CTabDDPM-based optical network fault diagnosis data augmentation method and system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH BEIJING
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies for optical network fault diagnosis suffer from problems such as a lack of physical consistency constraints in generated data, insufficient ability to maintain complex feature association structures, and insufficient controllable enhancement capabilities for specific fault categories, leading to reduced diagnostic reliability.
The CTabDDPM-based approach is adopted. By injecting fault labels as conditional vectors into the reverse denoising link, the CTabDDPM model is constructed. Rule constraints and consistency screening are introduced to build a multi-dimensional evaluation system, generate high-quality fault data that conforms to the target category distribution, and perform data mixing to train the fault diagnosis model.
It improves the identification capability and reliability of optical network fault diagnosis, and the generated data is physically and statistically consistent with real data, effectively alleviating the class imbalance problem and improving the generalization ability and robustness of the diagnostic model.
Smart Images
Figure CN122227115A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of optical communication network fault diagnosis technology, and in particular to a data enhancement method and system for optical network fault diagnosis based on CTabDDPM. Background Technology
[0002] Fault diagnosis is a fundamental technology in optical network operation and maintenance (O&M). Its purpose is to determine whether a device has malfunctioned based on its operational monitoring parameters and to further identify the type of fault. This technology directly impacts fault location efficiency, service recovery time, and network operational stability. With the expansion of optical network scale and the increase in equipment complexity, the number of monitoring parameters continues to rise, and the relationships between parameters become more complex. Traditional diagnostic methods relying on human experience, rule bases, and fixed thresholds are no longer sufficient to meet engineering requirements in terms of real-time performance, accuracy, and scalability. To address these issues, the industry is gradually adopting data-driven fault diagnosis methods. Data-driven methods involve training models using historical operational samples, enabling the models to learn the mapping relationship between device parameters and fault states, and then using this knowledge for the automatic identification and classification of new samples. Compared to rule-based methods, this type of method can improve diagnostic efficiency and automation levels under complex operating conditions, and has become an important technological direction for intelligent O&M of optical networks.
[0003] However, in real-world optical network scenarios, data-driven methods have long been constrained by the distribution of training data. First, the number of normal samples far exceeds the number of fault samples, and there are significant differences in sample numbers between different fault categories, resulting in a substantial class imbalance. Second, fault samples are primarily derived from real-world fault events, leading to a passive acquisition process that results in long lead times, high costs, and insufficient repeatability in obtaining high-quality labeled samples. These factors contribute to a persistent shortage of minority class fault samples, causing model training to tend to favor the majority class. This manifests as a decrease in minority class fault identification capabilities and an increase in false negative rates, ultimately impacting the effectiveness of fault diagnosis systems in high-reliability scenarios.
[0004] To alleviate the problem of insufficient minority class samples, existing technologies typically employ generative data augmentation methods. This involves synthesizing fault samples using generative models and co-training the diagnostic model with the original samples. Examples include Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). While GANs can be used for distributed learning and sample generation, they suffer from training instability and the risk of pattern collapse, potentially leading to insufficient diversity in generated samples. VAEs offer relatively stable training, but the generated samples are prone to over-smoothing, resulting in insufficient discrimination of fault features. Although these methods can statistically expand the sample size, they generally suffer from insufficient representation of physical constraints in optical network scenarios. Specifically, they fail to adequately maintain the coupling relationships between multiple device parameters, potentially generating samples with unreasonable mechanisms, thus affecting the reliability of downstream diagnostic models.
[0005] The development of the Denoising Diffusion Probabilistic Model (DDPM) has provided a new technical approach to this problem. DDPM learns complex distributions step by step through forward denoising and backward denoising, exhibiting superior training stability and pattern coverage compared to traditional adversarial generation methods, and has shown good generation performance in multiple fault diagnosis tasks. However, standard DDPM essentially learns the overall distribution and lacks the ability to precisely control the generation of target categories, making it difficult to directly meet the needs of optical network fault category-oriented enhancement.
[0006] In summary, the existing technologies have the following shortcomings: (1) The generated data lacks sufficient physical consistency constraints, which limits the engineering credibility; (2) The ability to maintain the complex feature association structure is insufficient, and the sample structure fidelity is not high; (3) The controllable enhancement capability for specific fault categories is insufficient, making it difficult to continuously improve diagnostic reliability under extreme imbalance conditions. Summary of the Invention
[0007] To address the aforementioned technical problems in the existing technology, this invention provides a method and system for enhancing optical network fault diagnosis data based on CTabDDPM. The technical solution is as follows: On the one hand, a data augmentation method for optical network fault diagnosis based on CTabDDPM is provided. The method includes: preprocessing an initial dataset consisting of performance parameters characterizing the operating status of optical network devices to obtain an initial input feature dataset; injecting fault labels as conditional vectors into the reverse denoising link of the TabDDPM model to construct a CTabDDPM model; augmenting the fault data in the initial input feature dataset based on the CTabDDPM model, and introducing rule constraints and consistency screening to obtain an augmented dataset; constructing a multi-dimensional evaluation system based on distribution similarity, structural authenticity, and task utility, and validating the augmented dataset based on the multi-dimensional evaluation system to obtain a validated augmented dataset; and mixing the validated augmented dataset with the initial input feature dataset according to a preset ratio to construct target training sets with different balances.
[0008] Optionally, the method further includes: training a fault diagnosis model based on the target training set to classify and identify the fault status of optical network devices.
[0009] Optionally, the CTabDDPM model includes a model training phase and a sampling generation phase; wherein, the model training phase constructs labeled training data based on the forward diffusion process, and trains a denoising neural network based on the labeled training data; the denoising neural network learns by predicting the noise injected in the forward diffusion based on the noisy data, the time step, and the conditional vector; the sampling generation phase utilizes the trained denoising neural network, inputs the target conditional vector, and generates new high-quality optical network device fault data through an iterative reverse denoising process; the training loss function of the CTabDDPM model includes:
[0010] The sampling and generation process of the CTabDDPM model includes:
[0011]
[0012] In the formula, ε represents the noise that needs to be estimated and trained using a diffusion model. This represents the noise predicted by the denoising neural network at time step t. t For hyperparameters, α t It is the predefined variance at time step t, z represents standard Gaussian noise, x0 is the original sample, x t It is the noisy data after t-step diffusion, x t-1 It is the denoised data from step t-1.
[0013] Optionally, the rule constraint and consistency screening includes: performing threshold range constraints and correlation checks on parameter combinations, and retaining sample data that meet the conditions.
[0014] Optionally, the multidimensional evaluation system includes: evaluating distribution similarity based on the maximum mean difference, Wasserstein distance, and L2 distance; evaluating structural authenticity based on complementary criteria consisting of Pearson correlation coefficient, mutual information, and adversarial evaluation; and evaluating task utility based on a combination of accuracy, F1 score, false negative rate, and false positive rate.
[0015] Optionally, the method further includes: jointly determining the preset ratio under reliability constraints and computing resource constraints.
[0016] On the other hand, a data augmentation system for optical network fault diagnosis based on CTabDDPM is also provided to implement the data augmentation method for optical network fault diagnosis based on CTabDDPM provided in the embodiments of the present invention. The system includes: a preprocessing module, a model building module, a data augmentation module, a data validation module, and a data mixing module. The preprocessing module preprocesses an initial dataset composed of performance parameters characterizing the operating status of optical network devices to obtain an initial input feature dataset. The model building module injects fault labels as conditional vectors into the reverse denoising link of the TabDDPM model to construct the CTabDDPM model. The data augmentation module enhances the fault data in the initial input feature dataset based on the CTabDDPM model and introduces rule constraints and consistency screening to obtain an augmented dataset. The data validation module constructs a multi-dimensional evaluation system based on distribution similarity, structural authenticity, and task utility, and validates the augmented dataset based on the multi-dimensional evaluation system to obtain a validated augmented dataset. The data mixing module mixes the validated augmented dataset with the initial input feature dataset according to a preset ratio to construct target training sets with different balances.
[0017] Optionally, the system further includes a classification and identification module, used to train a fault diagnosis model based on the target training set, and to classify and identify the fault status of optical network devices.
[0018] On the other hand, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.
[0019] On the other hand, a computer-readable storage medium is also provided, wherein program code is stored in the computer-readable storage medium, and the program code can be called by a processor to execute the method provided in the embodiments of the present invention.
[0020] This invention provides a data augmentation method and system for optical network fault diagnosis based on CTabDDPM. By introducing fault category conditions into the diffusion generation process, a sample distribution learning mechanism oriented towards category constraints is constructed. This enables the targeted generation of samples for a minority of fault types and the optimization of training data distribution, reducing the impact of training samples being dominated by the majority class and alleviating the problems of insufficient minority class representation and decision boundary shift. By introducing physical consistency constraints in the forward noise addition and backward denoising processes, the synthesized samples are guaranteed to satisfy the mechanistic correlation and evolution law between key monitoring parameters, ensuring that the generated data simultaneously meets the requirements of fault diagnosis tasks in terms of physical interpretability, structural correlation preservation, and statistical distribution consistency. By incorporating distribution consistency, structural fidelity, and downstream diagnostic performance into a unified evaluation process, and determining the injection ratio of generated samples and key model parameters accordingly, the stability of the transformation of augmentation results into diagnostic performance gains can be improved. This invention alleviates the technical problems of insufficient controllable augmentation capability for specific fault categories and reduced diagnostic reliability under extreme imbalance conditions in existing technologies. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a flowchart of a data enhancement method for optical network fault diagnosis based on CTabDDPM provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the change in enhanced data performance with the enhancement ratio provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an optical network fault diagnosis data enhancement system based on CTabDDPM provided in an embodiment of the present invention. Detailed Implementation
[0023] The technical solution of the present invention will now be described with reference to the accompanying drawings.
[0024] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.
[0025] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.
[0026] Figure 1 This is a flowchart of a data enhancement method for optical network fault diagnosis based on CTabDDPM, provided by an embodiment of the present invention. Figure 1 As shown, the method specifically includes the following steps: Step S102: Preprocess the initial dataset consisting of performance parameters characterizing the operating status of optical network devices to obtain the initial input feature dataset.
[0027] Step S104: Inject the fault label as a conditional vector into the reverse denoising link of the TabDDPM model to construct the CTabDDPM model.
[0028] Step S106: Based on the CTabDDPM model, the fault data in the initial input feature dataset is enhanced, and rule constraints and consistency screening are introduced to obtain the enhanced dataset.
[0029] Step S108: Construct a multi-dimensional evaluation system based on distribution similarity, structural authenticity and task utility, and validate the augmented dataset based on the multi-dimensional evaluation system to obtain the validated augmented dataset.
[0030] Step S110: Mix the validated augmented dataset with the initial input feature dataset according to a preset ratio to construct target training sets with different balances.
[0031] Specifically, after step S110, the method provided in this embodiment of the invention further includes: A fault diagnosis model is trained based on the target training set to classify and identify the fault status of optical network devices.
[0032] This invention provides a data augmentation method for optical network fault diagnosis based on CTabDDPM, which aims to inject fault labels as conditional vectors into the reverse denoising link, so that the model changes from learning the overall distribution to learning the conditional distribution, thereby realizing targeted generation for the target fault category and solving the problems of uncontrollable supplementation of minority classes and high proportion of invalid samples.
[0033] Tabular Denoising Diffusion Probabilistic Model (TabDDPM) is a tabular data generation method based on the Denoising Diffusion Probabilistic Model (DDPM). Designed for structured tabular data, it exhibits good generation performance in data augmentation tasks. Forward diffusion and backward denoising constitute the core mechanism of TabDDPM. Forward diffusion is a fixed and training-free noise addition process that progressively injects Gaussian noise into the original input data, eventually degenerating it into approximately pure noise. Backward denoising, on the other hand, progressively reconstructs the data distribution from the noisy data, predicting and removing the noise introduced during the forward process, generating high-quality samples that conform to the original distribution through multiple iterations. The forward diffusion and backward denoising processes can be defined as follows:
[0034]
[0035] Where x0 is the original sample, x t It is the noisy data after t-step diffusion, x t-1 It is the denoised data from step t-1.
[0036] Due to the true inverse distribution p(x) t-1 |x t Since it is difficult to solve directly, TabDDPM uses a neural network ε with parameter θ. θ This approximates the distribution. The network's input includes noisy data x. t Given a time step t and a conditional vector y, the output is the predicted noise ε. θ Through multiple iterations, high-quality samples that conform to the original data distribution can be generated. The training process is achieved by optimizing the following loss function:
[0037] Among them, L simple Corresponding to numerical feature loss, L i The loss corresponding to the classification features.
[0038] Compared to the basic TabDDPM, the Conditional Tabular Denoising Diffusion Probabilistic Model (CTabDDPM) provided in this embodiment introduces a conditional vector y during the reverse denoising process to guide the model in generating samples that conform to the target class distribution. This mechanism has two key improvements: 1) Achieve precise conditional control and targeted generation: Unconditional TabDDPM learns the overall joint distribution P(x), and the generation process is random; CTabDDPM learns the conditional distribution p(x|y), and can generate target category samples by specifying conditional vectors.
[0039] 2) Enhanced learning ability of complex associations between labels and features: In real-world scenarios, different category labels typically correspond to multi-parameter linkage change patterns. The conditional generation mechanism prompts the model to learn the non-linear relationship between labels and state parameters, ensuring that the generated samples are not only reasonable at the single feature level, but also consistent with the true category pattern at the joint feature distribution level.
[0040] Specifically, the CTabDDPM model aims to alleviate class imbalance through data augmentation, targeting minority class fault data generation, including the model training phase and the sampling and generation phase; among which, During the model training phase, labeled training data is constructed based on the forward diffusion process, and a denoising neural network ε is trained based on the labeled training data. θ ; Denoising Neural Network ε θ The learning basis is the noisy data x t The time step t and the conditional vector (class label y) are used to predict the noise injected during forward diffusion.
[0041] Preferably, in this embodiment of the invention, a multilayer perceptron (MLP) is used as the denoising neural network.
[0042] The sampling generation stage utilizes a trained denoising neural network ε θ Input the target condition vector (i.e., the fault category label y to be generated) target New high-quality optical network device fault data is generated through an iterative reverse denoising process.
[0043] Specifically, this embodiment of the invention uses numerical table features for verification, and the training loss function of the CTabDDPM model includes:
[0044] The sampling and generation process of the CTabDDPM model includes:
[0045]
[0046] In the formula, ε represents the noise that needs to be estimated and trained using a diffusion model. This represents the noise predicted by the denoising neural network at time step t; t This is a hyperparameter, preferably a very small value; α tIt is the predefined variance at time step t, z represents standard Gaussian noise, x0 is the original sample, x t It is the noisy data after t-step diffusion, x t-1 It is the denoised data from step t-1.
[0047] To ensure physical consistency, this invention introduces rule constraints and consistency screening after generation, specifically including: performing threshold interval constraints and correlation verification on parameter combinations, and retaining sample data that meet the conditions, so that the samples maintain the physical rationality of feature combinations under category condition constraints.
[0048] Specifically, the multidimensional evaluation system includes: The similarity of distributions is assessed based on the maximum mean difference, Wasserstein distance, and L2 distance. The authenticity of the structure is assessed by using complementary criteria based on Pearson correlation coefficient, mutual information, and adversarial evaluation. The utility of the task is evaluated based on a combination of accuracy, F1 score, false negative rate, and false positive rate.
[0049] This invention can be further described as a hierarchical, indicator-driven evaluation paradigm. Its goal is not merely to verify the apparent similarity of generated samples, but to establish a progressive chain of argumentation from statistical consistency and structural consistency to task effectiveness, thereby supporting the verifiability and transferability of the technical solution. Specifically, the evaluation process follows a three-stage criterion: first, determining the homology of distribution between the generated domain and the real domain; second, determining the fidelity of the multivariate dependency structure; and finally, determining whether the first two can translate into a substantial improvement in diagnostic performance.
[0050] At the statistical distribution level, the maximum mean discrepancy (MMD), Wasserstein distance (WD), and L2 distance (L2) are used for joint characterization. MMD measures the mean embedding bias of two sets of samples in the reproducing kernel Hilbert space, essentially reflecting the overall alignment at the high-dimensional distribution level. WD characterizes the minimum cost required to "transfer" one distribution to another, exhibiting high sensitivity to distribution support set misalignment and geometric differences. L2 measures the overall Euclidean shift in the feature space, reflecting the magnitude of global numerical drift. Lower values for these three values generally indicate a smaller statistical bias between the generated and true distributions.
[0051] In the structure fidelity layer, Pearson correlation coefficient, mutual information, and adversarial evaluation constitute complementary criteria. The Pearson correlation coefficient characterizes the strength of linear dependencies between variables and is suitable for verifying whether the first-order correlation structure is preserved. Mutual information measures statistical dependencies of any form and can supplement information on nonlinear relationships. The adversarial evaluation indirectly measures the difference in joint structure between real and generated samples by discriminating the separability of the model; the more difficult the discriminating process, the higher the degree of structural reproducibility. Thus, structural consistency can be cross-validated from three dimensions: linearity, nonlinearity, and separability.
[0052] At the task utility layer, a combination of accuracy, F1 score, false negative rate, and false positive rate is used for evaluation to avoid biased interpretations by a single indicator. Accuracy reflects the overall proportion of correct classifications; the F1 score, as a harmonic statistic of precision and recall, is more suitable for comprehensive performance characterization in class imbalance scenarios; the false negative rate measures the probability of missed detections, corresponding to the risk of unidentified faults; the false positive rate measures the probability of incorrect detections, corresponding to false alarms and operational resource consumption. If accuracy and F1 score improve simultaneously while the false negative rate and false positive rate decrease simultaneously, it can be considered that the improvement in generation quality has been effectively mapped to the optimization of the discrimination boundary.
[0053] Preferably, the method provided in this embodiment of the invention further includes: jointly determining a preset ratio under reliability constraints and computing resource constraints.
[0054] Specifically, the conclusion of this invention regarding the augmentation ratio (i.e., the preset ratio) can be summarized as a sample complexity proposition concerning the stability of minority class learning: In highly imbalanced diagnostic tasks, the convergence of the minority class decision boundary is not determined solely by the number of samples, but by the effective coverage; where effective coverage is a coupling function of sample quality and sample size. Specifically, when there is a systematic deviation between the generated samples and the true minority class distribution, simply increasing the augmentation amount will amplify distribution noise and weaken boundary discernibility; when the generated quality meets the distribution consistency requirement but the sample size is still below the critical coverage level, the minority class manifold support is insufficient, and the classifier still struggles to form a stable boundary. Based on the above mechanism, under the dataset and parameter settings of this embodiment, a performance inflection point occurs around 70%, such as... Figure 2 As shown, after exceeding this critical interval, performance enters a stable gain phase as the enhancement ratio increases, manifested as a continuous improvement in minority class recognition capability and a simultaneous decrease in false positives and false negatives. This result indicates that the enhancement ratio should be modeled as a control variable of equal criticality to the generative model structure. In engineering deployments, a range-based optimization strategy should be adopted, that is, a feasible ratio range should be jointly determined under reliability and computational resource constraints, rather than a pre-set single-point fixed ratio.
[0055] As described above, the embodiments of the present invention provide a data enhancement method for optical network fault diagnosis based on CTabDDPM, which has the following advantages compared with the prior art: (1) Improved fault identification capability under extreme class imbalance: This invention introduces fault category information into the diffusion generation process, transforming the generation objective from overall distribution learning to conditional distribution learning oriented towards specific categories. This mechanism can provide targeted compensation for a minority of fault samples, reshape the category structure of the training data, reduce the model's excessive dependence on the majority class samples, and thus improve the sufficiency of representation and identification stability of a minority of faults.
[0056] (2) Improving the physical fidelity and statistical reliability of synthetic data: This invention learns the joint variation law of multiple monitoring parameters of optical network in the stepwise denoising modeling, so that the synthetic samples not only closely resemble real data at the single variable level, but also maintain the coupling relationship and related structure between parameters. Since the generation process is subject to physical consistency constraints, it can effectively reduce the risk of introducing samples that do not conform to the real mechanism and improve the quality of training data.
[0057] (3) Improve the effectiveness of augmented data for actual diagnostic tasks: This invention combines conditional control with step-by-step denoising to ensure that the generated samples are consistent with real fault samples in terms of category semantics, statistical distribution, and structural relationships. As a result, augmented data can participate more effectively in classification boundary learning, reduce misleading training caused by low-quality samples, and enhance the generalization ability and robustness of the diagnostic model in complex scenarios.
[0058] (4) Establishing verifiable criteria for judging the quality of data augmentation: This invention incorporates distribution similarity, structural fidelity, and task utility into a unified evaluation approach, so that the effectiveness of augmented data no longer depends on a single indicator, but is comprehensively verified based on multidimensional consistency. This criterion helps to stably identify high-quality synthetic samples and improve the reliability and reusability of data augmentation strategies.
[0059] Figure 3 This is a schematic diagram of a data enhancement system for optical network fault diagnosis based on CTabDDPM, according to an embodiment of the present invention. Figure 3 As shown, the system includes: a preprocessing module 10, a model building module 20, a data augmentation module 30, a data validation module 40, and a data mixing module 50.
[0060] Specifically, the preprocessing module 10 is used to preprocess the initial dataset consisting of performance parameters characterizing the operating status of optical network devices to obtain the initial input feature dataset. Model building module 20 is used to inject fault labels as conditional vectors into the reverse denoising link of the TabDDPM model to build the CTabDDPM model. The data augmentation module 30 is used to augment the fault data in the initial input feature dataset based on the CTabDDPM model, and introduce rule constraints and consistency screening to obtain the augmented dataset. The data validation module 40 is used to construct a multi-dimensional evaluation system based on distribution similarity, structural authenticity and task utility, and to validate the augmented dataset based on the multi-dimensional evaluation system to obtain the validated augmented dataset. The data mixing module 50 is used to mix the validated augmented dataset with the initial input feature dataset according to a preset ratio to construct target training sets with different balances.
[0061] Specifically, such as Figure 3 As shown, the system also includes a classification and recognition module 60, which is used to train a fault diagnosis model based on the target training set and to classify and identify the fault status of optical network devices.
[0062] The present invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided in the embodiments of the present invention.
[0063] The present invention also provides a computer-readable storage medium storing program code, which can be called by a processor to execute the method provided in the embodiments of the present invention.
[0064] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0065] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0066] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.
[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0068] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0069] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0070] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A data augmentation method for optical network fault diagnosis based on CTabDDPM, characterized in that, The method includes: The initial dataset, consisting of performance parameters characterizing the operating status of optical network devices, is preprocessed to obtain the initial input feature dataset. The fault labels are injected as conditional vectors into the inverse denoising link of the TabDDPM model to construct the CTabDDPM model. The fault data in the initial input feature dataset is enhanced based on the CTabDDPM model, and rule constraints and consistency filtering are introduced to obtain the enhanced dataset. A multidimensional evaluation system is constructed based on distribution similarity, structural authenticity and task utility, and the augmented dataset is validated based on the multidimensional evaluation system to obtain the validated augmented dataset. The validated augmented dataset is mixed with the initial input feature dataset at a preset ratio to construct target training sets with different balances.
2. The method according to claim 1, characterized in that, The method further includes: A fault diagnosis model is trained based on the target training set to classify and identify the fault status of optical network devices.
3. The method according to claim 1, characterized in that, The CTabDDPM model includes a model training phase and a sampling generation phase; wherein... The model training phase constructs labeled training data based on the forward diffusion process, and trains a denoising neural network based on the labeled training data; the denoising neural network learns by predicting the noise injected in the forward diffusion based on the noise-added data, the time step, and the conditional vector. The sampling generation stage utilizes a trained denoising neural network, inputs a target conditional vector, and generates new high-quality optical network device fault data through an iterative reverse denoising process. The training loss function of the CTabDDPM model includes: ； The sampling and generation process of the CTabDDPM model includes: ； ； In the formula, ε represents the noise that needs to be estimated and trained using a diffusion model. This represents the noise predicted by the denoising neural network at time step t. t For hyperparameters, α t It is the predefined variance at time step t, z represents standard Gaussian noise, x0 is the original sample, x t It is the noisy data after t-step diffusion, x t-1 It is the denoised data from step t-1.
4. The method according to claim 1, characterized in that, The rule constraints and consistency screening include: applying threshold range constraints and correlation checks to parameter combinations, and retaining sample data that meet the conditions.
5. The method according to claim 1, characterized in that, The multidimensional evaluation system includes: The similarity of distributions is assessed based on the maximum mean difference, Wasserstein distance, and L2 distance. The authenticity of the structure is assessed by using complementary criteria based on Pearson correlation coefficient, mutual information, and adversarial evaluation. The utility of the task is evaluated based on a combination of accuracy, F1 score, false negative rate, and false positive rate.
6. The method according to claim 1, characterized in that, The method further includes: jointly determining the preset ratio under reliability constraints and computing resource constraints.
7. A data enhancement system for optical network fault diagnosis based on CTabDDPM, characterized in that, This system is used to implement the CTabDDPM-based optical network fault diagnosis data enhancement method according to any one of claims 1-6; the system includes: a preprocessing module, a model building module, a data enhancement module, a data validation module, and a data mixing module; wherein, The preprocessing module is used to preprocess the initial dataset consisting of performance parameters characterizing the operating status of optical network devices to obtain the initial input feature dataset. The model building module is used to inject fault labels as conditional vectors into the reverse denoising link of the TabDDPM model to build the CTabDDPM model. The data augmentation module is used to augment the fault data in the initial input feature dataset based on the CTabDDPM model, and introduce rule constraints and consistency filtering to obtain the augmented dataset. The data verification module is used to construct a multi-dimensional evaluation system based on distribution similarity, structural authenticity and task utility, and to verify the augmented dataset based on the multi-dimensional evaluation system to obtain the verified augmented dataset. The data mixing module is used to mix the validated augmented dataset with the initial input feature dataset according to a preset ratio to construct target training sets with different balances.
8. The system according to claim 7, characterized in that, The system also includes a classification and identification module, used to train a fault diagnosis model based on the target training set, and to classify and identify the fault status of optical network devices.
9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1-6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 6.