Method and system for evaluating insulation aging state and predicting service life of power transmission equipment

Abstract

The application provides a power transmission network equipment insulation aging state evaluation and life prediction method and system, relates to the power equipment state monitoring technical field, and comprises the following steps: acquiring multi-dimensional operation monitoring data, extracting features through multi-scale time-frequency decomposition, and obtaining decoupling feature vectors through orthogonal transformation; calculating Mahalanobis distance to obtain aging state deviation metric value; constructing stress cumulative damage factor to nonlinearly modulate the degradation rate; and predicting the aging state trajectory and determining the remaining life based on the corrected degradation rate. The application can accurately evaluate the insulation aging state and predict the remaining life of the equipment, and improves the evaluation accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a method and system for assessing the insulation aging status and predicting the lifespan of power transmission network equipment. Background Technology

[0002] The safe and stable operation of power systems is of great significance to national economic and social development. As a core component of the power system, the insulation performance of transmission network equipment is a key factor in ensuring its safe operation. With the continuous expansion of the power grid and the extension of equipment service life, the problem of insulation aging in transmission network equipment is becoming increasingly prominent. The insulation performance of the equipment gradually deteriorates, seriously affecting its safety and reliability. Accurately assessing the insulation aging status of transmission network equipment and predicting its remaining lifespan is of great importance for rationally planning equipment maintenance, extending equipment service life, and reducing operation and maintenance costs.

[0003] Currently, technologies for assessing the aging status and predicting the lifespan of power transmission network equipment still have some shortcomings and deficiencies. Most existing assessment methods are based on single parameters or simple parameter combinations, making it difficult to comprehensively reflect the complex process of equipment insulation aging, resulting in a lack of comprehensiveness and accuracy in the assessment results. These methods cannot effectively handle the coupling relationships in multi-dimensional operational monitoring data and struggle to extract characteristic information that truly reflects the insulation aging status. Traditional lifespan prediction methods typically employ fixed model structures, failing to adapt to changes in the aging characteristics of equipment under different operating environments and load conditions. They also insufficiently consider the cumulative effects of environmental factors, load fluctuations, and other external stresses, reducing the reliability of the prediction results. Particularly under complex operating conditions, these methods struggle to accurately capture the nonlinear characteristics and abrupt changes in the insulation aging process. Existing technologies lack uncertainty analysis for aging status assessment and lifespan prediction, failing to provide confidence intervals for prediction results, leading to a lack of scientific basis for maintenance decisions. Because the equipment aging process is affected by various random factors, single-point prediction values ​​often cannot meet the needs of risk assessment and decision optimization, making it difficult to provide reliable support for equipment management. Summary of the Invention

[0004] This invention provides a method and system for assessing the insulation aging status and predicting the lifespan of power transmission network equipment, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for assessing the insulation aging status and predicting the lifespan of power transmission network equipment, comprising:

[0006] Acquire multi-dimensional operational monitoring data of power transmission network equipment;

[0007] Multi-scale time-frequency decomposition is performed on multi-dimensional operation monitoring data. Transient impact components and long-term trend components are extracted from each scale component after decomposition. The components are then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector.

[0008] The decoupled feature vectors are embedded into a high-dimensional state space, and the Mahalanobis distance between the current state point and the historical healthy state cluster is calculated to obtain the aging state deviation metric.

[0009] The cumulative action mode of stress components is identified from multidimensional operation monitoring data and a stress accumulation damage factor is constructed. The deviation of aging state is extracted and piecewise fitted to obtain the degradation rate. The degradation rate is nonlinearly modulated based on the stress accumulation damage factor to obtain the corrected degradation rate.

[0010] Using the modified degradation rate as the initial evolution rate, the historical sequence of stress accumulation damage factor is fitted and extrapolated to obtain the aging state prediction trajectory through recursive calculation. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining life prediction value.

[0011] In one optional embodiment, multi-scale time-frequency decomposition is performed on the multi-dimensional operational monitoring data. Transient impact components and long-term trend components are extracted from each scale component after decomposition, and then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector, which includes:

[0012] Multiple scale levels are set for multidimensional operation monitoring data. Time-frequency decomposition is performed at each scale level to obtain the corresponding scale components. The time-domain change rate of each scale component is calculated. Scale components with time-domain change rates exceeding a preset change rate threshold are extracted as transient impact components, and scale components with time-domain change rates below a preset change rate threshold are extracted as long-term trend components.

[0013] Calculate the cross-correlation matrix between the transient shock component and the long-term trend component, perform eigenvalue decomposition on the cross-correlation matrix to obtain a set of eigenvectors, select the eigenvector corresponding to the largest eigenvalue to determine the first basis vector, select the eigenvector corresponding to the smallest eigenvalue to determine the second basis vector, and combine the first basis vector and the second basis vector to construct the transformation matrix.

[0014] The transient impact component is projected onto the first coordinate axis of the decoupled feature space through a transformation matrix to obtain the transient decoupled component. The long-term trend component is projected onto the second coordinate axis through a transformation matrix to obtain the trend decoupled component. The rotation angle of the basis vector of the transformation matrix is ​​iteratively adjusted and orthogonalized. The transformation matrix is ​​adaptively updated until the residual correlation coefficient drops below the preset decoupled threshold. The transient decoupled component and the trend decoupled component are combined to form the decoupled feature vector.

[0015] In an optional embodiment, the step of iteratively adjusting the rotation angle of the basis vectors of the transformation matrix and performing orthogonalization, and adaptively updating the transformation matrix until the residual correlation coefficient drops below a preset decoupling threshold, includes:

[0016] Calculate the deviation between the absolute value of the residual correlation coefficient in the current iteration and the preset decoupling threshold, and determine the adjustment magnitude coefficient of the transformation matrix based on the ratio of the deviation to the preset deviation benchmark value;

[0017] A rotational perturbation is applied to the first and second basis vectors in the current transformation matrix. The angle of the rotational perturbation is determined by the product of the adjustment amplitude coefficient and the preset unit rotation angle. After the rotational perturbation, the first and second basis vectors are subjected to Schmitt orthogonalization to obtain the first and second updated basis vectors. The updated transformation matrix is ​​then recombined.

[0018] The transient shock component and the long-term trend component are reprojected using the updated transformation matrix to obtain the new transient decoupling component and the new trend decoupling component. The new residual correlation coefficient is calculated. The change in correlation coefficient is determined based on the new residual correlation coefficient and the residual correlation coefficient of the previous round. When the change in correlation coefficient is negative, the current preset unit rotation angle is kept unchanged. When the change in correlation coefficient is positive, the preset unit rotation angle is updated to the product of the current value and the preset reduction factor.

[0019] Repeat the iterative process until the absolute value of the residual correlation coefficient drops below the preset decoupling threshold.

[0020] In one alternative embodiment, identifying the cumulative action mode of stress components from multidimensional operational monitoring data and constructing a stress accumulation damage factor includes:

[0021] Temperature stress components, electric field stress components, and mechanical stress components are extracted from multidimensional operation monitoring data. The cumulative stress of each stress component is obtained by time integration over historical periods.

[0022] The cumulative stress of each stress component is constructed into a stress accumulation matrix in chronological order. Singular value decomposition is performed on the stress accumulation matrix to determine the left singular vector matrix and the right singular vector matrix. The dominant stress mode vector corresponding to the maximum singular value is extracted from the left singular vector matrix, and the dominant time evolution vector corresponding to the maximum singular value is extracted from the right singular vector matrix. The dominant stress mode vector and the dominant time evolution vector are reconstructed by the outer product of the dominant stress mode vector and the dominant time evolution vector to obtain the dominant mode component of stress accumulation.

[0023] The stress accumulation trajectory is obtained by reconstructing the phase space of the dominant mode components. The correlation dimension of the stress accumulation trajectory is calculated to determine the complexity factor. The maximum Lyapunov exponent of the stress accumulation trajectory is calculated to determine the chaos factor. The complexity factor and the chaos factor are nonlinearly fused to obtain the stress accumulation damage factor.

[0024] In one optional embodiment, the degradation rate is obtained by extracting the deviation of the aging state from the metric value and performing piecewise fitting, and the modified degradation rate is obtained by nonlinearly modulating the degradation rate based on the stress accumulation damage factor, including:

[0025] Extract the time series of aging state deviation from the metric value, set multiple time segment windows on the time series, and perform segmented fitting within each time segment window to obtain the degradation rate of each time segment window.

[0026] A two-dimensional coordinate space is constructed by constructing the degradation rate corresponding to each time segment window and the stress cumulative damage factor in the same time period. The probability density statistics of historical data points in the two-dimensional coordinate space are used to obtain the joint probability distribution. The value of the stress cumulative damage factor is fixed in the joint probability distribution and the average value of the corresponding degradation rate is calculated to determine the baseline degradation response. The difference between the actual degradation rate and the baseline degradation response is calculated to obtain the degradation anomaly degree.

[0027] The width of the modulation window is determined based on the degree of degradation anomaly, and the modulation window is constructed. The historical data points are then subjected to sliding weighted summation along the direction of the stress accumulation damage factor in the two-dimensional coordinate space to obtain the damage influence distribution. The values ​​at the corresponding positions of each time segment window are extracted from the damage influence distribution as damage influence coefficients. The damage influence coefficients are multiplied by the stress accumulation damage factor to determine the modulation amount. The degradation rate of each time segment window is added to the modulation amount of the corresponding time segment window to obtain the corrected degradation rate.

[0028] In one optional embodiment, using the modified degradation rate as the initial evolution rate, the historical sequence of the stress accumulation damage factor is fitted and extrapolated to recursively calculate the aging state prediction trajectory. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining lifetime prediction value, including:

[0029] The historical sequence of stress cumulative damage factor is extracted. The fitting relationship is determined by linear fitting between the value at each time point in the historical sequence and the values ​​at multiple previous time points. The pre-set prediction period is extrapolated step by step according to the fitting relationship to obtain the extrapolated value of stress cumulative damage factor at each time point in the prediction period.

[0030] The modified degradation rate is used as the initial evolution rate. Multiple time points are set in the prediction period. At each time point, the aging state deviation metric is calculated based on the current rate and the extrapolated value of the stress accumulation damage factor at the corresponding time point. The evolution rate is updated based on the difference between the aging state deviation metric values ​​of adjacent time points. The calculation is continued iteratively. The aging state deviation metric values ​​of each time point are connected to obtain the aging state prediction trajectory.

[0031] The historical sequence of stress accumulation damage factor is segmented and randomly sampled to obtain multiple sets of recombined sequences. The recombined sequences are fitted and extrapolated to obtain multiple sets of extrapolated values. Based on the multiple sets of extrapolated values, multiple aging state prediction trajectories are calculated. The maximum and minimum values ​​of the multiple aging state prediction trajectories are extracted at each time point and connected to form a fluctuation boundary.

[0032] Extract the moment when the preset failure threshold is first exceeded on the aging state prediction trajectory, determine the failure moment, and calculate the time difference between the failure moment and the current moment to determine the remaining life prediction value.

[0033] In one optional embodiment, the historical sequence of stress accumulation damage factor is segmented and randomly sampled to obtain multiple sets of recombined sequences, including:

[0034] Calculate the autocorrelation coefficient of the historical sequence, extract the lag step length when the autocorrelation coefficient first falls below the preset correlation threshold, and use the lag step length as the block length.

[0035] The historical sequence of stress accumulation damage factor is continuously divided according to the block length to obtain multiple historical blocks;

[0036] Randomly extract historical blocks with replacement, and extract a preset number of historical blocks each time. Then connect them in chronological order to form a recombined sequence.

[0037] Repeat the process until the preset number of times is reached, resulting in multiple sets of recombination sequences.

[0038] A second aspect of the present invention provides a system for assessing the insulation aging status and predicting the lifespan of power transmission network equipment, comprising:

[0039] The first unit is used to acquire multi-dimensional operation monitoring data of power transmission network equipment;

[0040] The second unit is used to perform multi-scale time-frequency decomposition on multi-dimensional operation monitoring data, extract transient impact components and long-term trend components from the decomposed scale components, and project them to the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector.

[0041] The third unit is used to embed the decoupled feature vector into the high-dimensional state space and calculate the Mahalanobis distance between the current state point and the historical healthy state cluster to obtain the aging state deviation metric.

[0042] The fourth unit is used to identify the cumulative action mode of stress components from multidimensional operation monitoring data and construct a stress cumulative damage factor, extract the deviation value of aging state and perform piecewise fitting to obtain the degradation rate, and obtain the corrected degradation rate by nonlinearly modulating the degradation rate based on the stress cumulative damage factor.

[0043] The fifth unit is used to fit and extrapolate the historical sequence of stress accumulation damage factor with the modified degradation rate as the initial evolution rate, recursively calculate the aging state prediction trajectory, calculate the fluctuation boundary by segmented random sampling, and extract the failure time to determine the remaining life prediction value.

[0044] A third aspect of the present invention provides an electronic device, comprising:

[0045] processor;

[0046] Memory used to store processor-executable instructions;

[0047] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0048] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0049] In this embodiment of the invention, by performing multi-scale time-frequency decomposition and orthogonal transformation on multi-dimensional operation monitoring data of power transmission network equipment, feature decoupling is achieved, effectively eliminating the interference between different features and improving the accuracy and reliability of condition assessment. The introduction of a stress accumulation damage factor to nonlinearly modulate the degradation rate overcomes the limitations of traditional linear life prediction methods, accurately capturing the nonlinear dynamic characteristics during equipment aging and adapting to life prediction needs under various complex operating conditions. By calculating the fluctuation boundary through segmented random sampling, a confidence interval for life prediction is constructed, providing not only a point estimate of the remaining life but also quantifying the uncertainty of the prediction, thus providing a more comprehensive risk assessment basis for maintenance decisions of power transmission network equipment. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the method for assessing the insulation aging status and predicting the lifespan of power transmission network equipment according to an embodiment of the present invention.

[0051] Figure 2 A flowchart for constructing the stress accumulation damage factor is provided. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0054] Figure 1 This is a flowchart illustrating the method for assessing the insulation aging status and predicting the lifespan of power transmission network equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0055] Acquire multi-dimensional operational monitoring data of power transmission network equipment;

[0056] Multi-scale time-frequency decomposition is performed on multi-dimensional operation monitoring data. Transient impact components and long-term trend components are extracted from each scale component after decomposition. The components are then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector.

[0057] The decoupled feature vectors are embedded into a high-dimensional state space, and the Mahalanobis distance between the current state point and the historical healthy state cluster is calculated to obtain the aging state deviation metric.

[0058] The cumulative action mode of stress components is identified from multidimensional operation monitoring data and a stress accumulation damage factor is constructed. The deviation of aging state is extracted and piecewise fitted to obtain the degradation rate. The degradation rate is nonlinearly modulated based on the stress accumulation damage factor to obtain the corrected degradation rate.

[0059] Using the modified degradation rate as the initial evolution rate, the historical sequence of stress accumulation damage factor is fitted and extrapolated to obtain the aging state prediction trajectory through recursive calculation. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining life prediction value.

[0060] In one optional implementation, the multi-dimensional operational monitoring data is decomposed into multi-scale time-frequency components. Transient impact components and long-term trend components are extracted from each scale component after decomposition, and then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector, which includes:

[0061] Multiple scale levels are set for multidimensional operation monitoring data. Time-frequency decomposition is performed at each scale level to obtain the corresponding scale components. The time-domain change rate of each scale component is calculated. Scale components with time-domain change rates exceeding a preset change rate threshold are extracted as transient impact components, and scale components with time-domain change rates below a preset change rate threshold are extracted as long-term trend components.

[0062] Calculate the cross-correlation matrix between the transient shock component and the long-term trend component, perform eigenvalue decomposition on the cross-correlation matrix to obtain a set of eigenvectors, select the eigenvector corresponding to the largest eigenvalue to determine the first basis vector, select the eigenvector corresponding to the smallest eigenvalue to determine the second basis vector, and combine the first basis vector and the second basis vector to construct the transformation matrix.

[0063] The transient impact component is projected onto the first coordinate axis of the decoupled feature space through a transformation matrix to obtain the transient decoupled component. The long-term trend component is projected onto the second coordinate axis through a transformation matrix to obtain the trend decoupled component. The rotation angle of the basis vector of the transformation matrix is ​​iteratively adjusted and orthogonalized. The transformation matrix is ​​adaptively updated until the residual correlation coefficient drops below the preset decoupled threshold. The transient decoupled component and the trend decoupled component are combined to form the decoupled feature vector.

[0064] In one specific implementation, multiple scale levels are set for the multidimensional operational monitoring data, including micro, meso, and macro scales. The micro scale corresponds to high-frequency signals with a sampling interval of 10 milliseconds; the meso scale corresponds to mid-frequency signals with a sampling interval of 1 second; and the macro scale corresponds to low-frequency signals with a sampling interval of 10 minutes. Time-frequency decomposition is performed at each scale level using wavelet transform technology, processing the monitoring data with wavelet basis functions of different scales. For the micro scale, a 5-level decomposition is performed using the db4 wavelet basis function; for the meso scale, a 4-level decomposition is performed using the sym8 wavelet basis function; and for the macro scale, a 3-level decomposition is performed using the coif3 wavelet basis function. In this way, the original signal is decomposed into multiple components with different frequency characteristics.

[0065] The time-domain rate of change of each scale component is calculated by dividing the absolute value of the difference between adjacent data points by the time interval. The threshold for the rate of change at the microscale is set to 0.5 units / millisecond, at the mesoscale to 0.2 units / second, and at the macroscale to 0.05 units / minute. Scale components with a time-domain rate of change exceeding the preset threshold are extracted as transient impact components, while those with a time-domain rate of change below the preset threshold are extracted as long-term trend components. Taking transformer operating data as an example, the extracted transient impact components include high-frequency pulse signals generated during insulation breakdown, while the long-term trend components reflect the slow degradation of the insulation material over time.

[0066] Calculate the cross-correlation matrix between the transient impact component and the long-term trend component. Perform eigenvalue decomposition on the cross-correlation matrix to obtain eigenvector groups and corresponding eigenvalues. Sort the eigenvalues ​​in descending order, select the eigenvector corresponding to the largest eigenvalue to determine the first basis vector, and select the eigenvector corresponding to the smallest eigenvalue to determine the second basis vector. Taking a transformer as an example, the largest eigenvalue is 3.75, and the corresponding eigenvector is [0.42, 0.38, 0.45, 0.36, 0.40, 0.41]; the smallest eigenvalue is 0.21, and the corresponding eigenvector is [0.15, -0.48, 0.22, 0.56, -0.38, 0.50]. Combine the first and second basis vectors to construct a transformation matrix with a dimension of n×2.

[0067] The transient impact component is projected onto the first coordinate axis of the decoupled feature space through a transformation matrix. The transient impact component vector is multiplied by the first column of the transformation matrix to obtain the transient decoupled component. The long-term trend component is projected onto the second coordinate axis through a transformation matrix. The long-term trend component vector is multiplied by the second column of the transformation matrix to obtain the trend decoupled component. The rotation angle of the basis vectors of the transformation matrix is ​​iteratively adjusted and orthogonalized. The correlation coefficient between the current transient decoupled component and the trend decoupled component is calculated. If the absolute value of the correlation coefficient is greater than the preset decoupled threshold of 0.05, the basis vectors are rotated by 5 degrees and orthogonalized using the Gram-Schmidt orthogonalization method. The correlation coefficient is recalculated, and the above process is repeated until the absolute value of the correlation coefficient is less than the preset decoupled threshold. The transformation matrix is ​​adaptively updated until the residual correlation coefficient drops below the preset decoupled threshold. The transient decoupled component and the trend decoupled component are combined to form the decoupled feature vector. In practical applications, the dimension of the decoupled feature vector is twice the dimension of the original feature vector.

[0068] A device insulation aging status assessment model was constructed based on decoupled feature vectors, employing a deep neural network structure with three hidden layers. Each layer has 32, 16, and 8 nodes respectively. The ReLU activation function was used, and the output layer used the Sigmoid activation function to output the aging status score. Model training utilized historical device operating data and their corresponding insulation aging labels, including four states: normal, mild aging, moderate aging, and severe aging, with corresponding score ranges of 0-0.25, 0.26-0.5, 0.51-0.75, and 0.76-1.0, respectively. During training, the cross-entropy loss function was used, with the Adam algorithm optimized, a learning rate of 0.001, a batch size of 64, and 200 training epochs.

[0069] The remaining life prediction is based on aging condition scores and historical degradation rates. By establishing a mapping relationship between aging condition scores and equipment age, a degradation rate model is derived. For transformers, the life prediction formula is to subtract the critical failure state score (0.85) from the current aging condition score, and then divide by the average annual aging rate to obtain the estimated remaining lifespan. The average annual aging rate is obtained through historical data statistics, with a typical value of 0.02-0.05 points / year. Experimental results show that for a transformer that has been in operation for 20 years, its aging condition score is 0.65, the average annual aging rate is 0.03 points / year, and the predicted remaining lifespan is 6.67 years, with an error of 4.7% compared to the actual lifespan of 7 years.

[0070] During online real-time monitoring, newly collected operational monitoring data is processed every 24 hours to update the insulation aging status assessment results and life prediction values. When an abnormal increase in transient impact components is detected, an alarm mechanism is triggered and the data processing frequency is increased to once every 4 hours to promptly detect potential faults.

[0071] In one optional implementation, the step of iteratively adjusting the rotation angle of the basis vectors of the transformation matrix and performing orthogonalization, and adaptively updating the transformation matrix until the residual correlation coefficient drops below a preset decoupling threshold, includes:

[0072] Calculate the deviation between the absolute value of the residual correlation coefficient in the current iteration and the preset decoupling threshold, and determine the adjustment magnitude coefficient of the transformation matrix based on the ratio of the deviation to the preset deviation benchmark value;

[0073] A rotational perturbation is applied to the first and second basis vectors in the current transformation matrix. The angle of the rotational perturbation is determined by the product of the adjustment amplitude coefficient and the preset unit rotation angle. After the rotational perturbation, the first and second basis vectors are subjected to Schmitt orthogonalization to obtain the first and second updated basis vectors. The updated transformation matrix is ​​then recombined.

[0074] The transient shock component and the long-term trend component are reprojected using the updated transformation matrix to obtain the new transient decoupling component and the new trend decoupling component. The new residual correlation coefficient is calculated. The change in correlation coefficient is determined based on the new residual correlation coefficient and the residual correlation coefficient of the previous round. When the change in correlation coefficient is negative, the current preset unit rotation angle is kept unchanged. When the change in correlation coefficient is positive, the preset unit rotation angle is updated to the product of the current value and the preset reduction factor.

[0075] Repeat the iterative process until the absolute value of the residual correlation coefficient drops below the preset decoupling threshold.

[0076] In one specific implementation, raw signal data containing transient impact components and long-term trend components, such as the vibration signal of a manufacturing equipment, is acquired. An initial two-dimensional transformation matrix is ​​constructed, containing two unit orthogonal basis vectors. Initially, the first basis vector can be set to [1, 0], and the second basis vector can be set to [0, 1], which together constitute the initial form of the transformation matrix. A projection transformation is performed on the raw signal data to obtain the initially decoupled transient and trend components, and the residual correlation coefficient between these two components is calculated, for example, an initial residual correlation coefficient of 0.75. Simultaneously, a preset decoupling threshold of 0.05, a preset unit rotation angle of 0.01 radians, a preset deviation baseline value of 0.1, and a preset reduction factor of 0.9 are set.

[0077] Calculate the deviation between the absolute value of the residual correlation coefficient in the current iteration and the preset decoupling threshold. For example, when the residual correlation coefficient is 0.75, the deviation is |0.75| - 0.05 = 0.7. The adjustment range coefficient of the transformation matrix is ​​determined based on the ratio of the deviation to the preset deviation benchmark value, calculated as 0.7 / 0.1 = 7. This means that the adjustment range in the current iteration is relatively large, requiring a larger basis vector rotation angle to accelerate the decoupling process.

[0078] A rotational perturbation is applied to the first and second basis vectors in the current transformation matrix. Specifically, the angle of the rotational perturbation is determined by the product of the adjustment amplitude coefficient and the preset unit rotation angle, i.e., 7 × 0.01 = 0.07 radians. The first basis vector is rotated clockwise by 0.07 radians, and the second basis vector is rotated counterclockwise by 0.07 radians, forming the initially rotated basis vectors. Since the rotation operation may cause the basis vectors to lose orthogonality, the first and second basis vectors are immediately subjected to Schmitt orthogonalization after the rotational perturbation to ensure that they remain strictly orthogonal. The rotated first basis vector remains unchanged, and the projection component of the second basis vector in the direction of the first basis vector is subtracted. The resulting vector is normalized to obtain the second updated basis vector of unit length. After processing, the first updated basis vector [0.9976, 0.0699] and the second updated basis vector [-0.0699, 0.9976] are obtained. These two updated basis vectors are recombined to construct the updated transformation matrix.

[0079] The transient impact component and the long-term trend component are reprojected using an updated transformation matrix to obtain new transient decoupling components and new trend decoupling components. The correlation coefficient between these two new decoupling components is calculated to obtain a new residual correlation coefficient, for example, reduced to 0.65. The change in correlation coefficient is determined based on the new residual correlation coefficient and the residual correlation coefficient of the previous round, calculated as 0.65 - 0.75 = -0.1. Since the change in correlation coefficient is negative, it indicates that the current rotation direction has effectively reduced the residual correlation. Maintaining the current preset unit rotation angle of 0.01 radians, the next round of rotation continues in the current direction.

[0080] In subsequent iterations, if the correlation coefficient changes positively after a certain iteration, for example, from 0.25 to 0.28 (a change of 0.03), this indicates that the current rotation direction or amplitude is no longer effective in reducing the residual correlation. The preset unit rotation angle is then updated to the product of the current value and the preset reduction factor, i.e., 0.01 × 0.9 = 0.009 radians, reducing the rotation step size for finer adjustments.

[0081] The above iterative process is repeated, adaptively adjusting the transformation matrix in each iteration until the absolute value of the residual correlation coefficient drops below the preset decoupling threshold of 0.05. After multiple iterations, for example, after the 32nd iteration, the residual correlation coefficient is 0.048, satisfying the decoupling requirement. At this point, the transformation matrix can effectively separate the transient impulse component and the long-term trend component in the original signal, and the final transformation matrix basis vectors are the first basis vector [0.8660, 0.5000] and the second basis vector [-0.5000, 0.8660]. Using this final transformation matrix to perform a projection transformation on the original signal, mutually independent transient impulse characteristics and long-term trend characteristics can be obtained, providing a foundation for subsequent signal analysis and processing.

[0082] This adaptive iterative approach allows for dynamic adjustment of the transformation matrix update strategy based on changes in residual correlation during decoupling. This ensures the effectiveness of decoupling while improving the algorithm's convergence efficiency, making it particularly suitable for industrial monitoring scenarios requiring real-time processing.

[0083] In one alternative implementation, identifying the cumulative action mode of stress components from multidimensional operational monitoring data and constructing a stress accumulation damage factor includes:

[0084] Temperature stress components, electric field stress components, and mechanical stress components are extracted from multidimensional operation monitoring data. The cumulative stress of each stress component is obtained by time integration over historical periods.

[0085] The cumulative stress of each stress component is constructed into a stress accumulation matrix in chronological order. Singular value decomposition is performed on the stress accumulation matrix to determine the left singular vector matrix and the right singular vector matrix. The dominant stress mode vector corresponding to the maximum singular value is extracted from the left singular vector matrix, and the dominant time evolution vector corresponding to the maximum singular value is extracted from the right singular vector matrix. The dominant stress mode vector and the dominant time evolution vector are reconstructed by the outer product of the dominant stress mode vector and the dominant time evolution vector to obtain the dominant mode component of stress accumulation.

[0086] The stress accumulation trajectory is obtained by reconstructing the phase space of the dominant mode components. The correlation dimension of the stress accumulation trajectory is calculated to determine the complexity factor. The maximum Lyapunov exponent of the stress accumulation trajectory is calculated to determine the chaos factor. The complexity factor and the chaos factor are nonlinearly fused to obtain the stress accumulation damage factor.

[0087] In one specific implementation, various stress components are extracted from multi-dimensional operation monitoring data. Temperature stress components are calculated based on thermodynamic principles by collecting equipment operating temperature data through temperature sensors. Electric field stress components are calculated by collecting electric field distribution data around the equipment through electric field strength sensors and combining dielectric theory. Mechanical stress components are calculated by collecting force data of the equipment through vibration sensors, pressure sensors, etc. Taking a transformer as an example, the temperature stress component can be expressed as the difference between the oil temperature and the ambient temperature multiplied by the temperature influence coefficient. For example, when the oil temperature is 85℃, the ambient temperature is 25℃, and the temperature influence coefficient is 0.08, the temperature stress component is 4.8. The electric field stress component can be expressed as the ratio of the measured electric field strength to the rated electric field strength multiplied by the electric field influence coefficient. For example, when the measured electric field strength is 4.2kV / mm, the rated electric field strength is 3.5kV / mm, and the electric field influence coefficient is 1.2, the electric field stress component is 1.44. The mechanical stress component can be expressed as the ratio of the measured vibration amplitude to the threshold vibration amplitude multiplied by the mechanical influence coefficient. For example, when the measured vibration amplitude is 0.15mm, the threshold vibration amplitude is 0.1mm, and the mechanical influence coefficient is 0.9, the mechanical stress component is 1.35.

[0088] The extracted stress components are integrated over time over a historical period to obtain the cumulative stress. The time integration is achieved using a discrete accumulation method, that is, the stress component values ​​at each time point are accumulated within the sampling time window. With a one-day time window and a sampling interval of one hour, the 24-hour cumulative value of the temperature stress component reaches 108.5, the 24-hour cumulative value of the electric field stress component reaches 32.6, and the 24-hour cumulative value of the mechanical stress component reaches 29.8.

[0089] The cumulative stress of each stress component is used to construct a stress accumulation matrix in chronological order. Rows in the matrix represent different stress components, and columns represent different time points. For example, a 3×30 stress accumulation matrix can be constructed to represent the cumulative values ​​of three stress components over 30 consecutive days. The first row of the matrix stores the cumulative value of temperature stress, the second row stores the cumulative value of electric field stress, and the third row stores the cumulative value of mechanical stress.

[0090] Singular value decomposition (SVD) is performed on the stress accumulation matrix to determine the left and right singular vector matrices. SVD decomposes the original matrix into the product of three matrices: the left singular vector matrix, the diagonal singular value matrix, and the transpose of the right singular vector matrix. In this embodiment, SVD is performed on the 3×30 stress accumulation matrix, resulting in a 3×3 left singular vector matrix, a 3×30 diagonal singular value matrix, and a 30×30 right singular vector matrix. The singular values, arranged from largest to smallest, are 685.2, 142.3, and 38.7.

[0091] Extract the dominant stress mode vector corresponding to the maximum singular value from the left singular vector matrix. In this example, the first column of the left singular vector matrix [0.82, 0.41, 0.39] is the dominant stress mode vector, representing the relative weights of the three stress components during the accumulation process. Extract the dominant time evolution vector corresponding to the maximum singular value from the right singular vector matrix. In this example, the first column of the right singular vector matrix is ​​the dominant time evolution vector, describing the change of stress accumulation over time.

[0092] The dominant mode components of stress accumulation are reconstructed by the outer product of the dominant stress mode vector and the dominant time evolution vector. The outer product operation involves multiplying the two vectors as row and column vectors respectively, resulting in a matrix. The dominant mode components retain the most important structural features of the original stress accumulation matrix while filtering out secondary components.

[0093] The stress accumulation trajectory is obtained by reconstructing the phase space of the dominant mode components. Phase space reconstruction maps a one-dimensional time series to a multi-dimensional space using a time delay method. With an embedding dimension of 3 and a time delay of 2, each row of the dominant mode components is transformed into a trajectory in a three-dimensional phase space. For example, the first row of data [108.5, 112.3, 115.7, 110.2, 105.8, ...] can be transformed into a three-dimensional point sequence [(108.5, 115.7, 105.8), (112.3, 110.2, 103.4), ...].

[0094] The complexity factor is determined by calculating the correlation dimension of the stress accumulation trajectory. The correlation dimension is obtained by calculating the statistical correlation between point pairs in phase space and reflects the complexity of the system. In this example, the correlation dimension of the stress accumulation trajectory is calculated to be 2.43, indicating high complexity. The complexity factor is defined as the ratio of the correlation dimension to the theoretical maximum dimension, which is 0.81.

[0095] The maximum Lyapunov exponent of the stress accumulation trajectory is calculated to determine the chaos factor. The maximum Lyapunov exponent characterizes the sensitivity of the system to initial conditions and is an important indicator of chaos. In this example, the calculated maximum Lyapunov exponent of the stress accumulation trajectory is 0.058, and the chaos factor is defined as the normalized value of the maximum Lyapunov exponent, which is 0.72.

[0096] The stress cumulative damage factor is obtained by nonlinearly fusing the complexity factor and the chaos factor. The nonlinear fusion uses a weighted geometric average method, taking the square root of the product of the complexity factor and the chaos factor, and then multiplying it by an adjustment coefficient. In this example, the adjustment coefficient is set to 1.2, and the calculated stress cumulative damage factor is 0.91. This value is greater than the warning threshold of 0.85, indicating a high risk of cumulative damage to the equipment, requiring timely repair or replacement of relevant components.

[0097] like Figure 2 The flowchart shown illustrates the construction process of the stress accumulation damage factor.

[0098] In one optional implementation, the degradation rate is obtained by extracting the deviation of the aging state from the metric value and performing piecewise fitting, and the modified degradation rate is obtained by nonlinearly modulating the degradation rate based on the stress accumulation damage factor, including:

[0099] Extract the time series of aging state deviation from the metric value, set multiple time segment windows on the time series, and perform segmented fitting within each time segment window to obtain the degradation rate of each time segment window.

[0100] A two-dimensional coordinate space is constructed by constructing the degradation rate corresponding to each time segment window and the stress cumulative damage factor in the same time period. The probability density statistics of historical data points in the two-dimensional coordinate space are used to obtain the joint probability distribution. The value of the stress cumulative damage factor is fixed in the joint probability distribution and the average value of the corresponding degradation rate is calculated to determine the baseline degradation response. The difference between the actual degradation rate and the baseline degradation response is calculated to obtain the degradation anomaly degree.

[0101] The width of the modulation window is determined based on the degree of degradation anomaly, and the modulation window is constructed. The historical data points are then subjected to sliding weighted summation along the direction of the stress accumulation damage factor in the two-dimensional coordinate space to obtain the damage influence distribution. The values ​​at the corresponding positions of each time segment window are extracted from the damage influence distribution as damage influence coefficients. The damage influence coefficients are multiplied by the stress accumulation damage factor to determine the modulation amount. The degradation rate of each time segment window is added to the modulation amount of the corresponding time segment window to obtain the corrected degradation rate.

[0102] In one specific implementation, a time series of aging condition deviation values ​​is extracted. These values ​​are derived from the percentage difference between the equipment's insulation characteristic parameters and their rated or initial values. Taking transformer insulation as an example, the aging condition deviation value can be calculated using parameters such as dielectric loss factor, insulation resistance, and absorption ratio. Specifically, the initial dielectric loss factor of the transformer insulation paperboard is 0.5%, and after ten years of operation, it is measured to be 1.2%, resulting in a calculated aging condition deviation value of 140%. Continuous monitoring data is collected to form a time series of aging condition deviation values. For example, during the 20-year operation of a transformer, insulation parameters are measured quarterly, resulting in a data series of 80 time points.

[0103] Multiple time-segmented windows are set on the time series, and the length of each window can be determined based on the equipment type and operating conditions. For large transformers, a 12-month window can be used; for distribution transformers, a 6-month window can be used; and for cables, an 18-month window can be used. The overlap rate of adjacent time-segmented windows is set to 50% to ensure data continuity and smooth transition. For the transformer example above, a 20-year operating period can be divided into 39 time-segmented windows. Piecewise fitting is performed within each time-segmented window to obtain the degradation rate for that window. The piecewise fitting uses a linear regression method, fitting the deviation of the aging state from the metric value within the window to time; the slope of the fitted line is the degradation rate for that window. For example, the degradation rate calculated for the 15th time-segmented window is 3.2% / year, indicating that the insulation performance deteriorates by 3.2% per year during this period.

[0104] A two-dimensional coordinate space is constructed by linking the degradation rate corresponding to each time segment window with the stress accumulation damage factor for the same time period. The stress accumulation damage factor comprehensively considers the cumulative effects of multiple stress factors, such as temperature, electric field strength, and mechanical vibration. Temperature stress is calculated using a thermal acceleration factor, electric field stress is calculated as the ratio of electric field strength to the rated electric field strength, and mechanical vibration stress is calculated as the ratio of vibration amplitude to a reference value. These three stresses are weighted and summed with weights of 0.6, 0.3, and 0.1, respectively, to obtain the comprehensive stress index. The stress accumulation damage factor is the cumulative value of the comprehensive stress index within the time segment window. For example, the calculated stress accumulation damage factor for the 15th time segment window is 1.35, indicating that the comprehensive stress accumulation effect within this time period is 1.35 times that of the standard operating condition.

[0105] A joint probability distribution was obtained by statistically analyzing the probability density of historical data points in a two-dimensional coordinate space. A kernel density estimation method was used, with the degradation rate as the horizontal axis and the stress cumulative damage factor as the vertical axis, to establish a two-dimensional kernel density function. A Gaussian kernel was chosen as the kernel function, and the bandwidth parameter was set to 0.1. Kernel density estimation was performed on all 39 time-segmented data points to obtain the joint probability distribution of degradation rate and stress cumulative damage factor. The value of the stress cumulative damage factor was fixed in the joint probability distribution, and the average value of the corresponding degradation rate was calculated to determine the baseline degradation response. For example, when the stress cumulative damage factor is 1.0, the baseline degradation response is 2.5% / year; when the stress cumulative damage factor is 1.5, the baseline degradation response is 3.8% / year; and when the stress cumulative damage factor is 2.0, the baseline degradation response is 5.2% / year. The difference between the actual degradation rate and the baseline degradation response was calculated to obtain the degradation anomaly degree. For example, the actual degradation rate of the 15th time segment window is 3.2% / year, the corresponding stress cumulative damage factor is 1.35, and the calculated baseline degradation response is 3.4% / year. Therefore, the degradation anomaly is -0.2% / year, indicating that the actual degradation rate of this window is slightly lower than the expected value.

[0106] The width of the modulation window is determined and constructed based on the degradation anomaly degree. The width of the modulation window is proportional to the degradation anomaly degree, specifically calculated as the base width plus the product of the absolute value of the degradation anomaly degree and a scaling factor. The base width is set to 0.2, and the scaling factor is set to 0.5. For example, if the degradation anomaly degree of the 15th time segment window is -0.2% / year, the calculated modulation window width is 0.2 + 0.2 × 0.5 = 0.3. The modulation window adopts a Gaussian shape, with the current value of the stress accumulation damage factor at its center, and the width being the calculated window width.

[0107] The damage impact distribution is obtained by sliding a weighted summation of historical data points along the direction of the stress cumulative damage factor in a two-dimensional coordinate space using a modulation window. The modulation window slides from the minimum to the maximum value of the stress cumulative damage factor with a step size of 0.05. At each position, the weight of each data point within the window is calculated. The weight value is determined by a Gaussian function, with the highest weight of 1 at the center point and decreasing towards both sides. The degradation rate of each data point is multiplied by its corresponding weight, summed, and divided by the total weight to obtain the weighted average degradation rate at that position, which constitutes the damage impact distribution. For example, when the stress cumulative damage factor is 1.35, the damage impact distribution value is 3.25% / year.

[0108] The values ​​corresponding to each time segment window in the damage impact distribution are extracted as damage impact coefficients. For the 15th time segment window, the stress cumulative damage factor is 1.35, and the corresponding damage impact coefficient is 3.25% / year. The modulation amount is determined by multiplying the damage impact coefficient and the stress cumulative damage factor. For example, the modulation amount for the 15th time segment window is 3.25% / year × 1.35 = 4.39% / year. The modified degradation rate is obtained by adding the degradation rate of each time segment window to the corresponding modulation amount. For the 15th time segment window, the original degradation rate is 3.2% / year, the modulation amount is 4.39% / year, and the modified degradation rate is 7.59% / year.

[0109] Based on the corrected degradation rate, the remaining life of equipment insulation systems can be predicted more accurately. The estimated remaining life is obtained by calculating the difference between the cumulative damage and the critical damage threshold, and dividing by the corrected degradation rate. For example, if a transformer currently has 65% cumulative damage, a critical damage threshold of 90%, and a corrected degradation rate of 7.59% / year in the most recent time window, then the estimated remaining life is approximately (90%-65%) / 7.59% / year ≈ 3.29 years. Practical verification shows that the average error between the predicted and actual remaining life of equipment using this method is within 15%, which is better than the error of over 30% for traditional prediction methods.

[0110] In one optional implementation, using the modified degradation rate as the initial evolution rate, the historical sequence of the stress accumulation damage factor is fitted and extrapolated to recursively calculate the aging state prediction trajectory. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining lifetime prediction value, including:

[0111] The historical sequence of stress cumulative damage factor is extracted. The fitting relationship is determined by linear fitting between the value at each time point in the historical sequence and the values ​​at multiple previous time points. The pre-set prediction period is extrapolated step by step according to the fitting relationship to obtain the extrapolated value of stress cumulative damage factor at each time point in the prediction period.

[0112] The modified degradation rate is used as the initial evolution rate. Multiple time points are set in the prediction period. At each time point, the aging state deviation metric is calculated based on the current rate and the extrapolated value of the stress accumulation damage factor at the corresponding time point. The evolution rate is updated based on the difference between the aging state deviation metric values ​​of adjacent time points. The calculation is continued iteratively. The aging state deviation metric values ​​of each time point are connected to obtain the aging state prediction trajectory.

[0113] The historical sequence of stress accumulation damage factor is segmented and randomly sampled to obtain multiple sets of recombined sequences. The recombined sequences are fitted and extrapolated to obtain multiple sets of extrapolated values. Based on the multiple sets of extrapolated values, multiple aging state prediction trajectories are calculated. The maximum and minimum values ​​of the multiple aging state prediction trajectories are extracted at each time point and connected to form a fluctuation boundary.

[0114] Extract the moment when the preset failure threshold is first exceeded on the aging state prediction trajectory, determine the failure moment, and calculate the time difference between the failure moment and the current moment to determine the remaining life prediction value.

[0115] In one specific implementation, a historical sequence of stress accumulation damage factors is extracted. This stress accumulation damage factor comprehensively reflects the cumulative damage effect of multiple stress factors, such as temperature, electric field strength, and mechanical vibration, on insulating materials. Taking transformer insulation as an example, temperature stress is calculated using the thermal aging index, which doubles for every 6°C increase in temperature; electric field stress is calculated by power of the cube of the ratio of electric field strength to the design electric field strength; and mechanical vibration stress is calculated by the ratio of vibration acceleration to the reference acceleration. Considering all three stress factors, a comprehensive stress index is obtained by weighting them according to a ratio of 0.65:0.25:0.10, and then the stress accumulation damage factor is obtained by accumulating over time. Taking a power transformer as an example, with a 20-year operating history, the stress accumulation damage factor is recorded quarterly, forming a historical sequence of 80 data points with values ​​ranging from 0.95 to 1.85 and an average value of 1.25.

[0116] The fitting relationship was determined by linearly fitting the values ​​at each moment in the historical sequence with the values ​​at multiple preceding moments. The data from the previous 8 moments were selected as input variables, and the data from the current moment was used as the output variable. A multiple linear regression model was used to establish the fitting model. Taking the 60th moment as an example, its stress cumulative damage factor was 1.48. The stress cumulative damage factors for the previous 8 moments were 1.42, 1.45, 1.39, 1.41, 1.44, 1.47, 1.43, and 1.40, respectively. The fitting coefficients obtained by multiple linear regression were 0.15, 0.22, 0.08, 0.12, 0.18, 0.10, 0.09, and 0.06, respectively, with a fitting bias term of 0.05. Based on the fitting relationship, the model was extrapolated stepwise over a preset prediction period to obtain the extrapolated values ​​of the stress cumulative damage factor at each moment in the prediction period. The prediction period was set to the next 5 years, with 20 quarterly data points. For the first prediction time, the extrapolated value is calculated using the data from the last 8 time points in the historical sequence and the fitting coefficient, resulting in 1.52. For the second prediction time, the extrapolated value from the first prediction time is added to the data from the last 7 time points in the historical sequence, and the fitting coefficient is applied to calculate the value, resulting in 1.55. This process is repeated to calculate the extrapolated values ​​of the stress cumulative damage factor for 20 prediction time points, which are 1.52, 1.55, 1.58, 1.60, 1.63, 1.65, 1.68, 1.70, 1.73, 1.75, 1.78, 1.80, 1.83, 1.85, 1.88, 1.90, 1.93, 1.95, 1.98, and 2.00.

[0117] The modified degradation rate is used as the initial evolution rate, derived from the aforementioned degradation rate correction process. For this transformer, the most recent modified degradation rate is 3.8% / year. Multiple time points are set within the prediction period, with a quarterly interval. At each time point, the aging state deviation metric is calculated based on the current rate and the extrapolated value of the stress cumulative damage factor at the corresponding time point. The calculation process for the aging state deviation metric is to add the aging state deviation metric of the previous time point to the product of the current evolution rate and the time interval. Assuming the aging state deviation metric of the current time point is 65%, the evolution rate of the first prediction time point is 3.8% / year, and the time interval is 0.25 years, then the aging state deviation metric of the first prediction time point is 65% + 3.8% / year × 0.25 years = 65.95%. The evolution rate is updated based on the difference in aging state deviation metric values ​​between adjacent time points. The update formula is the current evolution rate multiplied by the ratio of the stress cumulative damage factor at the current time point to the stress cumulative damage factor at the previous time point. For the period from the first prediction time to the second prediction time, the evolution rate is updated to 3.8% / year × (1.55 / 1.52) = 3.87% / year. The iteration continues, and the aging state deviation metric at the second prediction time is 65.95% + 3.87% / year × 0.25 years = 66.92%. Following this method, the aging state deviation metric is calculated progressively for all prediction times, and connecting the aging state deviation metric values ​​at each time point yields the aging state prediction trajectory.

[0118] The historical sequence of stress cumulative damage factor is segmented and randomly sampled to obtain multiple sets of reconstructed sequences. Specifically, 80 historical data points are divided into 10 segments of 8 data points each. Six data points are randomly selected from each segment and arranged in their original order to form a reconstructed sequence containing 60 data points. This process is repeated 20 times to obtain 20 sets of reconstructed sequences. The reconstructed sequences are then fitted and extrapolated to obtain multiple sets of extrapolated values, using the same methods as described above. For example, the extrapolated stress cumulative damage factor value obtained from the 5th reconstructed sequence is 1.79 at the 10th prediction time, while the extrapolated value from the 12th reconstructed sequence at the same time is 1.71. Based on these extrapolated values, multiple aging state prediction trajectories are calculated using the same method as described above. The maximum and minimum values ​​of the multiple aging state prediction trajectories at each time point are extracted and connected to form a fluctuation boundary. For example, at the 10th prediction time, the maximum deviation of the aging state metric for the 20 trajectories is 74.8%, and the minimum is 71.2%, forming the fluctuation boundary at that time.

[0119] The moment when the aging state deviation first exceeds the preset failure threshold is extracted from the aging state prediction trajectory to determine the failure time. For transformer insulation, the preset failure threshold is typically 85%, meaning that the insulation is considered to have failed when the aging state deviates from the metric value by 85%. On the main prediction trajectory, the aging state deviation reaches 85.2% at the 18th prediction time, exceeding the failure threshold. Therefore, the failure time is determined to be the 18th prediction time, corresponding to 4.5 years after the current time point. The time difference between the failure time and the current time is calculated to determine the remaining lifetime prediction value, which is 4.5 years. Simultaneously, considering the influence of fluctuation boundaries, in the most unfavorable case (maximum value trajectory), the failure time is the 16th prediction time, corresponding to a remaining lifetime prediction value of 4 years; in the most favorable case (minimum value trajectory), the failure time is the 20th prediction time, corresponding to a remaining lifetime prediction value of 5 years. Therefore, the remaining lifetime prediction range for this transformer insulation system is 4-5 years, with a median of 4.5 years.

[0120] By extrapolating the stress accumulation damage factor and using multi-trajectory simulation, the uncertainties and variability in the insulation aging process are effectively considered, resulting in more reliable prediction results.

[0121] In one optional implementation, the historical sequence of stress accumulation damage factor is segmented and randomly extracted to obtain multiple sets of recombined sequences, including:

[0122] Calculate the autocorrelation coefficient of the historical sequence, extract the lag step length when the autocorrelation coefficient first falls below the preset correlation threshold, and use the lag step length as the block length.

[0123] The historical sequence of stress accumulation damage factor is continuously divided according to the block length to obtain multiple historical blocks;

[0124] Randomly extract historical blocks with replacement, and extract a preset number of historical blocks each time. Then connect them in chronological order to form a recombined sequence.

[0125] Repeat the process until the preset number of times is reached, resulting in multiple sets of recombination sequences.

[0126] In one specific implementation, the autocorrelation coefficient of the historical sequence is calculated. The autocorrelation coefficient reflects the degree of correlation between the time series and itself at different lag steps. For the historical sequence of the stress accumulation damage factor, its autocorrelation reveals the time-dependent characteristics of stress changes. When calculating the autocorrelation coefficient, correlation analysis is performed between the historical sequence and sequences with different lag steps. For the original historical sequence and the sequence with lags of k steps, the covariance of the two is calculated and divided by the variance of the original sequence to obtain the autocorrelation coefficient with lags of k steps. Taking the insulation of a transformer as an example, its historical sequence of the stress accumulation damage factor contains 80 quarterly data points. The calculated autocorrelation coefficients were 0.92 for lag 1, 0.85 for lag 2, 0.76 for lag 3, 0.68 for lag 4, 0.58 for lag 5, 0.47 for lag 6, 0.36 for lag 7, 0.25 for lag 8, 0.14 for lag 9, 0.06 for lag 10, and -0.03 for lag 11. The autocorrelation coefficients gradually decreased with increasing lag length, indicating a significant time dependence of the sequence, but this dependence weakened with increasing time intervals.

[0127] The lag step length at which the autocorrelation coefficient first falls below a preset correlation threshold is extracted and used as the block length. The preset correlation threshold is typically set to 0.2, meaning that when the autocorrelation coefficient is below 0.2, the correlation of the sequence at the corresponding lag step is negligible, and the data points are essentially independent. In the transformer example above, the autocorrelation coefficient is 0.14 at a lag of 9 steps, falling below the preset correlation threshold of 0.2 for the first time, thus determining the block length to be 9. This means that the stress cumulative damage factor data at 9 quarters intervals can be considered mutually independent. Determining the block length is the foundation for historical sequence reconstruction. A reasonable block length can preserve the time dependence structure within the sequence while ensuring the relative independence between different blocks, providing a scientific basis for subsequent random sampling. For different types of power equipment, the autocorrelation characteristics of their stress cumulative damage factor sequences may differ, and the block length will vary accordingly. For example, the autocorrelation decay of the historical stress cumulative damage factor sequence for power cables is relatively fast, and the block length may be 6-7; while the autocorrelation decay of the historical stress cumulative damage factor sequence for large transformers is relatively slow, and the block length may be 10-12.

[0128] The historical sequence of stress cumulative damage factors is continuously divided according to the block length to obtain multiple historical blocks. Starting from the beginning of the historical sequence, it is divided once every block length to form multiple historical blocks of equal length. For a transformer historical sequence containing 80 data points and a block length of 9, it can be divided into 8 complete blocks, plus the last incomplete block containing 8 data points, for a total of 9 historical blocks. The first historical block contains data points 1 to 9, with values ​​of 1.05, 1.08, 1.10, 1.12, 1.15, 1.18, 1.20, 1.22, and 1.25; the second historical block contains data points 10 to 18, with values ​​of 1.28, 1.30, 1.32, 1.35, 1.37, 1.40, 1.42, 1.45, and 1.47; and so on, until the ninth historical block contains data points 73 to 80, with values ​​of 1.68, 1.70, 1.73, 1.75, 1.78, 1.80, 1.82, and 1.85. Each historical block retains the temporal evolution characteristics of the stress accumulation damage factor, while different blocks are relatively independent and can be randomly sampled and recombined.

[0129] Multiple historical blocks are randomly sampled with replacement, with a preset number of blocks selected each time. These blocks are then concatenated in chronological order to form a reconstructed sequence. The preset number of blocks to be sampled is usually set to be equal to the original number of historical blocks; in this example, it is 9. The random sampling process uses a random number generator to ensure that each historical block has an equal probability of being selected. Assuming that a random sampling results in historical blocks numbered 3, 1, 5, 2, 7, 4, 9, 6, and 8, these blocks are concatenated in sequence to form a reconstructed sequence containing 81 data points. Specifically, the first nine data points of the recombined sequence come from the third historical block, with values ​​of 1.50, 1.52, 1.55, 1.57, 1.60, 1.62, 1.65, 1.67, and 1.70; the next nine data points come from the first historical block, with values ​​of 1.05, 1.08, 1.10, 1.12, 1.15, 1.18, 1.20, 1.22, and 1.25; and so on, until the last nine data points come from the eighth historical block, with values ​​of 1.58, 1.60, 1.63, 1.65, 1.67, 1.70, 1.72, 1.75, and 1.77. In this way, the recombined sequence retains the local temporal structure of the original historical sequence while introducing global randomness, enabling the simulation of different stress evolution scenarios.

[0130] The random sampling and sequence recombination process described above is repeated until a preset number of repetitions is reached, resulting in multiple recombined sequences. The preset number of repetitions is typically set to 20-30 times to ensure sufficiently diverse recombined sequences are generated. In this embodiment, the process is repeated 25 times, generating 25 recombined sequences. These recombined sequences, while preserving the local temporal structure of the original historical sequences, exhibit different global evolution trends, effectively simulating potential changes in the future stress accumulation damage factor. For example, in the 10th recombined sequence, the stress accumulation damage factor generally shows an upward trend, but experiences a brief decrease in the middle; the 18th recombined sequence exhibits a phased, fluctuating upward trend; and the 22nd recombined sequence is relatively stable in the first half and rises rapidly in the second half. These diverse recombined sequences provide rich stress scenario inputs for subsequent aging state prediction.

[0131] By fitting and extrapolating multiple sets of recombined sequences, extrapolated values ​​of multiple stress cumulative damage factors can be obtained. For each extrapolated value, the corresponding aging state prediction trajectory is calculated, thus obtaining a series of aging evolution scenarios. The distribution characteristics of these prediction trajectories reflect the uncertainty of the prediction results, providing a more comprehensive reference for equipment lifespan decisions.

[0132] The system for assessing the insulation aging status and predicting the lifespan of power transmission network equipment according to embodiments of the present invention includes:

[0133] The first unit is used to acquire multi-dimensional operation monitoring data of power transmission network equipment;

[0134] The second unit is used to perform multi-scale time-frequency decomposition on multi-dimensional operation monitoring data, extract transient impact components and long-term trend components from the decomposed scale components, and project them to the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector.

[0135] The third unit is used to embed the decoupled feature vector into the high-dimensional state space and calculate the Mahalanobis distance between the current state point and the historical healthy state cluster to obtain the aging state deviation metric.

[0136] The fourth unit is used to identify the cumulative action mode of stress components from multidimensional operation monitoring data and construct a stress cumulative damage factor, extract the deviation value of aging state and perform piecewise fitting to obtain the degradation rate, and obtain the corrected degradation rate by nonlinearly modulating the degradation rate based on the stress cumulative damage factor.

[0137] The fifth unit is used to fit and extrapolate the historical sequence of stress accumulation damage factor with the modified degradation rate as the initial evolution rate, recursively calculate the aging state prediction trajectory, calculate the fluctuation boundary by segmented random sampling, and extract the failure time to determine the remaining life prediction value.

[0138] A third aspect of the present invention provides an electronic device, comprising:

[0139] processor;

[0140] Memory used to store processor-executable instructions;

[0141] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0142] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0143] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the aging condition of insulation and predicting the lifespan of power transmission network equipment, characterized in that, include: Acquire multi-dimensional operational monitoring data of power transmission network equipment; Multi-scale time-frequency decomposition is performed on multi-dimensional operation monitoring data. Transient impact components and long-term trend components are extracted from each scale component after decomposition. The components are then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector. The decoupled feature vectors are embedded into a high-dimensional state space, and the Mahalanobis distance between the current state point and the historical healthy state cluster is calculated to obtain the aging state deviation metric. Identify the cumulative action modes of stress components from multidimensional operational monitoring data and construct a stress accumulation damage factor, including: Temperature stress components, electric field stress components, and mechanical stress components are extracted from multidimensional operation monitoring data. The cumulative stress of each stress component is obtained by time integration over historical periods. The cumulative stress of each stress component is constructed into a stress accumulation matrix in chronological order. Singular value decomposition is performed on the stress accumulation matrix to determine the left singular vector matrix and the right singular vector matrix. The dominant stress mode vector corresponding to the maximum singular value is extracted from the left singular vector matrix, and the dominant time evolution vector corresponding to the maximum singular value is extracted from the right singular vector matrix. The dominant stress mode vector and the dominant time evolution vector are reconstructed by the outer product of the dominant stress mode vector and the dominant time evolution vector to obtain the dominant mode component of stress accumulation. The stress accumulation trajectory is obtained by reconstructing the phase space of the dominant mode components. The correlation dimension of the stress accumulation trajectory is calculated to determine the complexity factor. The maximum Lyapunov exponent of the stress accumulation trajectory is calculated to determine the chaos factor. The complexity factor and the chaos factor are nonlinearly fused to obtain the stress accumulation damage factor. The degradation rate is obtained by extracting the deviation of the aging state from the metric value and performing piecewise fitting. The degradation rate is then nonlinearly modulated based on the stress accumulation damage factor to obtain the corrected degradation rate. Using the modified degradation rate as the initial evolution rate, the historical sequence of stress accumulation damage factor is fitted and extrapolated to obtain the aging state prediction trajectory through recursive calculation. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining life prediction value.

2. The method according to claim 1, characterized in that, Multi-scale time-frequency decomposition is performed on the multi-dimensional operation monitoring data. Transient impact components and long-term trend components are extracted from each scale component after decomposition. These components are then projected onto the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector, which includes: Multiple scale levels are set for multidimensional operation monitoring data. Time-frequency decomposition is performed at each scale level to obtain the corresponding scale components. The time-domain change rate of each scale component is calculated. Scale components with time-domain change rates exceeding a preset change rate threshold are extracted as transient impact components, and scale components with time-domain change rates below a preset change rate threshold are extracted as long-term trend components. Calculate the cross-correlation matrix between the transient shock component and the long-term trend component, perform eigenvalue decomposition on the cross-correlation matrix to obtain a set of eigenvectors, select the eigenvector corresponding to the largest eigenvalue to determine the first basis vector, select the eigenvector corresponding to the smallest eigenvalue to determine the second basis vector, and combine the first basis vector and the second basis vector to construct the transformation matrix. The transient impact component is projected onto the first coordinate axis of the decoupled feature space through a transformation matrix to obtain the transient decoupled component. The long-term trend component is projected onto the second coordinate axis through a transformation matrix to obtain the trend decoupled component. The rotation angle of the basis vector of the transformation matrix is ​​iteratively adjusted and orthogonalized. The transformation matrix is ​​adaptively updated until the residual correlation coefficient drops below the preset decoupled threshold. The transient decoupled component and the trend decoupled component are combined to form the decoupled feature vector.

3. The method according to claim 2, characterized in that, The step of iteratively adjusting the rotation angle of the basis vectors of the transformation matrix and performing orthogonalization, and adaptively updating the transformation matrix until the residual correlation coefficient drops below a preset decoupling threshold, includes: Calculate the deviation between the absolute value of the residual correlation coefficient in the current iteration and the preset decoupling threshold, and determine the adjustment magnitude coefficient of the transformation matrix based on the ratio of the deviation to the preset deviation benchmark value; A rotational perturbation is applied to the first and second basis vectors in the current transformation matrix. The angle of the rotational perturbation is determined by the product of the adjustment amplitude coefficient and the preset unit rotation angle. After the rotational perturbation, the first and second basis vectors are subjected to Schmitt orthogonalization to obtain the first and second updated basis vectors. The updated transformation matrix is ​​then recombined. The transient shock component and the long-term trend component are reprojected using the updated transformation matrix to obtain the new transient decoupling component and the new trend decoupling component. The new residual correlation coefficient is calculated. The change in correlation coefficient is determined based on the new residual correlation coefficient and the residual correlation coefficient of the previous round. When the change in correlation coefficient is negative, the current preset unit rotation angle is kept unchanged. When the change in correlation coefficient is positive, the preset unit rotation angle is updated to the product of the current value and the preset reduction factor. Repeat the iterative process until the absolute value of the residual correlation coefficient drops below the preset decoupling threshold.

4. The method according to claim 1, characterized in that, The degradation rate is obtained by extracting the deviation of the aging state from the metric value and performing piecewise fitting. The modified degradation rate is then obtained by nonlinearly modulating the degradation rate based on the stress accumulation damage factor, including: Extract the time series of aging state deviation from the metric value, set multiple time segment windows on the time series, and perform segmented fitting within each time segment window to obtain the degradation rate of each time segment window. A two-dimensional coordinate space is constructed by constructing the degradation rate corresponding to each time segment window and the stress cumulative damage factor in the same time period. The probability density statistics of historical data points in the two-dimensional coordinate space are used to obtain the joint probability distribution. The value of the stress cumulative damage factor is fixed in the joint probability distribution and the average value of the corresponding degradation rate is calculated to determine the baseline degradation response. The difference between the actual degradation rate and the baseline degradation response is calculated to obtain the degradation anomaly degree. The width of the modulation window is determined based on the degree of degradation anomaly, and the modulation window is constructed. The historical data points are then subjected to sliding weighted summation along the direction of the stress accumulation damage factor in the two-dimensional coordinate space to obtain the damage influence distribution. The values ​​at the corresponding positions of each time segment window are extracted from the damage influence distribution as damage influence coefficients. The damage influence coefficients are multiplied by the stress accumulation damage factor to determine the modulation amount. The degradation rate of each time segment window is added to the modulation amount of the corresponding time segment window to obtain the corrected degradation rate.

5. The method according to claim 1, characterized in that, Using the modified degradation rate as the initial evolution rate, the historical sequence of stress-accumulated damage factor is fitted and extrapolated to recursively calculate the aging state prediction trajectory. The fluctuation boundary is calculated by segmented random sampling, and the failure time is extracted to determine the remaining lifetime prediction value, including: The historical sequence of stress cumulative damage factor is extracted. The fitting relationship is determined by linear fitting between the value at each time point in the historical sequence and the values ​​at multiple previous time points. The pre-set prediction period is extrapolated step by step according to the fitting relationship to obtain the extrapolated value of stress cumulative damage factor at each time point in the prediction period. The modified degradation rate is used as the initial evolution rate. Multiple time points are set in the prediction period. At each time point, the aging state deviation metric is calculated based on the current rate and the extrapolated value of the stress accumulation damage factor at the corresponding time point. The evolution rate is updated based on the difference between the aging state deviation metric values ​​of adjacent time points. The calculation is continued iteratively. The aging state deviation metric values ​​of each time point are connected to obtain the aging state prediction trajectory. The historical sequence of stress accumulation damage factor is segmented and randomly sampled to obtain multiple sets of recombined sequences. The recombined sequences are fitted and extrapolated to obtain multiple sets of extrapolated values. Based on the multiple sets of extrapolated values, multiple aging state prediction trajectories are calculated. The maximum and minimum values ​​of the multiple aging state prediction trajectories are extracted at each time point and connected to form a fluctuation boundary. Extract the moment when the preset failure threshold is first exceeded on the aging state prediction trajectory, determine the failure moment, and calculate the time difference between the failure moment and the current moment to determine the remaining life prediction value.

6. The method according to claim 5, characterized in that, Segmented random sampling of the historical sequence of stress cumulative damage factor yielded multiple sets of recombinant sequences, including: Calculate the autocorrelation coefficient of the historical sequence, extract the lag step length when the autocorrelation coefficient first falls below the preset correlation threshold, and use the lag step length as the block length. The historical sequence of stress accumulation damage factor is continuously divided according to the block length to obtain multiple historical blocks; Randomly extract historical blocks with replacement, and extract a preset number of historical blocks each time. Then connect them in chronological order to form a recombined sequence. Repeat the process until the preset number of times is reached, resulting in multiple sets of recombination sequences.

7. A system for assessing the insulation aging status and predicting the lifespan of power transmission network equipment, used to implement the method described in any one of claims 1-6, characterized in that, include: The first unit is used to acquire multi-dimensional operation monitoring data of power transmission network equipment; The second unit is used to perform multi-scale time-frequency decomposition on multi-dimensional operation monitoring data, extract transient impact components and long-term trend components from the decomposed scale components, and project them to the decoupled feature space through orthogonal transformation to obtain the decoupled feature vector. The third unit is used to embed the decoupled feature vector into the high-dimensional state space and calculate the Mahalanobis distance between the current state point and the historical healthy state cluster to obtain the aging state deviation metric. The fourth unit is used to identify the cumulative action mode of stress components from multidimensional operation monitoring data and construct a stress cumulative damage factor, extract the deviation value of aging state and perform piecewise fitting to obtain the degradation rate, and obtain the corrected degradation rate by nonlinearly modulating the degradation rate based on the stress cumulative damage factor. The fifth unit is used to fit and extrapolate the historical sequence of stress accumulation damage factor with the modified degradation rate as the initial evolution rate, recursively calculate the aging state prediction trajectory, calculate the fluctuation boundary by segmented random sampling, and extract the failure time to determine the remaining life prediction value.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images