An intelligent detection and early warning method based on multi-level dynamic analysis

Abstract

The present invention discloses an intelligent detection and early warning method based on multi-level dynamic analysis, and the method includes data collection, data preprocessing, establishing an intelligent detection and early warning model, and intelligent detection and early warning based on multi-level dynamic analysis. The present invention belongs to the field of anomaly detection technology, and specifically refers to an intelligent detection and early warning method based on multi-level dynamic analysis. This scheme improves the accuracy of anomaly detection by iteratively selecting samples with information and combining a weighted sampling strategy. Through iterative sampling strategies and convergence checks, it avoids the overfitting or underfitting problems that are prone to occur in traditional methods, and achieves efficient and accurate early warning; optimizes the position of individuals with the help of gravitational strength and attenuation coefficients and random disturbances to accelerate the convergence of the model; controls the optimization direction by dynamically adjusting the importance score; and introduces a heavy load ratio that enables the model to have the ability to adjust adaptively; and can maintain efficient and accurate detection and early warning capabilities in dynamically changing systems.

Description

Technical Field

[0001] The present invention relates to the technical field of anomaly detection, and in particular to an intelligent detection and early warning method based on multi-level dynamic analysis. Background Art

[0002] Detection and early warning methods are a type of method that combines artificial intelligence algorithms and traditional monitoring methods to provide early warning and detection of anomalies, faults, risks, etc. in complex systems or environments. This type of method can automatically analyze large amounts of data, identify potential risks and abnormalities, and give early warnings, thereby providing protection for system operation and improving system reliability. However, general detection and early warning methods have the problem of using random or fixed sample sets for training, which leads to single training and cannot adapt to the complex operating environment of the detection and early warning model. The prediction ability of boundary data is weak, resulting in insufficient detection accuracy. General detection and early warning methods have the problem of slow convergence speed when adjusting model parameters, unable to effectively control the optimization direction, and unable to maintain the effectiveness of global search, which leads to the problem that the model cannot efficiently capture key features and issue accurate early warning signals. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent detection and early warning method based on multi-level dynamic analysis. The general detection and early warning method has the problem that the training is single and cannot adapt to the complex operating environment of the detection and early warning model, and the prediction ability of the boundary data is weak, resulting in insufficient detection accuracy. This scheme iteratively selects samples with information and combines it with a weighted sampling strategy to improve the accuracy of anomaly detection while reducing samples. The sampling strategy focuses on samples near the boundary of potential limit states, which can significantly improve the model's ability to recognize fault boundaries. Through iterative sampling strategies and convergence checks, the model is effectively prevented from being overly dependent on initial data, avoiding overfitting or underfitting problems that are prone to occur in traditional methods, and achieving efficient and accurate early warning. The general detection and early warning method has the problem that the convergence speed is slow when adjusting the model parameters and cannot effectively control the optimal This solution uses gravitational strength, attenuation coefficient and random disturbance to optimize the position of individuals and accelerate the convergence of the model. It ensures that the intelligent early warning model can find the optimal solution through multi-dimensional exploration when facing complex nonlinear data, and timely identify potential risks. By introducing the importance score, the model can identify which individuals contribute the most to model training, so as to focus on optimizing these important individuals in the next iteration, dynamically adjust the importance score of the model, and effectively control the optimization direction. The introduction of the overload ratio enables the model to have the ability of adaptive adjustment, ensuring that the effectiveness of the global search can be maintained in multiple iterations, so that the model has stronger self-correction ability and can adapt to different risk scenarios. It can maintain efficient and accurate detection and early warning capabilities in dynamically changing systems.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent detection and early warning method based on multi-level dynamic analysis, the method comprising the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: Establishing an intelligent detection and early warning model;

[0008] Step S4: Intelligent detection and early warning based on multi-level dynamic analysis.

[0009] Furthermore, in step S1, the data collection is to collect multi-level dynamic data; the multi-level dynamic data includes physical layer data, model layer data, function layer data, security layer data and status type; the status type includes normal status and abnormal status; the status type is used as a data label.

[0010] Furthermore, in step S2, the data preprocessing is to clean the collected data, convert the data and divide the data set; the data cleaning is to process missing values, duplicate values ​​and outliers; the data conversion is to convert the data into vector form and perform standardization; the data set division is to generate N vectors containing the original probability distribution of the original data. MC The Monte Carlo population of samples, as the initial candidate sample, is expressed as: ;Use uniform sampling for preliminary selection, the weight coefficient is , get the data set and divide it into test set and training set; where S MC is the generated Monte Carlo sample set, is the i-th sample, is the original probability distribution value.

[0011] Furthermore, in step S3, the establishment of the intelligent detection and early warning model specifically includes the following steps:

[0012] Step S31: initial training; using the training set to train the AAE-SDR neural network of the encoder structure; the AAE-SDR neural network includes an encoder and a decoder, which is optimized by the encoder loss function, expressed as: ; Initially generate latent representations in the latent space , and learn the potential limit state function Where, L AE is the encoder loss function, x is the input data, is the encoder function, is the decoder function;

[0013] Step S32: Sample identification: Use the initially trained AAE-SDR neural network to identify misclassified samples. Misclassified samples are samples that are predicted incorrectly in the potential limit state function, expressed as: ; Candidate samples are selected based on the distance from the potential representation to the potential limit state boundary, expressed as: ; In the formula, is the predicted label; error is the wrong classification, y is the true label, is the sample set after screening, k is the convergence coefficient, τ is the current number of training times, LSF is the limit state function, d(·) is the Euclidean distance, is the maximum allowable distance from the potential failure specimen to the limit state boundary;

[0014] Step S33: weighted sampling: select samples from candidate samples by weighted sampling, and the weight coefficient is based on the probability density of the potential representation of the sample, expressed as: ; In the formula, is the weighted probability for the potential representation z, and p(z) is the probability density function of the potential representation z;

[0015] Step S34: model training: retrain the AAE-SDR neural network using the updated data set, and return to step S32 after each training is completed;

[0016] Step S35: Estimate the failure probability; predict the limit state function value of each sample in the Monte Carlo population through the AAE-SDR neural network, and estimate the failure probability. The formula used is as follows:

[0017] ;

[0018] In the formula, is the estimated failure probability; I[·] is the indicator function; N MC is the sample size;

[0019] Step S36: Convergence check; verify the relative difference between the failure probabilities in adjacent iterations. When the relative difference is less than the predetermined threshold value ecr, the training of the intelligent detection and early warning model is completed. If the classification accuracy of the trained intelligent detection and early warning model for the test set is higher than the accuracy threshold, the intelligent detection and early warning model is established. Otherwise, the data set is re-divided and the step S37 is performed to adjust the model parameters. The formula used is as follows:

[0020] ;

[0021] In the formula, and are the failure probabilities at the τth and τ-1th training times, respectively; It is a relative difference;

[0022] Step S37: Adjust model parameters; specifically including:

[0023] Step S371: Establish parameter optimization space based on distance threshold, convergence coefficient, truncated distribution parameter, convergence threshold and initial parameters of AAE-SDR neural network; randomly initialize individual positions in the search population, and use the prediction accuracy of the intelligent detection and early warning model trained based on individual positions for the test set as the individual fitness value;

[0024] Step S372: Location update; the formula used is as follows:

[0025] ;

[0026] In the formula, and are the positions of the i1th individual in the jth dimension at the t+1th and tth iterations, respectively; β0 is a constant that controls the strength of gravity; μ is the attenuation coefficient, and r is the distance between the current individual in the jth dimension and the global optimal solution; is the global optimal position of the jth dimension at the tth iteration; is the fitness value of the i1th individual; rand is a random number between 0 and 1;

[0027] Step S373: define the importance score; the formula used is as follows:

[0028] ;

[0029] In the formula, f t+1 and f t are the importance scores of the population at the t+1th iteration and the tth iteration respectively; pn is the number of individuals in the population; is the historical highest fitness value of the i1th individual at the tth iteration; is the average fitness value of the population;

[0030] Step S374: Calculate the overload ratio; an importance threshold is preset, and when the population importance score is lower than the importance threshold, the individual positions of individuals with low fitness values ​​in the population are reinitialized based on the overload ratio; the formula used to calculate the overload ratio is as follows:

[0031] ;

[0032] Where, pp is the overload ratio; pp min is the minimum overload ratio; pp max is the maximum overload ratio;

[0033] Step S375: search judgment; a fitness threshold is set in advance. When there is an individual fitness value higher than the fitness threshold, an intelligent detection and early warning model based on the individual position is obtained; otherwise, if the maximum number of iterations is reached, go to step S371, otherwise go to step S372.

[0034] Furthermore, in step S4, the intelligent detection and early warning based on multi-level dynamic analysis is based on the established intelligent detection and early warning model, and real-time collection of physical layer data, model layer data, functional layer data and security layer data are input into the intelligent detection and early warning model after preprocessing. If the status type output by the intelligent detection and early warning model is an abnormal status, early warning processing is performed.

[0035] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0036] (1) In view of the problem that general detection and early warning methods use random or fixed sample sets for training, which leads to single training and cannot adapt to the complex operating environment of the detection and early warning model, and the weak prediction ability of boundary data leads to insufficient detection accuracy, this scheme iteratively selects informative samples and combines them with weighted sampling strategies to improve the accuracy of anomaly detection while reducing samples. The sampling strategy focuses on samples near the boundary of potential limit states, which can significantly improve the model's ability to identify fault boundaries. Through iterative sampling strategies and convergence checks, it effectively prevents the model from being overly dependent on initial data, avoids the overfitting or underfitting problems that are prone to occur in traditional methods, and achieves efficient and accurate early warning.

[0037] (2) In view of the problems that general detection and early warning methods have slow convergence speed when adjusting model parameters, cannot effectively control the optimization direction, and cannot maintain the effectiveness of global search, which leads to the model's inability to efficiently capture key features and issue accurate early warning signals, this solution uses gravitational strength and attenuation coefficient as well as random perturbations to optimize the position of individuals and accelerate the convergence of the model; ensure that the intelligent early warning model finds the optimal solution through multi-dimensional exploration when facing complex nonlinear data and identifies potential risks in a timely manner; by introducing importance scores, the model can identify which individuals contribute the most to model training, so that these important individuals can be optimized in the next iteration, the importance scores of the model can be dynamically adjusted, and the optimization direction can be effectively controlled; the introduction of the overload ratio enables the model to have the ability of adaptive adjustment, ensuring that the effectiveness of global search can be maintained in multiple iterations, so that the model has stronger self-correction capabilities and can adapt to different risk scenarios; it can maintain efficient and accurate detection and early warning capabilities in dynamically changing systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a process flow of an intelligent detection and early warning method based on multi-level dynamic analysis provided by the present invention;

[0039] Figure 2 It is a schematic diagram of the process of step S3.

[0040] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0042] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0043] Example 1, see Figure 1 The present invention provides an intelligent detection and early warning method based on multi-level dynamic analysis, which comprises the following steps:

[0044] Step S1: data collection, collecting multi-level dynamic data;

[0045] Step S2: Data preprocessing, data cleaning, data conversion and data set division of the collected data;

[0046] Step S3: Establish an intelligent detection and early warning model, select samples iteratively based on weighted sampling strategy and sample identification; perform convergence check based on failure probability, disturb the position of parameter optimization individuals by using gravity strength and attenuation coefficient, introduce importance score to optimize individual screening, and finally complete the establishment of the intelligent detection and early warning model;

[0047] Step S4: Intelligent detection and early warning based on multi-level dynamic analysis, and realizing intelligent early warning detection on the data to be detected based on the established intelligent detection and early warning model.

[0048] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the multi-level dynamic data includes physical layer data, model layer data, function layer data, security layer data and status type; the status type includes normal status and abnormal status; the status type is used as a data label.

[0049] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, data cleaning is to process missing values, duplicate values ​​and outliers; data conversion is to convert data into vector form and perform standardization; data set partitioning is to generate N vectors containing the original probability distribution of the original data. MC The Monte Carlo population of samples, as the initial candidate sample, is expressed as: ;Use uniform sampling for preliminary selection, the weight coefficient is , get the data set and divide it into test set and training set; where S MC is the generated Monte Carlo sample set, is the i-th sample, is the original probability distribution value.

[0050] Example 4, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S3, establishing an intelligent detection and early warning model specifically includes the following steps:

[0051] Step S31: initial training; using the training set to train the AAE-SDR neural network of the encoder structure; the AAE-SDR neural network includes an encoder and a decoder, which is optimized by the encoder loss function, expressed as: ; Initially generate latent representations in the latent space , and learn the potential limit state function Where, L AE is the encoder loss function, x is the input data, is the encoder function, is the decoder function;

[0052] Step S32: Sample identification: Use the initially trained AAE-SDR neural network to identify misclassified samples. Misclassified samples are samples that are predicted incorrectly in the potential limit state function, expressed as: ; Candidate samples are selected based on the distance from the potential representation to the potential limit state boundary, expressed as: ; In the formula, is the predicted label; error is the wrong classification, y is the true label, is the sample set after screening, k is the convergence coefficient, τ is the current number of training times, LSF is the limit state function, d(·) is the Euclidean distance, is the maximum allowable distance from the potential failure specimen to the limit state boundary;

[0053] Step S33: weighted sampling: select samples from candidate samples by weighted sampling, and the weight coefficient is based on the probability density of the potential representation of the sample, expressed as: ; In the formula, is the weighted probability for the potential representation z, and p(z) is the probability density function of the potential representation z;

[0054] Step S34: model training: retrain the AAE-SDR neural network using the updated data set, and return to step S32 after each training is completed;

[0055] Step S35: Estimate the failure probability; predict the limit state function value of each sample in the Monte Carlo population through the AAE-SDR neural network, and estimate the failure probability. The formula used is as follows:

[0056] ;

[0057] In the formula, is the estimated failure probability; I[·] is the indicator function; N MC is the sample size;

[0058] Step S36: Convergence check; verify the relative difference between the failure probabilities in adjacent iterations. When the relative difference is less than the predetermined threshold value ecr, the training of the intelligent detection and early warning model is completed. If the classification accuracy of the trained intelligent detection and early warning model for the test set is higher than the accuracy threshold, the intelligent detection and early warning model is established. Otherwise, the data set is re-divided and the step S37 is performed to adjust the model parameters. The formula used is as follows:

[0059] ;

[0060] In the formula, and are the failure probabilities at the τth and τ-1th training times, respectively; It is a relative difference;

[0061] Step S37: Adjust model parameters.

[0062] By performing the above operations, the general detection and early warning methods have the problem of using random or fixed sample sets for training, which leads to single training and cannot adapt to the complex operating environment of the detection and early warning model. The weak prediction ability of the boundary data leads to insufficient detection accuracy. This solution iteratively selects informative samples and combines them with weighted sampling strategies to improve the accuracy of anomaly detection while reducing samples. The sampling strategy focuses on samples near the boundary of potential limit states, which can significantly improve the model's ability to identify fault boundaries. Through iterative sampling strategies and convergence checks, the model is effectively prevented from being overly dependent on initial data, avoiding overfitting or underfitting problems that are prone to occur in traditional methods, and achieving efficient and accurate early warning.

[0063] Example 5, see Figure 2 This embodiment is based on the above embodiment. In step S37, adjusting the model parameters specifically includes:

[0064] Step S371: Establish parameter optimization space based on distance threshold, convergence coefficient, truncated distribution parameter, convergence threshold and initial parameters of AAE-SDR neural network; randomly initialize individual positions in the search population, and use the prediction accuracy of the intelligent detection and early warning model trained based on individual positions for the test set as the individual fitness value;

[0065] Step S372: Location update; the formula used is as follows:

[0066] ;

[0067] In the formula, and are the positions of the i1th individual in the jth dimension at the t+1th and tth iterations, respectively; β0 is a constant that controls the strength of gravity; μ is the attenuation coefficient, and r is the distance between the current individual in the jth dimension and the global optimal solution; is the global optimal position of the jth dimension at the tth iteration; is the fitness value of the i1th individual; rand is a random number between 0 and 1;

[0068] Step S373: define the importance score; the formula used is as follows:

[0069] ;

[0070] In the formula, f t+1 and f t are the importance scores of the population at the t+1th iteration and the tth iteration respectively; pn is the number of individuals in the population; is the historical highest fitness value of the i1th individual at the tth iteration; is the average fitness value of the population;

[0071] Step S374: Calculate the overload ratio; an importance threshold is preset. When the population importance score is lower than the importance threshold, the individual positions of individuals with low fitness values ​​in the population are reinitialized based on the overload ratio; the formula used to calculate the overload ratio is as follows:

[0072] ;

[0073] Where, pp is the overload ratio; pp min is the minimum overload ratio; pp max is the maximum overload ratio;

[0074] Step S375: search judgment; a fitness threshold is set in advance. When there is an individual fitness value higher than the fitness threshold, an intelligent detection and early warning model based on the individual position is obtained; otherwise, if the maximum number of iterations is reached, go to step S371, otherwise go to step S372.

[0075] By performing the above operations, the general detection and early warning methods have the problems of slow convergence speed when adjusting model parameters, inability to effectively control the optimization direction, and inability to maintain the effectiveness of global search, which leads to the model's inability to efficiently capture key features and issue accurate early warning signals. This solution uses gravity intensity and attenuation coefficient as well as random disturbances to optimize the position of individuals and accelerate the convergence of the model; ensure that the intelligent early warning model finds the optimal solution through multi-dimensional exploration when facing complex nonlinear data, and identifies potential risks in a timely manner; by introducing importance scores, the model can identify which individuals contribute the most to model training, so as to focus on optimizing these important individuals in the next iteration, dynamically adjust the importance scores of the model, and effectively control the optimization direction; the introduction of the overload ratio enables the model to have the ability of adaptive adjustment, ensuring that the effectiveness of global search can be maintained in multiple iterations, so that the model has stronger self-correction capabilities and adapts to different risk scenarios; it can maintain efficient and accurate detection and early warning capabilities in dynamically changing systems.

[0076] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S4, the intelligent detection and early warning based on multi-level dynamic analysis is based on the established intelligent detection and early warning model, and real-time collection of physical layer data, model layer data, functional layer data and security layer data are input into the intelligent detection and early warning model after preprocessing. If the status type output by the intelligent detection and early warning model is abnormal status, early warning processing is performed.

[0077] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0078] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0079] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent detection and early warning method based on multi-level dynamic analysis, characterized in that: The method comprises the following steps: Step S1: data collection, collecting multi-level dynamic data; Step S2: data preprocessing; Step S3: Establish an intelligent detection and early warning model, select samples iteratively based on weighted sampling strategy and sample identification; perform convergence check based on failure probability, disturb the position of parameter optimization individuals by using gravity strength and attenuation coefficient, introduce importance score to optimize individual screening, and finally complete the establishment of the intelligent detection and early warning model; Step S4: Intelligent detection and early warning based on multi-level dynamic analysis, based on the established intelligent detection and early warning model to realize intelligent early warning detection of the detection data; In step S1, the multi-level dynamic data includes physical layer data, model layer data, function layer data, security layer data and status type; the status type includes normal status and abnormal status; the status type is used as a data label; In step S4, the intelligent detection and early warning based on multi-level dynamic analysis is based on the established intelligent detection and early warning model, and real-time data of physical layer, model layer, function layer and security layer are collected, and input into the intelligent detection and early warning model after preprocessing. If the state type output by the intelligent detection and early warning model is abnormal state, early warning processing is performed; Step S3 includes step S37: adjusting model parameters; specifically including: Step S371: Establish parameter optimization space based on distance threshold, convergence coefficient, truncated distribution parameter, convergence threshold and initial parameters of AAE-SDR neural network; randomly initialize individual positions in the search population, and use the prediction accuracy of the intelligent detection and early warning model trained based on individual positions for the test set as the individual fitness value; Step S372: Location update; the formula used is as follows: ； In the formula, and are the positions of the i1th individual in the jth dimension at the t+1th and tth iterations respectively; β0 is the constant that controls the gravitational strength; μ is the attenuation coefficient, and r is the distance between the current individual in the jth dimension and the global optimal solution; is the global optimal position of the jth dimension at the tth iteration; is the fitness value of the i1th individual; rand is a random number between 0 and 1; Step S373: define the importance score; the formula used is as follows: ； In the formula, f t+1 and f t are the importance scores of the population at the t+1th iteration and the tth iteration respectively; pn is the number of individuals in the population; is the historical highest fitness value of the i1th individual at the tth iteration; is the average fitness value of the population; Step S374: Calculate the overload ratio; an importance threshold is preset, and when the population importance score is lower than the importance threshold, the individual positions of individuals with low fitness values ​​in the population are reinitialized based on the overload ratio; the formula used to calculate the overload ratio is as follows: ； Where, pp is the overload ratio; pp min is the minimum overload ratio; pp max is the maximum overload ratio; Step S375: search judgment; a fitness threshold is set in advance. When there is an individual fitness value higher than the fitness threshold, an intelligent detection and early warning model based on the individual position is obtained; otherwise, if the maximum number of iterations is reached, go to step S371, otherwise go to step S372.

2. The intelligent detection and early warning method based on multi-level dynamic analysis according to claim 1 is characterized in that: In step S3, the establishment of the intelligent detection and early warning model specifically includes the following steps: Step S31: initial training; using the training set to train the AAE-SDR neural network of the encoder structure; the AAE-SDR neural network includes an encoder and a decoder, which is optimized by the encoder loss function, expressed as: ; Initially generate latent representations in the latent space , and learn the potential limit state function Where, L AE is the encoder loss function, x is the input data, is the encoder function, is the decoder function; Step S32: Sample identification: Use the initially trained AAE-SDR neural network to identify misclassified samples. Misclassified samples are samples that are predicted incorrectly in the potential limit state function, expressed as: ; Candidate samples are selected based on the distance from the potential representation to the potential limit state boundary, expressed as: ; In the formula, is the predicted label; error is the wrong classification, y is the true label, is the sample set after screening, k is the convergence coefficient, τ is the current number of training times, LSF is the limit state function, d(·) is the Euclidean distance, is the maximum allowable distance from the potential failure sample to the limit state boundary, is the i-th sample, x is the input data, S MC is the generated Monte Carlo sample set, z is the potential representation; is the encoder function, is the decoder function; Step S33: weighted sampling: select samples from candidate samples by weighted sampling, and the weight coefficient is based on the probability density of the potential representation of the sample, expressed as: ; In the formula, is the weighted probability for the potential representation z, and p(z) is the probability density function of the potential representation z; Step S34: model training: retrain the AAE-SDR neural network using the updated data set, and return to step S32 after each training is completed; Step S35: Estimate the failure probability; predict the limit state function value of each sample in the Monte Carlo population through the AAE-SDR neural network, and estimate the failure probability. The formula used is as follows: ； In the formula, is the estimated failure probability; I[·] is the indicator function; N MC is the sample size; Step S36: Convergence check; verify the relative difference between the failure probabilities in adjacent iterations. When the relative difference is less than the predetermined threshold value ecr, the training of the intelligent detection and early warning model is completed. If the classification accuracy of the trained intelligent detection and early warning model for the test set is higher than the accuracy threshold, the intelligent detection and early warning model is established. Otherwise, the data set is re-divided and the step S37 is performed to adjust the model parameters. The formula used is as follows: ； In the formula, and are the failure probabilities at the τth and τ-1th training times, respectively; It is a relative difference; Step S37: Adjust model parameters.

3. The intelligent detection and early warning method based on multi-level dynamic analysis according to claim 1 is characterized in that: In step S2, data cleaning is to process missing values, duplicate values ​​and outliers; data conversion is to convert data into vector form and perform standardization; data set partitioning is to generate N vectors containing the original probability distribution of the original data. MC The Monte Carlo population of samples, as the initial candidate sample, is expressed as: ;Use uniform sampling for preliminary selection, the weight coefficient is , get the data set and divide it into test set and training set; where S MC is the generated Monte Carlo sample set, is the i-th sample, is the original probability distribution value.

Images