCC risk, p(x,y) represents the joint probability distribution of the feature X and the target variable Y, p(x) represents the marginal probability distribution of the feature X, p(y) represents the marginal probability distribution of the target variable Y, p(x i ) represents the marginal probability of the i-th value interval of the feature X, N represents the total number of historical working condition time series data, N(X=x i ) represents the number of times the feature X takes the value x i , p(x i ,y j ) represents the joint probability distribution of the feature X=x i and the target variable Y=y j , N(X=x i ,Y=y j ) represents the number of times that X=x i and Y=y j are simultaneously satisfied.The number of historical operating condition time series data. For example, assume that the ambient temperature parameter is discretized to obtain: x1: [0℃, 10℃), x2: [10℃, 20℃), x3: [20℃, 30℃) and x4: [30℃, 40℃); the SCC index is discretized to obtain: y1: low risk [0, 0.3), y1: medium risk [0.3, 0.7) and y1: high risk [0.7, 1.0], then This means that the ambient temperature is within [10°C, 20°C) and the SCC index is high, which corresponds to the joint probability distribution.

[0077] Preferably, the parameters included in the historical working condition time series data involve environmental parameters, mechanical parameters and material parameters. A complete multi-dimensional data acquisition system is established, and multiple influencing factors such as environmental parameters, mechanical parameters, material parameters, etc. are incorporated into a unified analysis framework, overcoming the limitation of traditional methods that only consider a single factor.

[0078] In a preferred embodiment, the service performance characteristics of the above materials include: temperature-humidity interaction characteristics, stress-environment interaction characteristics, stress corrosion sensitivity index, stress intensity factor, corrosion rate, historical statistical characteristics and historical time-frequency characteristics

[0079] The above-mentioned material service performance characteristics corresponding to the above-mentioned overhead ground wire samples are determined based on the historical working condition time series data, including:

[0080] Performing standardized preprocessing on the above historical operating condition time series data to obtain preprocessed historical operating condition time series data;

[0081] The temperature-humidity interaction characteristics are calculated based on the historical ambient temperature and historical relative humidity in the preprocessed historical operating condition time series data;

[0082] Specifically, the temperature-humidity interaction characteristics are calculated using the following formula:

[0083]

[0084] Where, F TH represents the temperature-humidity interaction feature, T represents the historical ambient temperature, H represents the historical relative humidity, and T opt represents the optimum corrosion temperature (usually 30°C), σ T Represents the temperature sensitivity parameter (usually 10).

[0085] The stress-environment interaction characteristics are calculated based on the historical stress, historical chloride ion concentration and historical pH value in the pre-processed historical operating condition time series data;

[0086] Specifically, the stress-environment interaction characteristics are calculated using the following formula:

[0087]

[0088] In the formula, F SE represents the stress-environment interaction characteristic, σ represents the historical stress, T ref represents the reference temperature (298K), and R represents the gas constant (8.314 J / (mol·K)).

[0089] According to the historical environmental temperature, the historical stress, the historical chloride ion concentration, and the historical PH value in the pretreated historical working condition time series data, a stress corrosion sensitivity index is calculated;

[0090] Specifically, the stress corrosion sensitivity index is calculated by the following formula:

[0091]

[0092] In the formula, SCC index represents the stress corrosion sensitivity index, σ yield represents the material yield strength, [Cl - ]threshold represents the chloride ion critical concentration, E a represents the activation energy, and β pH and γ represent empirical parameters.

[0093] According to the historical stress in the pretreated historical working condition time series data, a stress intensity factor is calculated;

[0094] Specifically, the stress intensity factor is calculated by the following formula:

[0095]

[0096] In the formula, K SCC represents the stress intensity factor, Y' represents the geometric correction factor, a represents the initial crack length, f geometry represents the geometric shape function of the initial crack.

[0097] According to the historical relative humidity, the historical environmental temperature, the historical chloride ion concentration, and the historical PH value in the pretreated historical working condition time series data, a corrosion rate is calculated;

[0098] Specifically, the corrosion rate is calculated by the following formula:

[0099]

[0100] In the formula, v corr represents the corrosion rate, k0 represents the frequency factor, α', β', and γ' represent the reaction order, and RH is the historical relative humidity.

[0101] According to the pretreated historical working condition time series data, the historical mean, historical standard deviation, skewness, kurtosis and coefficient of variation corresponding to each pretreated historical working condition time series data are calculated, and the historical mean, historical standard deviation, skewness, kurtosis and coefficient of variation are taken as the historical statistical characteristics.

[0102] Specifically, the historical mean for reflecting the central tendency of data is calculated by the following formula:

[0103]

[0104] In the formula, μ represents the historical mean, and n represents the total number of pretreated historical working condition time series data.

[0105] Specifically, the historical standard deviation for reflecting the dispersion degree of data is calculated by the following formula:

[0106]

[0107] In the formula, σ represents the historical standard deviation. z

[0108] Specifically, the skewness for reflecting the symmetry of data distribution is calculated by the following formula:

[0109]

[0110] In the formula, Skew represents the skewness.

[0111] Specifically, the kurtosis for reflecting the sharpness of data distribution is calculated by the following formula:

[0112]

[0113] In the formula, Kurt represents the kurtosis.

[0114] Specifically, the coefficient of variation for reflecting the relative variability is calculated by the following formula:

[0115]

[0116] In the formula, CV represents the variation.

[0117] According to the pretreated historical working condition time series data, the trend feature, periodic feature and autocorrelation feature corresponding to each pretreated historical working condition time series data are calculated, and the trend feature, periodic feature and autocorrelation feature are taken as the historical time-frequency characteristics of the historical working condition time series data.

[0118] Specifically, the trend feature is represented by the linear regression slope: ​​

[0119]

[0120] wherein β" represents a trend characteristic, t i′ represents a time point corresponding to the i'th normalized historical operating condition time series data, represents a time mean value.

[0121] Specifically, the main frequency of the data is analyzed by a fast Fourier transform (FFT) to obtain the above periodic characteristic:

[0122]

[0123] wherein X(k) represents the periodic characteristic at the k'th frequency, N represents the sequence length of the normalized historical operating condition time series data, and x(n) represents the value of then'th sampling point in the time domain.

[0124] Specifically, the autocorrelation characteristic is calculated by the following formula:

[0125]

[0126] wherein R(τ) represents the autocorrelation characteristic, and τ represents the lag order.

[0127] In this preferred embodiment, the temperature-humidity interaction characteristic, the stress-environment interaction characteristic, the stress corrosion sensitivity index, the stress intensity factor, the corrosion rate, the historical statistical characteristic, and the historical time-frequency characteristic are calculated from the preprocessed historical operating condition time series data.

[0128] Step S102: The first M features with the largest importance scores in the first features are taken as the second features; principal component analysis is performed on the second features, and the principal components with a cumulative contribution rate greater than a preset contribution rate threshold are retained to obtain the third features; wherein N and M are positive integers;

[0129] Specifically, the importance score of each first feature is calculated by the following formula:

[0130]

[0131] wherein Importance(f i ) represents the importance score value corresponding to the i'th first feature, B represents the number of decision trees in the recursive feature elimination method, b" represents the b'th decision tree in the recursive feature elimination method, T b represents the set of decision trees in the recursive feature elimination method, p(t') represents the sample proportion of node t', and Δ i (t') represents the importance contribution of feature f i at node t'.

[0132] Specifically, the principal component analysis method is used to reduce the dimension of the second features:

[0133] Y" = W T X'

[0134] In the formula, Y" represents the feature matrix corresponding to the second feature after dimension reduction, X' represents the original high-dimensional feature matrix, and W represents the principal component projection matrix.

[0135] The principal component projection matrix is calculated by the following formula:

[0136] C = WAW T

[0137] In the formula, C represents the covariance matrix of the original high-dimensional feature matrix X'.

[0138] Finally, the principal components whose cumulative contribution rate exceeds 95% are retained:

[0139]

[0140] In the formula, λ i″ represents the ith characteristic value, k represents the number of retained principal components, and p' represents the number of second features.

[0141] Step S103: Obtain the working condition time series data of the overhead ground wire to be predicted in the current period, and determine the corresponding second feature value and third feature value according to the above time series data;

[0142] Specifically, since the parameters corresponding to each second feature and third feature have been determined through the foregoing steps, in step S103, the feature values corresponding to the overhead ground wire to be predicted in the current period are directly obtained as the second feature values and the third feature values. Illustratively, if it has been determined that the second feature is temperature, then the second feature value of the overhead ground wire to be predicted in the current period is the temperature time series data in the current period.

[0143] In another preferred embodiment, the historical working condition time series data is standardized and preprocessed to obtain preprocessed historical working condition time series data, which includes:

[0144] According to the historical working condition time series data, the average value corresponding to each historical working condition time series data is calculated;

[0145] Specifically, the average value corresponding to each historical working condition time series data is calculated by the following formula:

[0146]

[0147] In the formula, μ represents the average value, n' represents the total number of historical working condition time series data, and x idenotes the i-th historical working condition time series data.

[0148] According to the average value and the corresponding historical working condition time series data, the standard deviation corresponding to each historical working condition time series data is calculated;

[0149] Specifically, the standard deviation is calculated by the following formula:

[0150]

[0151] In the formula, σ' represents the standard deviation.

[0152] According to the average value and the standard deviation, the historical working condition time series data is corrected for outliers to obtain first corrected historical sample data;

[0153] The first corrected historical sample data is denoised to obtain second corrected historical sample data;

[0154] Specifically, the adaptive sliding window median filtering algorithm is used to remove data noise, and the filtering window size is adaptively adjusted according to the local variance of the signal:

[0155]

[0156] In the formula, w i′ denotes the filtering window size of the i'-th sampling point in any first corrected historical sample data, w base denotes the basic window size, which is usually 5, α" denotes the adjustment coefficient, which is usually 0.5, σ local (i') denotes the local variance centered on the i'-th sampling point, σ global denotes the global variance of the historical working condition time series data.

[0157] Specifically, the data denoised by the median filtering algorithm is represented by the following formula:

[0158]

[0159] In the formula, x filtered (i') denotes the filtered value of the i'-th sampling point, and x(i') denotes the data corresponding to the i'-th sampling point.

[0160] For each historical working condition time series data, the trend term and the seasonal term corresponding to the missing value are calculated according to the historical working condition time series data;

[0161] Specifically, the multiple imputation method based on time series characteristics is used to process the missing value. Time series data can usually be decomposed into trend, seasonality and random components, represented as:

[0162] x(t) = T(t) + S(t) + R(t)

[0163]

[0164] Where x(t) represents the historical operating time series data, T(t) represents the trend term, S(t) represents the seasonal term, and R(t) represents the random term, which is the remainder after removing the trend and seasonal components. 2m′+1 represents the moving average window size. For hourly data, m′=12 (24-hour window) is usually used; for minute-level or second-level data, m′=60 is used. m′ represents the moving window size in the multiple interpolation method based on time series characteristics. x(t+j) represents the observation value offset by j time units relative to time t. t represents time t, j represents the index of the time offset, and p represents the period length. represents the seasonal mean of the mod p-th phase position.

[0165] According to the above trend item and seasonal item, the corresponding missing values ​​are interpolated to obtain the second corrected historical sample data;

[0166] Specifically, the missing value interpolation formula is:

[0167]

[0168] Where, Indicates missing value t miss The interpolated value of Indicates missing value t miss The estimated value of the trend term, Indicates missing value t miss The seasonal term estimate of Δt represents the time interval, ω1 and ω2 represent weight coefficients, which are usually 0.25, and w i′ Indicates the filter window size of the i'th data point, x filtered Represents the data after median filtering.

[0169] For each historical operating condition time series data, determine the median and absolute median difference of the historical operating condition time series data according to the historical operating condition time series data;

[0170] According to the above median and the above absolute median difference, the corresponding historical operating condition time series data is standardized to obtain the standardized historical operating condition time series data, and the preprocessed historical operating condition time series data is obtained.

[0171] Specifically, the improved Z-score standardization method is used to standardize the data considering the skewness of the data distribution:

[0172]

[0173] Where z i′Xi' represents the normalized value of the sampling point i, Xi represents the median value, which is used to represent the robust mean, Xi represents the robust standard deviation estimate, and MAD represents the median absolute deviation.

[0174] Preferably, through the above data preprocessing process, high-quality standardized data is obtained, providing a reliable data basis for the subsequent method process and ensuring the accuracy and stability of the subsequent analysis results.

[0175] In this preferred embodiment, the preprocessed historical working condition time series data is obtained by performing outlier detection, noise filtering, missing value processing, and data standardization on the historical working condition time series data.

[0176] In another preferred embodiment, the above outlier correction of the historical working condition time series data according to the average value and the standard deviation to obtain the first corrected historical sample data includes:

[0177] For each historical working condition time series data, the first quartile, the third quartile, and the interquartile range are extracted from the historical working condition time series data.

[0178] Specifically, the first quartile is the 25th percentile, and the third quartile is the 75th percentile, which is used to measure the distribution range of the middle 50% of the data. The first quartile, the third quartile, and the interquartile range are represented as:

[0179] Q1 = P 25 ({x i}), Q3 = P 75 ({x i}), IQR = Q3 - Q1

[0180] In the formula, Q1 represents the first quartile, P 25 25th percentile, Q3 represents the third quartile, P 75 75th percentile, and IQR represents the interquartile range.

[0181] The first threshold is calculated according to the first quartile and the interquartile range.

[0182] Specifically, the first threshold is calculated by the following formula:

[0183] A1 = Q1 - 1.5 × IQR

[0184] In the formula, A represents the first threshold.

[0185] The second threshold is calculated according to the third quartile and the interquartile range.

[0186] Specifically, the second threshold is calculated by the following formula:

[0187] A2 = Q3 + 1.5 x IQR

[0188] In the formula, A2 represents the second threshold.

[0189] In the case that the historical operating condition time series data is less than the first threshold, greater than the second threshold, or the absolute value of the difference between the historical operating condition time series data and the corresponding average value is greater than the standard deviation of the preset multiple, the historical operating condition time series data is regarded as an abnormal value, and the abnormal value is corrected to obtain first corrected historical sample data.

[0190] Specifically, the abnormal value judgment criterion is represented by the following formula:

[0191] Abnormal value = {x i | x i < Q1 - 1.5 x IQR or x i > Q3 + 1.5 x IQR or |x i - μ | > 3σ

[0192] Specifically, the abnormal value judgment criterion combines the robust method based on quantile and the classical method based on standard deviation, and can effectively identify various types of abnormal values.

[0193] Preferably, the detected abnormal value is replaced by a boundary value or corrected by an interpolation method.

[0194] In this preferred embodiment, the historical operating condition time series data is corrected for abnormal values by the average value and the standard deviation to obtain first corrected historical sample data.

[0195] Step S104: Based on a plurality of corrosion prediction models constructed by different algorithms, SCC prediction values corresponding to each corrosion prediction model are calculated according to the operating condition time series data, a plurality of second characteristic values and third characteristic values.

[0196] In a preferred embodiment, the SCC prediction values corresponding to each corrosion prediction model are calculated according to the operating condition time series data, a plurality of second characteristic values and third characteristic values based on a plurality of corrosion prediction models constructed by different algorithms, including:

[0197] The operating condition time series data is input into a first preset corrosion prediction model, so that the first preset corrosion prediction model predicts a first SCC prediction value according to the operating condition time series data.

[0198] Specifically, the first preset corrosion prediction model is constructed based on a deep neural network, and therefore the first preset corrosion prediction model adopts a multi-layer feedforward neural network structure and has strong function approximation capability. The corresponding network forward propagation is represented by the following formula:

[0199] a (l) (l) (W (l) a (l-1) +b (l) )

[0200] In the formula, a (l) represents the output of the lth layer, f (l) represents the activation function of the lth layer, W (l) represents the weight matrix of the lth layer, a (l-1) represents the output of the (l-1)th layer, and b (l) represents the bias vector of the lth layer.

[0201] Specifically, the activation function adopts a ReLU function, which is represented by the following formula:

[0202] f(x)=max(0,x)

[0203] In the formula, x represents the net input of the current layer.

[0204] The second preset corrosion prediction model and the third preset corrosion prediction model are input with the second characteristic values, so that the second preset corrosion prediction model and the third preset corrosion prediction model predict the second SCC prediction value and the third SCC prediction value according to the second characteristic values.

[0205] Specifically, the preset second stress corrosion prediction model is a support vector regression model with a radial basis function kernel, which is suitable for processing nonlinear regression problems. The kernel function of the second preset corrosion prediction model is:

[0206]

[0207] In the formula, K(x i , x j ) represents the kernel function value, and represent the historical working condition time series data of two different historical sample overhead ground wires, and γ represents the kernel parameter.

[0208] The optimization objective function corresponding to the second preset corrosion prediction model is:

[0209]

[0210] ​where w represents a weight vector, b represents a bias term, C represents a regularization parameter, n' represents a total number of training samples, and ξ i and denotes a slack variable.

[0211] A constraint condition corresponding to the second preset corrosion prediction model is:

[0212] y i -w T φ(x i )-b≤ε+ξ i

[0213]

[0214] where y i represents a true corrosion measurement result of the i-th sample, ε represents an insensitive loss function parameter, and φ(x) represents a feature mapping function.

[0215] Specifically, the third preset corrosion prediction model is constructed based on a random forest model. The random forest model improves prediction stability by integrating multiple decision trees and has good anti-overfitting capability. A splitting criterion of a single decision tree is expressed by minimizing a mean square error (MSE) as follows:

[0216]

[0217] where MSE represents the mean square error, y i represents a true label corresponding to the i-th sample, represents a mean value of all third SCC prediction values, and n represents a sample number.

[0218] An optimal split point of the third preset corrosion prediction model is determined by the following formula:

[0219]

[0220] where split * represents the optimal split point, n L represents a sample number corresponding to a left child node after splitting, n R represents a sample number corresponding to a right child node after splitting, MSE L represents a mean square error corresponding to the left child node after splitting, and MSE R represents a mean square error corresponding to the right child node after splitting.

[0221] A prediction result of the third preset corrosion prediction model is expressed by the following formula:

[0222]

[0223] where denotes the third SCC prediction value, B denotes the number of decision trees, b’ denotes the b’th decision tree, T b′ (x) denotes the prediction result of the b’th decision tree.

[0224] The out-of-bag error estimate of the third preset corrosion prediction model is represented by the following formula:

[0225]

[0226] In the formula, OOB Error represents the out-of-bag error, denotes the out-of-bag prediction value of sample i.

[0227] A plurality of third characteristic values are input into a fourth preset corrosion prediction model, so that the fourth preset corrosion prediction model predicts a fourth SCC prediction value according to the plurality of third characteristic values;

[0228] Specifically, the fourth preset corrosion prediction model is constructed based on a gradient boosting decision tree model, and the prediction performance is gradually optimized by a gradient boosting algorithm, and has strong non-linear modeling capability. The loss function of the fourth preset corrosion prediction model is:

[0229]

[0230] In the formula, L(y, F(x)) represents the loss function of the fourth preset corrosion prediction model, y represents the true SCC risk label (i.e. the corresponding historical SCC value), and F(x) represents the model prediction function of the fourth preset corrosion prediction model.

[0231] The gradient of the fourth preset corrosion prediction model in the training process is calculated by the following formula:

[0232]

[0233] In the formula, r im denotes the negative gradient residual of sample i in the mth iteration, L(y i , F(x i )) denotes the loss function of sample i in the fourth preset corrosion prediction model, and F(x i ) denotes the model prediction function of sample i in the fourth preset corrosion prediction model.

[0234] The updating process of the fourth preset corrosion prediction model in the mth iteration is represented by the following formula:

[0235] F m (x) = F m-1 (x) + v·h m (x)

[0236] In the formula, F m(x) represents the fourth preset corrosion prediction model obtained after the mth iteration, F m-1 (x) represents the fourth preset corrosion prediction model obtained after the m-1th iteration, v represents the learning rate, h m (x) represents the mth regression tree used to fit the negative gradient residual.

[0237] The first preset corrosion prediction model is constructed based on a deep neural network, the second preset corrosion prediction model is constructed based on a support vector machine model, the third preset corrosion prediction model is constructed based on a random forest model, and the fourth preset corrosion prediction model is constructed based on a gradient boosting decision tree model.

[0238] In this preferred embodiment, the first preset corrosion prediction model is constructed based on a deep neural network, the second preset corrosion prediction model is constructed based on a support vector machine model, the third preset corrosion prediction model is constructed based on a random forest model, and the fourth preset corrosion prediction model is constructed based on a gradient boosting decision tree model.

[0239] In another preferred embodiment, the training of the first preset corrosion prediction model comprises:

[0240] The preprocessed historical operating condition time series data and the corresponding historical SCC values are input into the first preset corrosion prediction model to be trained for iterative training until the loss function converges, thereby generating the first preset corrosion prediction model.

[0241] In each iteration training, the current first historical SCC prediction value is predicted according to the current preprocessed historical operating condition time series data, the current loss function is calculated according to the current first historical SCC prediction value and the corresponding historical SCC value, and it is determined whether the current loss function converges. If the current loss function converges, the current first preset corrosion prediction model is used as the trained first preset corrosion prediction model. Otherwise, the parameters in the current first preset corrosion prediction model are adjusted and the training is continued.

[0242] Specifically, the loss function of the first preset corrosion prediction model uses the mean square error plus a regularization term, which is represented as:

[0243]

[0244] In the formula, L' represents the loss function of the first preset corrosion prediction model, represents the ith historical operating condition time series data, λ represents the regularization coefficient, represents the Frobenius norm, and L represents the total number of layers in the first preset corrosion prediction model.

[0245] The first preset corrosion prediction model updates the weight through a back propagation algorithm, and a process of updating the weight is represented by the following formula:

[0246]

[0247] In the formula, η represents a learning rate.

[0248] In this preferred embodiment, the trained first preset corrosion prediction model is obtained by iteratively training the first preset corrosion prediction model.

[0249] Step S105: The prediction probability of corrosion fracture of the to-be-predicted overhead ground wire is calculated according to the preset probability mapping function and the SCC prediction values corresponding to different corrosion prediction models.

[0250] In a preferred embodiment, the prediction probability of corrosion fracture of the to-be-predicted overhead ground wire is calculated according to the preset probability mapping function and the SCC prediction values corresponding to different corrosion prediction models, and includes:

[0251] The model weight corresponding to each corrosion prediction model is obtained;

[0252] The final SCC value is calculated according to the model weight, the first SCC prediction value, the second SCC prediction value, the third SCC prediction value and the fourth SCC prediction value;

[0253] Specifically, the adaptive weighted average method is used to fuse the SCC prediction values obtained by each corrosion prediction model to fully give play to the advantages of each algorithm. The model weight can be determined according to the following formula:

[0254]

[0255] In the formula, w k is the weight of the kth model, β1 is a temperature parameter, RMSE k is the root mean square error of the kth corrosion prediction model, and j represents an index variable of the model number, which is used to traverse all K sub-models.

[0256] Integrated prediction result:

[0257]

[0258] In the formula, S represents the final SCC value, represents the SCC prediction value of the kth corrosion prediction model.

[0259] Dynamic weight adjustment:

[0260] w k (t) = w k (t-1) + α1 × sign (ek (t)) x exp(-|e k (t)|

[0261] wherein e k (t) is the prediction error of the kth corrosion prediction model at time t, and a1 is an adaptive adjustment coefficient.

[0262] According to a preset probability mapping function, the final SCC value is subjected to probability mapping to obtain a prediction probability of corrosion fracture of the above-mentioned overhead ground wire to be predicted.

[0263] Specifically, the final SCC value is converted into a probability, and a mapping relationship between an index and the probability is established. The probability mapping function adopts a Sigmoid function:

[0264]

[0265] wherein P failure represents the prediction probability, S represents the final SCC value, S0 represents a critical threshold value (usually 0.5), and k' represents a steepness parameter (usually 10).

[0266] Specifically, in the training process, the probability calibration adopts a Platt scaling method:

[0267]

[0268] wherein P calibrated represents a calibrated failure probability, and A and B represent parameters in the probability calibration.

[0269] Specifically, the parameters A and B are obtained through maximum likelihood estimation:

[0270]

[0271] wherein y i represents a true probability value, and p i represents a prediction probability in the training process.

[0272] Preferably, taking a No. 157 tower of a 110kV transmission line in a certain coastal area as an example, the overhead conductor material on the tower is galvanized steel strand OPGW-24B1, the operation life is 18.5 years, the conductor diameter is 12.4mm, the environmental temperature is 38.2℃ (exceeding the design temperature 35℃), the relative humidity is 92.5% (close to saturation), the pH value is 4.1 (strong acid environment), the chloride ion concentration is 156.7mg / L (exceeding the critical value 50mg / L), and the SO2 concentration is 45.3mg / m 3, rainfall is 4.2mm / h, wind speed is 15.8m / s; stress is 285.4MPa (close to yield strength 350MPa), tension is 17850N (exceeds design tension 15000N), vibration acceleration is 1.23m / s 2 (exceeds safety threshold 0.5m / s 2 ), frequency is 3.8Hz, strain is 0.001427, surface roughness is 4.8μm (far more than new line standard 1.5μm), coating thickness is 18.5μm (lower than standard value 50μm), surface damage level is 4 / 5 (severe damage). Illustratively, based on the above data, the coastal area environmental condition risk assessment schematic diagram is shown in Figure 2 , the mechanical condition assessment schematic diagram is shown in Figure 3 , and the risk factor contribution degree analysis schematic diagram is shown in Figure 4 .

[0273] Subsequently, model prediction is performed through the above data, and the obtained prediction probability is 0.913. Illustratively, the evaluation result schematic diagram is shown in Figure 5 , and it can be seen from Figure 5 that the final evaluation result is: the risk level is extremely high, and the warning level is red warning.

[0274] In this preferred embodiment, by performing weighted calculation on the SCC prediction values corresponding to different corrosion prediction models, the prediction probability of corrosion fracture of the to-be-predicted overhead ground wire is calculated by using a preset probability mapping function.

[0275] On the basis of the above method embodiment, the present application correspondingly provides a device embodiment.

[0276] As shown in Figure 6 , an embodiment of the present application provides an overhead ground wire stress corrosion cracking prediction device based on machine learning, comprising:

[0277] a feature extraction module, a principal component analysis module, a data acquisition module, an SCC prediction module and a probability mapping module;

[0278] The feature extraction module is used to acquire historical working condition time series data of an overhead ground wire sample, determine historical SCC values corresponding to the overhead ground wire sample according to the historical working condition time series data, determine a plurality of material service performance characteristics corresponding to the overhead ground wire sample according to the historical working condition time series data, calculate mutual information of each material service performance characteristic and the corresponding historical SCC value, and take the first N material service performance characteristics with the maximum mutual information as a plurality of first characteristics.

[0279] The principal component analysis module is configured to take the first M features with the largest importance scores in each first feature as a plurality of second features, perform principal component analysis on the plurality of second features, and retain principal components with a cumulative contribution rate greater than a preset contribution rate threshold to obtain a plurality of third features, wherein N and M are positive integers.

[0280] The data acquisition module is configured to acquire working condition time series data of the overhead ground wire to be predicted in a current period, and determine a plurality of second feature values and third feature values corresponding to the time series data.

[0281] The SCC prediction module is configured to calculate SCC prediction values corresponding to each corrosion prediction model based on the working condition time series data, the plurality of second feature values and the third feature values according to a plurality of corrosion prediction models constructed by different algorithms.

[0282] The probability mapping module is configured to calculate a prediction probability of corrosion fracture of the overhead ground wire to be predicted according to a preset probability mapping function and the SCC prediction values corresponding to different corrosion prediction models.

[0283] It should be noted that the apparatus embodiments described above are merely illustrative, and the modules described above as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiments provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor. The above schematic diagram is only an example of an overhead ground wire stress corrosion cracking prediction apparatus based on machine learning, and does not constitute a limitation on an overhead ground wire stress corrosion cracking prediction apparatus based on machine learning, which can include more or fewer components than the diagram, or combine certain components, or different components.

[0284] On the basis of the above method embodiments, the present application correspondingly provides terminal device embodiments.

[0285] Another embodiment of the present application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor executes the computer program to implement the above-mentioned overhead ground wire stress corrosion cracking prediction method based on machine learning according to any one of the embodiments of the present application.

[0286] For example, in this embodiment, the computer program described above can be divided into one or more modules, the one or more modules are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the device;

[0287] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The device can include, but is not limited to, a processor, a memory, and the like.

[0288] The processor can be a central processing module (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or any conventional processor, and the like. The processor is the control center of the device, and is connected to various parts of the device through various interfaces and lines.

[0289] The memory can be used to store the computer program and / or the module, and the processor realizes various functions of the device by running or executing the computer program and / or the module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, and the like; in addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0290] On the basis of the method embodiment, the present application correspondingly provides a storage medium embodiment.

[0291] Another embodiment of the present application provides a storage medium including a stored computer program, wherein the computer program, when executed, controls a device in which the storage medium is located to perform the machine learning-based prediction method for stress corrosion cracking of an overhead ground wire according to any one of the embodiments of the present application.

[0292] In this embodiment, the storage medium is a computer-readable storage medium, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, an executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunications signal, a software distribution medium, and the like.

[0293] The above is a preferred embodiment of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements are also considered within the scope of protection of the present application.

Claims

1. A method for predicting stress corrosion cracking of overhead ground wires based on machine learning, characterized in that: include: Obtaining historical operating condition time series data of overhead ground wire samples, and determining historical SCC values ​​corresponding to the overhead ground wire samples based on the historical operating condition time series data; Determining several material service performance characteristics corresponding to the overhead ground wire sample based on historical operating condition time series data; Calculate the mutual information between each material's service performance characteristic and the corresponding historical SCC value; take the top N material service performance characteristics with the largest mutual information as the first features; The first M features with the largest importance scores among the first features are used as several second features; principal component analysis is performed on the several second features, and the principal components with cumulative contribution rates greater than a preset contribution rate threshold are retained to obtain several third features; where N and M are positive integers; Obtaining time series data of the working condition of the overhead ground wire to be predicted in the current time period, and determining corresponding second eigenvalues ​​and third eigenvalues ​​according to the time series data; Based on several corrosion prediction models constructed by different algorithms, the SCC prediction value corresponding to each corrosion prediction model is calculated according to the working condition time series data, several second eigenvalues ​​and third eigenvalues; According to the preset probability mapping function and the SCC prediction values ​​corresponding to different corrosion prediction models, the predicted probability that the overhead ground wire to be predicted will corrode and break is calculated.

2. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 1, characterized in that: The service performance characteristics of the material include: temperature-humidity interaction characteristics, stress-environment interaction characteristics, stress corrosion sensitivity index, stress intensity factor, corrosion rate, historical statistical characteristics and historical time-frequency characteristics The determining of several material service performance characteristics corresponding to the overhead ground wire sample based on the historical operating condition time series data includes: Performing standardized preprocessing on the historical operating condition time series data to obtain preprocessed historical operating condition time series data; The temperature-humidity interaction characteristics are calculated based on the historical ambient temperature and historical relative humidity in the preprocessed historical operating condition time series data; The stress-environment interaction characteristics are calculated based on the historical stress, historical chloride ion concentration and historical pH value in the pre-processed historical operating condition time series data; The stress corrosion sensitivity index is calculated based on the historical ambient temperature, historical stress, historical chloride ion concentration and historical pH value in the pre-processed historical operating condition time series data; The stress intensity factor is calculated based on the historical stress in the pre-processed historical working condition time series data; The corrosion rate is calculated based on the historical relative humidity, historical ambient temperature, historical chloride ion concentration and historical pH value in the pre-processed historical operating condition time series data; According to the preprocessed historical operating condition time series data, the historical mean, historical standard deviation, skewness, kurtosis and coefficient of variation corresponding to each preprocessed historical operating condition time series data are calculated, and the historical mean, historical standard deviation, skewness, kurtosis and coefficient of variation are used as the historical statistical features; According to the preprocessed historical operating condition time series data, the trend characteristics, periodic characteristics and autocorrelation characteristics corresponding to each preprocessed historical operating condition time series data are calculated, and the trend characteristics, periodic characteristics and autocorrelation characteristics are used as the historical time-frequency characteristics of the historical operating condition time series data.

3. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 2, characterized in that: The standardization preprocessing of the historical operating condition time series data to obtain the preprocessed historical operating condition time series data comprises: Calculate the average value corresponding to each historical operating condition time series data according to the historical operating condition time series data; Calculate the standard deviation of each historical operating condition time series data based on the average value and the corresponding historical operating condition time series data; Performing outlier correction on the historical operating condition time series data according to the mean value and the standard deviation to obtain first corrected historical sample data; De-noising the first corrected historical sample data to obtain second corrected historical sample data; For each historical operating condition time series data, calculate the trend term and seasonal term corresponding to the missing value according to the historical operating condition time series data; According to the trend item and the seasonal item, interpolation processing is performed on the corresponding missing values ​​to obtain second corrected historical sample data; For each historical operating condition time series data, determine the median and absolute median difference of the historical operating condition time series data according to the historical operating condition time series data; According to the median and the absolute median difference, the corresponding historical operating condition time series data is standardized to obtain the standardized historical operating condition time series data, and the preprocessed historical operating condition time series data is obtained.

4. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 3, characterized in that: The step of performing outlier correction on the historical operating condition time series data according to the mean value and the standard deviation to obtain first corrected historical sample data includes: For each historical operating condition time series data, extract the first quartile, the third quartile, and the interquartile range from the historical operating condition time series data; Calculating a first threshold value according to the first quartile and the interquartile range; Calculate a second threshold value according to the third quartile and the interquartile range; When the historical operating condition time series data is less than the first threshold value, the historical operating condition time series data is greater than the second threshold value; or the absolute value of the difference between the historical operating condition time series data and the corresponding average value is greater than a preset multiple of the standard deviation, the historical operating condition time series data is regarded as an abnormal value, and the abnormal value is corrected to obtain the first corrected historical sample data.

5. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 4, characterized in that: The method of calculating the SCC prediction value corresponding to each corrosion prediction model based on the operating condition time series data, the second eigenvalues, and the third eigenvalues ​​based on the corrosion prediction models constructed by different algorithms includes: Inputting the operating condition time series data into a first preset corrosion prediction model, so that the first preset corrosion prediction model predicts a first SCC prediction value based on the operating condition time series data; Inputting the plurality of second characteristic values ​​into a second preset corrosion prediction model and a third preset corrosion prediction model, respectively, so that the second preset corrosion prediction model and the third preset corrosion prediction model predict a second SCC prediction value and a third SCC prediction value based on the plurality of second characteristic values; inputting the plurality of third eigenvalues ​​into a fourth preset corrosion prediction model, so that the fourth preset corrosion prediction model predicts a fourth SCC prediction value based on the plurality of third eigenvalues; Among them, the first preset corrosion prediction model is constructed based on a deep neural network, the second preset corrosion prediction model is constructed based on a support vector machine model, the third preset corrosion prediction model is constructed based on a random forest model, and the fourth preset corrosion prediction model is constructed based on a gradient boosting decision tree model.

6. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 5, characterized in that: The training of the first preset corrosion prediction model includes: Inputting the preprocessed historical operating condition time series data and the corresponding historical SCC values ​​into a first preset corrosion prediction model to be trained for iterative training until the loss function converges, thereby generating the first preset corrosion prediction model; Among them, during each iterative training, the current first historical SCC prediction value is predicted based on the current preprocessed historical operating condition time series data; the current loss function is calculated based on the current first historical SCC prediction value and the corresponding historical SCC value, and it is judged whether the current loss function converges; if the current loss function converges, the current first preset corrosion prediction model is used as the trained first preset corrosion prediction model; otherwise, the parameters in the current first preset corrosion prediction model are adjusted and training continues.

7. The method for predicting stress corrosion cracking of overhead ground wires based on machine learning according to claim 6, characterized in that: The method of calculating the predicted probability of the overhead ground wire to be predicted to be corroded and fractured according to the preset probability mapping function and the SCC predicted values ​​corresponding to different corrosion prediction models includes: Obtain the model weight corresponding to each corrosion prediction model; Calculating a final SCC value according to the model weight, the first SCC prediction value, the second SCC prediction value, the third SCC prediction value, and the fourth SCC prediction value; The final SCC value is probability mapped according to a preset probability mapping function to obtain a predicted probability that the overhead ground wire to be predicted will corrode and break.

8. A device for predicting stress corrosion cracking of overhead ground wires based on machine learning, characterized in that: include: Feature extraction module, principal component analysis module, data acquisition module, SCC prediction module and probability mapping module; The feature extraction module is used to obtain historical operating condition time series data of overhead ground wire samples, determine the historical SCC value corresponding to the overhead ground wire samples based on the historical operating condition time series data; and determine several material service performance characteristics corresponding to the overhead ground wire samples based on the historical operating condition time series data; Calculate the mutual information between each material's service performance characteristic and the corresponding historical SCC value; take the top N material service performance characteristics with the largest mutual information as the first features; The principal component analysis module is configured to select the first M features with the largest importance scores among the first features as a plurality of second features; perform principal component analysis on the plurality of second features, and retain principal components whose cumulative contribution rates are greater than a preset contribution rate threshold, to obtain a plurality of third features; wherein N and M are positive integers; The data acquisition module is used to obtain time series data of the working condition of the overhead ground wire to be predicted in the current time period, and determine the corresponding second eigenvalues ​​and third eigenvalues ​​according to the time series data; The SCC prediction module is used to calculate the SCC prediction value corresponding to each corrosion prediction model based on the operating condition time series data, the second eigenvalues ​​and the third eigenvalues ​​based on the corrosion prediction models constructed by different algorithms; The probability mapping module is used to calculate the predicted probability that the overhead ground wire to be predicted will corrode and break based on a preset probability mapping function and the SCC prediction values ​​corresponding to different corrosion prediction models.

9. A terminal device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements a method for predicting stress corrosion cracking of overhead ground wires based on machine learning as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute a method for predicting stress corrosion cracking of an overhead ground wire based on machine learning as described in any one of claims 1 to 7.

Images