A method for diagnosing defects of secondary equipment in intelligent substation

Abstract

The present invention relates to the field of power system equipment monitoring and diagnosis, and in particular to a method for diagnosing defects in secondary equipment of intelligent substations. The content includes: real-time collection of multi-source signals of the operating status of substation equipment and filtering processing to obtain real-time monitoring signals of intelligent substations, reverse rearrangement and segmentation, and then constructing an inverse dynamic relationship matrix; mapping the inverse dynamic relationship matrix to a high-dimensional nonlinear feature space, and dynamically optimizing it, dynamically fusing the final eigenvalues ​​of the obtained high-dimensional nonlinear feature vector to obtain a final fused feature vector, and then obtaining a defect classification result; performing trend prediction based on the final fused feature vector and the defect classification result. It solves the problem that traditional equipment monitoring methods are insufficient in dynamic feature extraction capabilities, insufficient in adaptability to nonlinear complex patterns, lack of effective connection between defect classification results and operating status trend prediction, and affect the overall effect.

Description

Technical Field

[0001] The present invention relates to the field of power system equipment monitoring and diagnosis, and in particular to a method for diagnosing defects in secondary equipment of an intelligent substation. Background Art

[0002] With the rapid development of smart grid technology, substations, as important nodes in the operation of power grids, have continuously improved their automation and intelligence levels. Secondary equipment in substations (including protection devices, measurement and control devices, communication equipment, etc.) undertake core functions such as monitoring, controlling and protecting the operation status of the power grid. The reliability and safety of their operation are directly related to the stability of the power system and the efficient transmission of electric energy. In recent years, the number and types of secondary equipment in smart substations have increased rapidly, their operating environment has become increasingly complex, and the workload of the equipment has gradually increased. The potential defects and operating hazards of secondary equipment have posed new challenges to the safe operation of the power grid.

[0003] Traditional equipment monitoring systems often only perform simple threshold alarms or rule analysis on a single variable, which is difficult to meet the modern power grid's comprehensive analysis needs for multi-dimensional status data. Smart substation secondary equipment not only involves electrical quantities such as voltage and current, but also environmental variables such as temperature and humidity. These multi-source data have complex time dependencies and dynamic characteristics in the process of equipment defect formation. How to effectively mine the correlation information in multi-source signals and dynamically capture the changing characteristics of equipment operating status has become an important technical requirement for smart substation equipment management.

[0004] Therefore, the traditional equipment monitoring methods have the following technical problems: insufficient ability to extract dynamic features of equipment operation signals, lack of deep modeling of signal time dependence and dynamic correlation, and inability to effectively characterize the intrinsic connection between signal segments; insufficient adaptability to nonlinear complex patterns in feature extraction, and difficulty in coping with the diverse defect modes of equipment under complex operating conditions, resulting in insufficient classification ability and accuracy of defect diagnosis; obvious shortcomings in the combination of equipment defect classification and future operating status trend prediction, lack of effective connection between defect classification results and operating status trend prediction, resulting in poor correlation between defect diagnosis and operating status trend prediction results, affecting the overall effect. Summary of the invention

[0005] The present invention provides a method for diagnosing defects in secondary equipment of an intelligent substation, so as to solve the problems that the traditional equipment monitoring method has insufficient dynamic feature extraction capability for equipment operation signals, lacks deep modeling of signal time dependency and dynamic correlation, and cannot effectively characterize the intrinsic connection between signal segments; exhibits insufficient adaptability to nonlinear complex patterns in feature extraction, and is difficult to cope with diversified defect patterns under complex equipment operation states, resulting in insufficient classification capability and accuracy of defect diagnosis; has obvious shortcomings in combining equipment defect classification with future operation status trend prediction, lacks effective connection between defect classification results and operation status trend prediction, resulting in poor correlation between defect diagnosis and operation status trend prediction results, affecting the overall effect.

[0006] A method for diagnosing defects in secondary equipment of a smart substation of the present invention specifically includes the following technical solutions:

[0007] A method for diagnosing defects in secondary equipment of a smart substation comprises the following steps:

[0008] S1. Real-time data collection and filtering of multi-source signals of substation equipment operation status are performed to obtain real-time monitoring signals of smart substations; reverse rearrangement and segmentation of real-time monitoring signals of smart substations are performed to obtain reverse segmented signals; based on the reverse segmented signals, reverse dynamic matrices and reverse dynamic relationship matrices are constructed;

[0009] S2. Map the inverse dynamic relationship matrix to a high-dimensional nonlinear feature space to obtain a high-dimensional nonlinear feature vector; introduce a nested recursive update mechanism to dynamically optimize the high-dimensional nonlinear feature vector to obtain the final eigenvalue of the high-dimensional nonlinear feature vector, and perform dynamic fusion to obtain the final fused feature vector; based on the final fused feature vector, obtain the defect classification result; based on the final fused feature vector and the defect classification result, perform trend prediction to obtain the trend prediction result.

[0010] Preferably, the S1 specifically includes:

[0011] Through the reverse segmentation processing technology, the real-time monitoring signal of the smart substation is reversely rearranged and segmented according to the time interval to obtain a reverse segmentation signal.

[0012] Preferably, the S1 specifically includes:

[0013] By calculating the dot product integral of the inverse segmented signals in different time periods in the time dimension, the correlation strength of the inverse segmented signals is obtained and the inverse dynamic matrix is ​​constructed.

[0014] Preferably, the S1 specifically includes:

[0015] Based on the inverse dynamic matrix, the correlation between the inverse segmented signals is mapped through regularization processing and inverse analysis to obtain the inverse dynamic relationship matrix.

[0016] Preferably, the S2 specifically includes:

[0017] Through the projection operator, the inverse dynamic relationship matrix is ​​transformed into a high-dimensional nonlinear feature vector based on cosine transform and nonlinear activation.

[0018] Preferably, the S2 specifically includes:

[0019] The nested recursive update mechanism introduces a feature autoregressive weight matrix and a feature cross weight matrix, integrates temporal dynamics and feature interaction dependencies, and dynamically optimizes high-dimensional nonlinear feature vectors. The recursive update formula is as follows:

[0020] ,

[0021] in, It is The high-dimensional nonlinear feature vector is The value after recursion; is a nonlinear activation function; It is The high-dimensional nonlinear feature vector is The value after recursion; is the feature autoregressive weight matrix; is the cross-weight matrix between features; It is The high-dimensional nonlinear feature vector is The value after recursion;

[0022] When the preset convergence condition is met, the recursion is stopped and the final eigenvalue of the high-dimensional nonlinear eigenvector is obtained.

[0023] Preferably, the S2 specifically includes:

[0024] Based on the hierarchical contribution of the final eigenvalue of each high-dimensional nonlinear eigenvector, dynamic fusion weights are assigned, and an inverse dynamic relationship matrix is ​​introduced to perform dynamic fusion to generate the final fused eigenvector.

[0025] Preferably, the S2 specifically includes:

[0026] The final fused feature vector is input into the defect pattern classifier to obtain the defect classification result.

[0027] Preferably, the S2 specifically includes:

[0028] Based on the final fusion feature vector and defect classification results, the initial prediction value is obtained; the error feedback and recursive optimization mechanism are introduced, and the prediction state vector at the previous moment is combined to optimize the prediction state vector at the future moment to obtain the trend prediction result; the trend prediction result is dynamically adjusted through the following recursive formula:

[0029] ,

[0030] ,

[0031] ,

[0032] in, is the initial prediction value, indicating The predicted state vector at time t; The future The predicted state vector after time steps; The future The predicted state vector after time steps; is the multi-objective error function; and They are the final fused feature vectors And defect classification results The mapping weight matrix of is the learning rate, which is used to control the magnitude of each step in the gradient descent process; It's time The actual observed value of is the weight of historical forecast errors; is the weight of the dynamic feature consistency; It is the inverse dynamic relationship matrix.

[0033] The beneficial effects of the technical solution of the present invention are:

[0034] 1. The present invention can deeply explore the temporal dependency and correlation of the operating status of substation equipment through the inverse segmentation processing technology and the inverse dynamic matrix construction method. The inverse dynamic relationship matrix overcomes the instability problem caused by the high correlation between the inverse segmentation signals through regularization processing, and significantly improves the accuracy of feature extraction; through high-dimensional nonlinear feature projection based on cosine transform and nonlinear activation, the operating status of the substation equipment is mapped to the high-dimensional nonlinear feature space, so that complex dynamic relationships can be more effectively distinguished and expressed, and the adaptability to diversified defect modes is enhanced.

[0035] 2. The present invention introduces a nested recursive update mechanism to dynamically optimize high-dimensional nonlinear feature vectors, which not only includes time dynamics, but also integrates the interaction between different high-dimensional nonlinear feature vectors, thereby generating globally optimized high-dimensional nonlinear features, further improving the accuracy of defect classification; dynamic fusion combines the final eigenvalues ​​of the high-dimensional nonlinear feature vector with the inverse dynamic relationship matrix to generate a final fused feature vector with high discrimination for the defect mode of substation equipment; the classifier of the defect model performs defect classification based on the final fused feature vector, and at the same time, the defect classification results and the final fused feature vector are used to collaboratively complete the dynamic prediction of the operating status of the substation equipment at future times, ensuring that the defect classification results and the trend prediction results are closely linked.

[0036] 3. The present invention introduces error feedback and recursive optimization mechanisms to dynamically adjust the predicted state vector at future moments. The multi-objective error function not only weighs the deviation between the predicted state vector and the actual observed value, but also ensures the consistency of the dynamic feature expression of the substation equipment, further improving the stability of the trend prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a method for diagnosing defects in secondary equipment of a smart substation according to the present invention. DETAILED DESCRIPTION

[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0040] The specific scheme of the intelligent substation secondary equipment defect diagnosis method provided by the present invention is described in detail below with reference to the accompanying drawings.

[0041] Refer to the attached Figure 1 , which shows a flow chart of a method for diagnosing defects in secondary equipment of a smart substation provided by an embodiment of the present invention, the method comprising the following steps:

[0042] S1. Real-time data collection and filtering of multi-source signals of substation equipment operation status are performed to obtain real-time monitoring signals of smart substations; reverse rearrangement and segmentation of real-time monitoring signals of smart substations are performed to obtain reverse segmented signals; based on the reverse segmented signals, reverse dynamic matrices and reverse dynamic relationship matrices are constructed;

[0043] In order to realize the defect diagnosis of secondary equipment in smart substations, firstly, multi-source signals of the operating status of substation equipment are collected in real time through multiple channels. The multi-source signals include electrical quantities (such as voltage, current, frequency) and environmental status quantities of substation equipment (such as temperature, humidity, etc.). In order to reduce the impact of sampling noise on subsequent processing, the multi-source signals are filtered through filters to obtain the filtered multi-source signals; the filter adopts the Butterworth low-pass filtering method to ensure that the main frequency components of the multi-source signals are captured while suppressing high-frequency noise.

[0044] The multi-source signals after filtering are recorded in the form of time series to obtain the real-time monitoring signal of the smart substation. The sampling interval is determined by the dynamic characteristics of the substation equipment, generally within the millisecond range, and is optimized according to the dynamic response rate of the substation equipment.

[0045] Real-time monitoring signal through smart substation , using reverse segment processing technology to reversely rearrange the real-time monitoring signal of the smart substation into , and segment according to the time interval to generate Reverse segment signal , in order to capture the characteristics of the historical operating status of substation equipment and enhance the ability to describe the historical operating status of substation equipment. The reverse segmented signal is represented by the weighted integral in its time interval, and the formula is as follows:

[0046] ,

[0047] in, It is A reverse segment signal, indicating that in the time interval The weighted cumulative value of the real-time monitoring signals of the smart substations within; and It is the starting and ending point of the time segment; It is the real-time monitoring signal of the smart substation at the time point The value of is the time integral; is the time attenuation coefficient, which is used to adjust the contribution of the monitoring signals of the smart substation at different time points to the signal accumulation value. Each real-time monitoring signal of the smart substation is weighted to ensure that the real-time monitoring signal of the smart substation in the most recent time period contributes more to the reverse segmented signal, while the real-time monitoring signal in the more distant time period has a controlling effect on the smoothness of the overall trend.

[0048] After the real-time monitoring signal of the smart substation is processed in reverse segmentation, in order to further extract the dynamic correlation characteristics between the reverse segmented signals in different time periods, it is necessary to construct an inverse dynamic matrix , to systematically describe the correlation between the reverse segmented signal segments; each element of the reverse dynamic matrix Reflects the The reverse segment signal and the The correlation strength of the reverse segmented signals is obtained by calculating the dot product integral of the two reverse segmented signals in the time dimension; the value of each element in the reverse dynamic matrix depends not only on the amplitude change of the two reverse segmented signals, but also on their synchronization and similarity in time. The specific formula is:

[0049] ,

[0050] in, and They are The reverse segment signal and the The inverse dynamic matrix uses the dot product integral between the inverse segmented signals to represent the correlation strength, revealing the intrinsic dynamic relationship between different inverse segmented signals. Through the construction of the inverse dynamic matrix, the mutual influence characteristics of the inverse segmented signals in each time period are captured. The calculation of the inverse dynamic matrix elements not only reflects the characteristics of a single inverse segmented signal, but also structurally represents the correlation of different inverse segmented signals through the dot product integral, which is convenient for subsequent processing.

[0051] After the inverse dynamic matrix is ​​constructed, in order to further explore the causal relationship between the inverse segmented signals and enhance the stability of the inverse dynamic matrix, it is necessary to construct the inverse dynamic relationship matrix. Through regularization processing and inverse analysis, the correlation between the inverse segmented signals is mapped into a reversible matrix that can characterize the dynamic characteristics of the inverse segmented signals. At the same time, it effectively suppresses the numerical instability problem that may be caused by the high correlation of the inverse segmented signals, and enhances the stability of the inverse dynamic matrix and the ability to analyze the inverse dependency relationship. The formula for constructing the inverse dynamic relationship matrix is ​​as follows:

[0052] ,

[0053] in, It is the inverse dynamic relationship matrix; is the identity matrix; is the transpose of the inverse dynamic matrix; It is a reversible matrix, which is used to avoid the problem of numerical instability caused by the strong correlation of the reverse segmented signals. Each element of the reverse dynamic relationship matrix reflects the reverse dependency relationship between the reverse segmented signals, further deepening the causal relationship expression of the reverse dynamic matrix.

[0054] S2. Map the inverse dynamic relationship matrix to a high-dimensional nonlinear feature space to obtain a high-dimensional nonlinear feature vector; introduce a nested recursive update mechanism to dynamically optimize the high-dimensional nonlinear feature vector to obtain the final eigenvalue of the high-dimensional nonlinear feature vector, and perform dynamic fusion to obtain the final fused feature vector; based on the final fused feature vector, obtain the defect classification result; based on the final fused feature vector and the defect classification result, perform trend prediction to obtain the trend prediction result.

[0055] In order to further enhance the expressiveness and distinguishing ability of the inverse dynamic relationship features, the inverse dynamic relationship matrix is ​​mapped to a high-dimensional nonlinear feature space. Through the projection operator, the inverse dynamic relationship matrix is ​​converted into a high-dimensional nonlinear feature vector, so as to extract the discriminative nonlinear features from the dynamic relationship of the inverse segmented signal to support the subsequent recursive calculation and the final defect classification. The projection operator adopts a combination of cosine transform and nonlinear activation to convert each row of the inverse dynamic relationship feature of the inverse dynamic relationship matrix into a high-dimensional nonlinear feature vector. The formula is as follows:

[0056] ,

[0057] in, It is A high-dimensional nonlinear feature vector represents the mapping result of the inverse dynamic relationship matrix in the high-dimensional nonlinear feature space; is the inverse dynamic relationship matrix OK; is the weight of the projection operator, indicating the The importance of the projection direction; It is A projection direction vector, indicating the direction of the feature map; is the number of projection directions, i.e., the dimension of the high-dimensional nonlinear feature space; It is a nonlinear activation function used to enhance the nonlinear performance of features. By performing multi-directional projection on the inverse dynamic relationship matrix and combining cosine transform and weighted summation, the low-dimensional inverse dynamic relationship features are mapped into high-dimensional nonlinear feature vectors, which have higher classification capabilities.

[0058] After feature projection, the high-dimensional nonlinear feature vector is further optimized, and a nested recursive update mechanism is introduced to capture the intrinsic dependencies between each high-dimensional nonlinear feature vector and form a deep expression of the operating status of the substation equipment. By dynamically optimizing each high-dimensional nonlinear feature vector, it not only reflects the dynamic changes of its own characteristics, but also integrates the interactive effects from other high-dimensional nonlinear feature vectors, thereby generating a globally optimized high-dimensional nonlinear feature vector; the recursive process is carried out in multiple rounds of iterations. In each iteration, the update of each high-dimensional nonlinear feature vector is affected by its previous state and constrained by the interaction with other high-dimensional nonlinear feature vectors. The recursive update formula is as follows:

[0059] ,

[0060] in, It is The high-dimensional nonlinear feature vector is The value after recursion; It is The high-dimensional nonlinear feature vector is The value after recursion; is the feature autoregressive weight matrix, which is used to control the temporal dynamics of each high-dimensional nonlinear eigenvector; It is the cross-weight matrix between features, which is used to quantify the mutual correlation between different high-dimensional nonlinear feature vectors; It is The high-dimensional nonlinear feature vector is The value after recursion. The feature expression is gradually optimized through the recursive process, and the interactive dependencies between high-dimensional nonlinear feature vectors are dynamically updated until convergence. When the convergence conditions are met (for example When the value is less than the preset convergence threshold, the recursion stops and the The final eigenvalue of a high-dimensional nonlinear eigenvector , After multiple recursive iterations, the temporal dynamics (through the feature autoregressive weight matrix ) and feature interactions (through the cross-weight matrix between features ) After High-dimensional nonlinear feature vector The final representation of .

[0061] Combining the relative importance of different recursive features and the global dependency of the inverse dynamic relationship matrix, a fusion feature with higher discrimination of defect patterns is constructed; in the dynamic fusion process, the dynamic fusion weight is assigned based on the hierarchical contribution of the final eigenvalue of each high-dimensional nonlinear feature vector. At the same time, the inverse dynamic relationship matrix is ​​introduced for dynamic fusion to generate the final fusion feature vector with higher discrimination of defect patterns; the dynamic fusion formula is:

[0062] ,

[0063] in, is the final fused feature vector; is the dynamic fusion weight, which is allocated according to the importance of the high-dimensional nonlinear feature vector; is the inverse dynamic adjustment coefficient, which is used to control the inverse dynamic relationship matrix influence; is the total number of high-dimensional nonlinear feature vectors, that is, the number of high-dimensional nonlinear feature vectors obtained by recursive calculation in the high-dimensional nonlinear feature space, The final fusion feature vector Input the classifier of the defect pattern and output the defect classification result:

[0064] ,

[0065] in, is the defect classification result, which represents the probability distribution of the classification categories of substation equipment defects; is the classifier weight, which is used to map the final fused feature vector to the classification space; is the bias, which is used to adjust the classification boundary.

[0066] Trend prediction is a key part of the defect diagnosis of secondary equipment in smart substations. Trend prediction is performed based on the final fusion feature vector and defect classification results. Error feedback and recursive optimization mechanisms are introduced to take the predicted state vector at the previous moment as input, and gradually optimize the predicted state vector at the future moment, providing an accurate prediction of the future operating state of substation equipment. The trend prediction results are dynamically adjusted through the following recursive formula:

[0067] ,

[0068] ,

[0069] ,

[0070] in, is the initial prediction value, indicating The predicted state vector at time t; The future The predicted state vector after time steps is the predicted state vector of the substation equipment operating state at the future moment, which specifically quantifies the continuity or possible change trend of the current operating state of the substation equipment; The future The predicted state vector after time steps; is the multi-objective error function; and The final fused feature vectors are And defect classification results The mapping weight matrix of is the learning rate, which is used to control the magnitude of each step in the gradient descent process. ; It's time The actual observed values ​​are obtained from the real equipment operation records; is the weight of the historical forecast error, which is determined by the size of the historical forecast error and is set experimentally; It is the weight of the dynamic feature consistency, ensuring the importance of feature consistency to prediction, through experimental settings. Through continuous iterative optimization, the trend prediction result is finally obtained. Each dimension of the trend prediction result represents a substation equipment operating state variable, such as voltage, current, fault probability, remaining life, etc.

[0071] In summary, a method for diagnosing defects in secondary equipment of smart substations has been completed.

[0072] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for diagnosing defects in secondary equipment of a smart substation, characterized in that: The following steps are involved: S1. Real-time data collection of multi-source signals of the operating status of substation equipment is performed, and filtering is performed to obtain real-time monitoring signals of smart substations; the real-time monitoring signals of smart substations are reversely rearranged and segmented to obtain reverse segmented signals; the correlation strength of reverse segmented signals is obtained by calculating the dot product integral of reverse segmented signals in different time periods in the time dimension, and a reverse dynamic matrix is ​​constructed; Based on the inverse dynamic matrix, the correlation between the inverse segmented signals is mapped through regularization processing and inverse analysis to obtain the inverse dynamic relationship matrix; S2, mapping the inverse dynamic relationship matrix to a high-dimensional nonlinear feature space to obtain a high-dimensional nonlinear feature vector; Introduction The nested recursive update mechanism combines the feature autoregressive weight matrix and the feature cross weight matrix, integrates the temporal dynamics and the interactive dependency between features, and dynamically optimizes the high-dimensional nonlinear feature vector; the recursive update formula is as follows: ， in, It is The high-dimensional nonlinear feature vector is The value after recursion; is a nonlinear activation function; It is The high-dimensional nonlinear feature vector is The value after recursion; is the feature autoregressive weight matrix; is the cross-weight matrix between features; It is The high-dimensional nonlinear feature vector is The value after recursion; When the preset convergence condition is met, the recursion is stopped, the final eigenvalue of the high-dimensional nonlinear feature vector is obtained, and dynamic fusion is performed to obtain the final fused feature vector; based on the final fused feature vector, the defect classification result is obtained; based on the final fused feature vector and the defect classification result, trend prediction is performed to obtain the trend prediction result.

2. A method for diagnosing secondary equipment defects in a smart substation according to claim 1, characterized in that: The S1 specifically includes: Through the reverse segmentation processing technology, the real-time monitoring signal of the smart substation is reversely rearranged and segmented according to the time interval to obtain a reverse segmentation signal.

3. A method for diagnosing secondary equipment defects in a smart substation according to claim 1, characterized in that: The S2 specifically includes: Through the projection operator, the inverse dynamic relationship matrix is ​​transformed into a high-dimensional nonlinear feature vector based on cosine transform and nonlinear activation.

4. A method for diagnosing secondary equipment defects in a smart substation according to claim 1, characterized in that: The S2 specifically includes: Based on the hierarchical contribution of the final eigenvalue of each high-dimensional nonlinear eigenvector, dynamic fusion weights are assigned, and an inverse dynamic relationship matrix is ​​introduced to perform dynamic fusion to generate the final fused eigenvector.

5. A method for diagnosing secondary equipment defects in a smart substation according to claim 4, characterized in that: The S2 specifically includes: The final fused feature vector is input into the defect pattern classifier to obtain the defect classification result.

6. A method for diagnosing secondary equipment defects in a smart substation according to claim 5, characterized in that: The S2 specifically includes: Based on the final fusion feature vector and defect classification results, the initial prediction value is obtained; the error feedback and recursive optimization mechanism are introduced, and the prediction state vector at the previous moment is combined to optimize the prediction state vector at the future moment to obtain the trend prediction result; the trend prediction result is dynamically adjusted through the following recursive formula: ， ， ， in, is the initial prediction value, indicating The predicted state vector at time t; The future The predicted state vector after time steps; The future The predicted state vector after time steps; is the multi-objective error function; and They are the final fused feature vectors And defect classification results The mapping weight matrix of is the learning rate, which is used to control the magnitude of each step in the gradient descent process; It's time The actual observed value of is the weight of historical forecast errors; is the weight of the dynamic feature consistency; It is the inverse dynamic relationship matrix.

Images