Industrial time sequence event analysis method and device based on causal regularization and medium

Abstract

The invention discloses an industrial time sequence event analysis method and device based on causal regularization and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the multi-scale time sequence feature coding of multivariate time sequence data, and obtaining time sequence feature data; performing potential vector reconstruction of probability distribution on the time sequence feature data to determine reconstruction loss and distance measurement loss; based on the prior causal graph, obtaining a causal consistency loss function through acyclic constraint balance analysis; performing hyper-parameter consistency processing on the causal consistency loss function to determine an overall loss function and obtain a minimum overall loss; a small number of samples are finely adjusted to model convergence, and a CT-VAE model of the implicit label is determined; and inputting the data window into the model to obtain an event analysis result and an event diagnosis report. According to the method, the technical problems that in the prior art, industrial time sequence event analysis robustness is poor, the identification error is high, and implicit key event identification cannot be met are solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an industrial time series event analysis method, device, and medium based on causal regularization. Background Art

[0002] In modern industrial systems, such as mining, power generation, and chemical manufacturing, production processes are continuously monitored by numerous sensors, generating massive amounts of high-dimensional, multivariate time series data. Accurately and promptly identifying critical events within this complex, multivariate time series data is crucial for ensuring production safety and improving operational efficiency. Critical events include both planned routine operations (such as switching a mine's main ventilation fan in reverse) and unplanned equipment failures or abnormal operating conditions.

[0003] Accurately identifying critical events is often impossible with a single sensor. Confirmation requires analyzing the complex, synchronously evolving patterns among multiple variables. First, because the relationships between variables are not simply statistical correlations but rather involve physical causal chains, traditional time series models often overlook this underlying causal structure. Inadequate integration of multidimensional time series data and causal dependencies results in extremely poor model robustness. Second, the frequency of critical production events, such as planned fan failures or unexpected failures, is extremely small relative to the system's total operating time. Due to the sparsity and class imbalance of critical production events, standard supervised learning models are heavily biased toward predicting all situations as normal operation, increasing the difficulty of accurately identifying abnormal events. Third, existing historical maintenance records and critical event logs may contain omissions or errors, resulting in noisy and incomplete labels for abnormal events. This label uncertainty makes relying solely on supervised learning approaches unfeasible. Fourth, some key events, such as unsuccessful reversing, are defined by the failure of a series of expected actions. These key events are the absence or deviation of the expected pattern, rather than a simple numerical overrun. In other words, there are no explicitly labeled (implicit) key events, and existing technologies cannot meet the requirements for identifying implicit key events. Summary of the Invention

[0004] The embodiments of the present application provide an industrial time series event analysis method, device, and medium based on causal regularization, which solve the technical problems in the prior art of industrial time series event analysis, such as poor robustness, high recognition error, reliance on supervised learning methods, and inability to meet the requirements of implicit key event identification.

[0005] In a first aspect, an embodiment of the present application provides an industrial time series event analysis method based on causal regularization, characterized in that the method includes: obtaining multivariate time series data, and performing multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; reconstructing the potential vector of the probability distribution of the time series feature data to determine the reconstruction loss and the distance metric loss; constructing a priori causal graph, and based on the priori causal graph, obtaining a causal consistency loss function through acyclic constraint balance analysis; performing hyperparameter consistency processing on the causal consistency loss function according to the reconstruction loss and the distance metric loss to determine the overall loss function, and performing unsupervised training on the overall loss function to minimize the overall loss; obtaining latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tuning the model with a small number of samples until the model converges, and determining the CT-VAE model with implicit labels; inputting the data window into the CT-VAE model with implicit labels, and performing event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports.

[0006] In one implementation of the present application, multi-scale time series feature encoding is performed on multivariate time series data to obtain time series feature data, specifically including: information extraction on the multivariate time series data to obtain dense information time series feature data; short-scale causal dilation convolution is performed on the dense information time series feature data to determine transient time series feature data; long-scale causal dilation convolution is performed on the dense information time series feature data to determine continuous time series feature data; and time series feature data is obtained based on the transient time series feature data and the continuous time series feature data.

[0007] In one implementation of the present application, the time series feature data is reconstructed with a latent vector of a probability distribution to determine the reconstruction loss and the distance measurement loss, specifically including: predicting the latent variables of the time series feature data with a Gaussian distribution to obtain the vectors corresponding to the latent variables; wherein the vectors corresponding to the latent variables include: a mean vector and a logarithmic variance vector; performing parameter resampling on the vectors corresponding to the latent variables to determine the latent vector; reconstructing the latent vector into the original input data window to obtain the reconstructed time series feature data; determining the reconstruction loss by calculating the mean square error function based on the time series feature data and the reconstructed time series feature data; and determining the distance measurement loss by analyzing the distribution differences based on the time series feature data and the reconstructed time series feature data.

[0008] In one implementation of the present application, based on the prior causal graph, a causal consistency loss function is obtained through acyclic constraint balance analysis, specifically including: knowledge encoding of the prior causal graph to determine the mandatory edges of the prior causal graph; based on the mandatory edges of the prior causal graph, an updated documented causal graph is obtained through temporal causal discovery analysis; the prior causal graph and the updated documented causal graph are integrated to determine the total causal graph; based on the total causal graph, a difference loss term is obtained through implicit causal graph difference analysis; and the difference loss term is subjected to acyclic constraint balance to obtain the causal consistency loss function.

[0009] In one implementation of the present application, the causal consistency loss function is subjected to hyperparameter consistency processing based on the reconstruction loss and the distance metric loss to determine the overall loss function, specifically including: determining the first hyperparameter based on the distance metric loss; determining the second hyperparameter based on the causal consistency loss function; and performing consistent configuration of the first hyperparameter and the second hyperparameter to determine the overall loss function.

[0010] In one implementation of the present application, based on the CT-VAE model corresponding to the latent representation data and the minimized overall loss, a small number of samples are fine-tuned until the model converges to determine the CT-VAE model of the implicit label, specifically including: performing confidence sample discrimination on the latent representation data to determine a high-confidence event sample set; wherein, the confidence sample discrimination includes: sample discrimination, confidence discrimination, and sample confidence threshold filtering; based on the high-confidence event sample set, abnormal event data clusters are deduced in the latent space to obtain abnormal cluster clustering; according to the abnormal cluster clustering, the CT-VAE model is trained until the model converges to determine the CT-VAE model of the implicit label.

[0011] In one implementation of the present application, event diagnostic reasoning is performed on a data window to obtain event analysis results and an event diagnosis report, specifically including: calculating a reconstruction score on the data window to obtain a reconstruction error coefficient; calculating a potential deviation on the data window to determine the deviation distance between the potential vector corresponding to the data window and the center of the normal data distribution; calculating a causal violation score on the data window to obtain the score of the causal consistency loss item on the data window; performing a weighted summation of the scores of the reconstruction error coefficient, the deviation distance, and the causal consistency loss item on the data window to determine the total anomaly score; and obtaining an event analysis result and an event diagnosis report based on the total anomaly score through event detection and classification.

[0012] In one implementation of the present application, after inputting the data window into the implicitly labeled CT-VAE model and performing event diagnostic reasoning on the data window to obtain event analysis results and event diagnosis reports, the method also includes: identifying unknown events in the event analysis results and marking the identified unknown events; visualizing the latent space trajectory of the event diagnosis report to obtain causal graph deviation analysis to determine a visualized event diagnosis report.

[0013] In a second aspect, an embodiment of the present application further provides an industrial time series event analysis method and device based on causal regularization, characterized in that the device comprises: at least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so as to enable the at least one processor to: obtain multivariate time series data, and perform multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; reconstruct the potential vector of the probability distribution of the time series feature data to determine the reconstruction loss and the distance metric loss; construct a priori causal graph, and based on the prior causal graph , through acyclic constraint balance analysis, the causal consistency loss function is obtained; according to the reconstruction loss and distance metric loss, the causal consistency loss function is subjected to hyperparameter consistency processing to determine the overall loss function, and the overall loss function is unsupervisedly trained to minimize the overall loss; the latent representation data is obtained, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, a small number of samples are fine-tuned until the model converges, and the CT-VAE model of the implicit label is determined; the data window is input into the CT-VAE model of the implicit label, and event diagnosis reasoning is performed on the data window to obtain event analysis results and event diagnosis reports.

[0014] In a third aspect, an embodiment of the present application further provides a non-volatile computer storage medium for an industrial time series event analysis method based on causal regularization, which stores computer executable instructions, characterized in that the computer executable instructions are set to: obtain multivariate time series data, and perform multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; reconstruct the potential vector of the probability distribution of the time series feature data to determine the reconstruction loss and the distance metric loss; construct a priori causal graph, and based on the priori causal graph, obtain a causal consistency loss function through acyclic constraint balance analysis; perform hyperparameter consistency processing on the causal consistency loss function according to the reconstruction loss and the distance metric loss to determine the overall loss function, and perform unsupervised training on the overall loss function to minimize the overall loss; obtain potential representation data, and based on the potential representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tune the model with a small number of samples until the model converges, and determine the CT-VAE model with implicit labels; input the data window into the CT-VAE model with implicit labels, and perform event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports.

[0015] The embodiments of the present application provide an industrial time series event analysis method, device and medium based on causal regularization. Through multi-scale time series feature encoding, probabilistic latent space analysis and decoding, causal graph regularization of prior causal graphs, CT-VAE model to obtain feature training and event diagnostic reasoning through causal graph deviation analysis, the present application solves the technical problems of poor robustness, high recognition error, reliance on supervised learning method and inability to meet the requirements of implicit key event identification in the prior art of industrial time series event analysis, realizes accurate identification of key production events, improves the model interpretability and model generalization ability of industrial time series event analysis, reduces the false alarm rate and missed alarm rate of industrial abnormal events, and optimizes the classification detection effect of known time and location time. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of an industrial time series event analysis method based on causal regularization provided in an embodiment of the present application; Figure 2 An architecture diagram of an industrial time series event analysis system based on causal regularization provided in an embodiment of the present application; Figure 3 A diagram of the architecture of a causal temporal event mining model provided in an embodiment of the present application; Figure 4 A CT-VAE model training flowchart provided in an embodiment of the present application; Figure 5 An event reasoning and diagnosis flow chart provided in an embodiment of the present application; Figure 6 A diagram for identifying a downtime event provided in an embodiment of the present application; Figure 7 A continuous downtime event identification diagram provided in an embodiment of the present application; Figure 8 An unsuccessful downtime event identification diagram provided in an embodiment of the present application; Figure 9 A schematic diagram of the internal structure of an industrial time series event analysis device based on causal regularization provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0018] The embodiments of the present application provide an industrial time series event analysis method, device and medium based on causal regularization. Through multi-scale time series feature encoding, probabilistic latent space analysis and decoding, causal graph regularization of prior causal graphs, CT-VAE model to obtain feature training and event diagnostic reasoning through causal graph deviation analysis, the present application solves the technical problems of poor robustness, high recognition error, reliance on supervised learning method and inability to meet the requirements of implicit key event identification in the prior art of industrial time series event analysis, realizes accurate identification of key production events, improves the model interpretability and model generalization ability of industrial time series event analysis, reduces the false alarm rate and missed alarm rate of industrial abnormal events, and optimizes the classification detection effect of known time and location time.

[0019] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0020] Figure 1 This is a flow chart of an industrial time series event analysis method based on causal regularization provided in an embodiment of the present application. Figure 1 As shown, an industrial time series event analysis method based on causal regularization provided in an embodiment of the present application specifically includes the following steps: Step 101: Acquire multivariate time series data, and perform multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data.

[0021] For example, the encoder is used to extract the multivariate time series data from the input window Extract an information-dense temporal feature embedding from In order to capture event features at different time scales simultaneously, the present invention preferably adopts a multi-scale temporal convolutional network (TCN) as the main structure of the encoder. Causal convolution ensures that at the prediction time point When outputting, only The core of TCN is the dilated causal convolution. This incorporates dilated convolution into the causal convolution. By inserting holes between the elements of the convolution kernel, the causal dilation convolution exponentially expands the receptive field, enabling it to capture very long-term dependencies with fewer layers. This avoids the vanishing / exploding gradient problem common in recurrent neural networks (RNNs), and convolution operations can be computed in parallel, improving model training efficiency.

[0022] Specifically, multi-scale time series feature encoding is performed on multivariate time series data to obtain time series feature data, including: information extraction on the multivariate time series data to obtain dense information time series feature data; short-scale causal dilation convolution is performed on the dense information time series feature data to determine transient time series feature data; long-scale causal dilation convolution is performed on the dense information time series feature data to determine continuous time series feature data; and time series feature data is obtained based on the transient time series feature data and the continuous time series feature data.

[0023] Figure 2 An architecture diagram of an industrial time series event analysis system based on causal regularization provided in an embodiment of the present application.

[0024] In one embodiment, the data acquisition and preprocessing module is responsible for acquiring multivariate time series data in real time or in batches from multiple sensors at industrial sites (measuring power, current, wind speed, pressure, vibration, temperature, etc.). The raw data is then cleaned, aligned, missing value processed, and normalized. A sliding window technique is then used to segment the continuous data stream into fixed-length, overlapping data windows, which serve as input samples for subsequent models.

[0025] The causal temporal model engine module is responsible for building, training, and storing the Causal-Temporal Variational Autoencoder (CT-VAE) model proposed in this invention.

[0026] The event reasoning and diagnosis presentation module is responsible for invoking the trained CT-VAE model to perform real-time analysis on new data windows. When an event is detected, the module not only issues an alarm but also generates a multi-dimensional diagnostic report, which is presented to the operator or engineer through a user interface to assist with root cause analysis.

[0027] The multi-scale design is achieved by connecting multiple TCN modules in parallel with different dilation factor sequences: The short-scale TCN module uses a smaller convolution kernel and a smaller dilation rate (such as d = 1, 2, 4), allowing it to focus on capturing fast, transient event features, such as the current spike at the moment of motor startup or the vibration shock at the initial stage of equipment failure.

[0028] The long-scale TCN module uses a larger convolution kernel and a larger dilation rate (such as d = 8, 16, 32) to have a wider receptive field and can capture the overall outline of events that last longer, such as a complete machine downtime process that lasts for several minutes. The output features of all scale TCN modules are finally concatenated or added to form the final temporal feature embedding. , providing high-quality input for subsequent probabilistic modeling and causal analysis.

[0029] Step 102: reconstruct the potential vector of the probability distribution of the time series feature data to determine the reconstruction loss and the distance metric loss.

[0030] For example, a variational autoencoder (VAE) framework is used to learn the probability distribution of the system's normal operating state. Unlike standard autoencoders, which deterministically map input to a single point, VAEs map input to a probability distribution, making them more robust to input noise and capable of generating new, normal-looking data by sampling from the latent space.

[0031] Specifically, the time series feature data is reconstructed with a latent vector of a probability distribution to determine the reconstruction loss and the distance metric loss, including: predicting the latent variables of the time series feature data with a Gaussian distribution to obtain the vectors corresponding to the latent variables; wherein the vectors corresponding to the latent variables include: a mean vector and a logarithmic variance vector; re-sampling the vectors corresponding to the latent variables to determine the latent vector; reconstructing the latent vector into the original input data window to obtain the reconstructed time series feature data; determining the reconstruction loss by calculating the mean square error function based on the time series feature data and the reconstructed time series feature data; and determining the distance metric loss by analyzing the distribution differences based on the time series feature data and the reconstructed time series feature data.

[0032] In one embodiment, first, the temporal features output by the TCN encoder are embedded is fed into two separate fully connected layers (or 1x1 convolutional layers) to predict the latent variables The mean vector of the Gaussian distribution it obeys and the logarithmic variance vector .

[0033] In order to make the sampling process differentiable and thus enable end-to-end gradient backpropagation training, the present invention adopts the reparameterization technique. By formula Sampling is performed, where is from the standard normal distribution Random noise sampled in , represents element-wise multiplication, is the distribution parameter, is the standard deviation, which can optimize gradient descent.

[0034] Then decoding is performed, and the latent vector is received by a decoder with a structure symmetrical to the TCN encoder , and try to reconstruct the original input data window The decoder usually uses transposed convolution to achieve upsampling.

[0035] Finally, for the mean square error function calculation and distribution difference analysis, the model basic loss function is the standard VAE loss, and the loss function is constructed as follows: Reconstruction loss , which measures the original input and reconstructed output The difference between the two can be calculated using the mean squared error (MSE) function.

[0036] Distance metric loss , which measures the potential distribution produced by the encoder With the standard normal prior distribution The difference between.

[0037] It should be noted that the distance metric loss This term is a regularization term that promotes the structure of the latent space. The KL divergence can be selected as the metric loss function.

[0038] Step 103: Construct a priori causal graph, and based on the priori causal graph, obtain a causal consistency loss function through acyclic constraint balance analysis.

[0039] For example, through a differentiable causal consistency loss function, the model is forced to ensure that its internal feature representation conforms to the preset causal structure representing physical laws while learning data patterns. This ensures that the feature representation learned by the model is not only statistically valid, but also physically meaningful, thereby injecting domain knowledge directly into the model architecture. It also solves the technical problem that traditional deep learning models run as "black boxes", have opaque decision-making logic, and cannot provide explanations that conform to physical intuition.

[0040] Specifically, based on the prior causal graph, a causal consistency loss function is obtained through acyclic constraint balance analysis, including: knowledge encoding of the prior causal graph to determine the mandatory edges of the prior causal graph; based on the mandatory edges of the prior causal graph, an updated documented causal graph is obtained through temporal causal discovery analysis; the prior causal graph and the updated documented causal graph are integrated to determine the total causal graph; based on the total causal graph, a difference loss term is obtained through implicit causal graph difference analysis; and acyclic constraint balance is performed on the difference loss term to obtain the causal consistency loss function.

[0041] Figure 3 This is a diagram of the architecture of a causal temporal event mining model provided in an embodiment of the present application.

[0042] In one embodiment, first, a priori causal graph is constructed. Before model training begins, a directed acyclic graph (DAG) needs to be constructed. , where the node set Corresponding to each variable (i.e. sensor) of the multivariate time series, the edge set Represents the direct causal relationship between variables. The construction of the prior causal graph adopts a "human-computer collaboration" model, integrating two sources of knowledge: First, deterministic knowledge from domain experts is encoded as mandatory edges in the graph. This knowledge can originate from physical laws (e.g., "increased motor power leads to increased fan speed"), equipment design (e.g., "vibration of fan A" is related to "power of fan A"), or operating procedures (e.g., "opening the damper of the backup fan" is a prerequisite for "starting the backup fan"). These edges have the highest confidence (equivalent to the underlying knowledge).

[0043] Second, apply temporal causal discovery algorithms (such as DYNOTEARS) to the training data set to mine potential causal relationships that may not be documented, and discover hidden association patterns in the data that experts may overlook.

[0044] Integrate the two sources of graphs to form the final During the integration process, when expert knowledge conflicts with the results of data discovery, expert knowledge prevails to ensure the accuracy and physical rationality of the prior causal diagram.

[0045] Then, define the causal consistency loss function Assume that the first layer weight matrix of the TCN encoder is A linear mapping is implicitly learned, which should approximate the linear structural equation model of the monitored system. The absolute value of Extract an adjacency matrix , whose elements Represents the variables learned by the model For variables The strength of the causal influence.

[0046] Introducing a new loss term , used to penalize the implicit causal graph learned by the model (given by the adjacency matrix Representative) and prior causal diagram Using the idea of ​​algorithms such as NOTERS, the acyclicity constraint of the graph is expressed as a differentiable function .

[0047] in, is the number of variables, is the trace of the matrix.

[0048] This is explained by the following formula.

[0049] (1) in, A priori causal diagram The edge set of is the indicator function.

[0050] It should be noted that the first The penalty graph has cycles. The second term is the sparsity penalty, whose main components are those in the adjacency matrix. exists in but not in the edges in (i.e., the abnormal causal chain); is a hyperparameter used to balance the acyclicity constraint and the structural consistency with the prior graph.

[0051] Step 104: Based on the reconstruction loss and the distance metric loss, the causal consistency loss function is subjected to hyperparameter consistency processing to determine the overall loss function, and the overall loss function is subjected to unsupervised training to minimize the overall loss.

[0052] For example, in order to enable the causal graph regularized model to both reconstruct normal data and reproduce the solution of prior causal knowledge in its internal structure, this application improves the generalization ability of the model and the interpretability of its latent space through consistency processing of hyperparameters.

[0053] Specifically, based on the reconstruction loss and the distance metric loss, the causal consistency loss function is subjected to hyperparameter consistency processing to determine the overall loss function, including: determining a first hyperparameter based on the distance metric loss; determining a second hyperparameter based on the causal consistency loss function; and performing consistent configuration of the first hyperparameter and the second hyperparameter to determine the overall loss function.

[0054] In one embodiment, a model overall loss function is constructed, and the overall loss function of the model is explained by the following formula.

[0055] (2) in, and is a hyperparameter used to balance the reconstruction accuracy, smoothness of the latent space, and consistency with physical knowledge.

[0056] By minimizing the total loss function, the model is guided to find a solution that both reconstructs normal data well and reproduces prior causal knowledge in its internal structure. This transforms the model from a simple "pattern recognizer" into a simplified, data-driven "physics simulator." The latent space learned by the model is therefore more interpretable, and different latent dimensions can naturally correspond to different physical subsystems, achieving meaningful feature decoupling.

[0057] Step 105: Obtain the latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tune the model with a small number of samples until the model converges, and determine the CT-VAE model of the implicit label.

[0058] For example, existing historical maintenance records and key event logs may contain omissions or errors. For example, events such as "unsuccessful reversal" and other operational failures are likely not to be formally recorded. This uncertainty in labels makes it impossible to rely entirely on supervised learning methods, and weakly supervised or unsupervised paradigms that can handle imperfect labels must be adopted. Due to the unreliability of historical event labels, this application uses the CT-VAE model and adopts a progressive strategy from unsupervised to weakly supervised. By fine-tuning a small number of samples until the model converges, the CT-VAE model with implicit labels is determined, which solves the label noise and event sparsity problems that are common in industrial data.

[0059] Specifically, based on the CT-VAE model corresponding to the latent representation data and minimizing the overall loss, a small number of samples are fine-tuned until the model converges to determine the CT-VAE model of implicit labels, including: performing confidence sample discrimination on the latent representation data to determine a high-confidence event sample set; wherein, confidence sample discrimination includes: sample discrimination, confidence discrimination, and sample confidence threshold filtering; based on the high-confidence event sample set, abnormal event data clusters are deduced in the latent space to obtain abnormal cluster clustering; according to the abnormal cluster clustering, the CT-VAE model is trained until the model converges to determine the CT-VAE model of implicit labels.

[0060] Figure 4 A CT-VAE model training flowchart provided in an embodiment of the present application.

[0061] In one embodiment, it is assumed that most of the data in the dataset represents the normal operation of the system. Under this assumption, the CT-VAE model is trained unsupervised on the entire training set. The training goal is to minimize the total loss function defined in the second part. .

[0062] The purpose of unsupervised training is to allow the model to first learn the basic dynamic characteristics and inherent physical causal structure of the system under normal operating conditions, build a deep understanding of "what is normal", and lay the foundation for subsequent event recognition.

[0063] Incomplete and noisy historical event logs in industrial scenarios are considered a "weakly labeled set." To filter out truly valuable high-confidence samples from this weakly labeled set, this application uses a filtering mechanism inspired by adversarial training to filter out high-confidence samples.

[0064] First, a simple discriminator network is trained whose task is to distinguish the latent representations of two classes of samples. Among them, one is the potential representation from a large amount of unlabeled data (most likely normal data), and the other is the potential representation from log events.

[0065] During the training process, samples that are marked as events but whose potential representation z is difficult to distinguish from normal data by the discriminator are likely to be normal working conditions that are incorrectly labeled, or minor anomalies with very unclear features. The confidence of these samples is low.

[0066] Then, the output of the discriminator is used to evaluate the abnormality or credibility of each log event sample, and those samples with scores below a certain threshold are filtered out, so that a small but high-quality set of high-confidence event samples can be obtained.

[0067] Finally, the pre-trained CT-VAE model is fine-tuned based on a set of high-confidence event samples. The training goal is to add a discriminative or contrastive loss term to the original loss function.

[0068] This loss term is derived from the idea of ​​open-set recognition, whose goal is to push the representation of known, labeled abnormal events (such as "unsuccessful machine downtime") away from the center of the normal data cluster (usually the origin) in the latent space.

[0069] Furthermore, a center loss or contrast loss can be added to penalize the fact that the potential representation of known abnormal samples is too close to the center of the normal data cluster.

[0070] After fine-tuning, the model can not only perceive "anomalies" (i.e. deviations from the normal cluster), but also clearly "recognize" a specific, known type of event, which forms its own unique cluster in the latent space, away from the normal cluster.

[0071] When the model discovers new, unknown anomalies in subsequent applications, these new samples can be added to the high-confidence event set after confirmation and labeling by human experts for the next round of iterative fine-tuning of the model.

[0072] Step 106: Input the data window into the implicitly labeled CT-VAE model, and perform event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports.

[0073] For example, the reconstruction method for event diagnostic reasoning in the existing technology relies on a single reconstruction error and cannot achieve an accurate description of the event state. The present application performs event diagnostic reasoning through a combined anomaly score composed of three complementary parts, realizes the visualization of the multi-dimensional state of the event, and improves the richness of the system state description.

[0074] Specifically, event diagnostic reasoning is performed on the data window to obtain event analysis results and event diagnosis reports, including: calculating the reconstruction score of the data window to obtain the reconstruction error coefficient; calculating the potential deviation of the data window to determine the deviation distance between the potential vector corresponding to the data window and the center of the normal data distribution; calculating the causal violation score of the data window to obtain the score of the causal consistency loss item on the data window; taking a weighted sum of the scores of the reconstruction error coefficient, the deviation distance and the causal consistency loss item on the data window to determine the total anomaly score; and obtaining the event analysis results and event diagnosis report based on the total anomaly score through event detection and classification.

[0075] Furthermore, after inputting the data window into the implicitly labeled CT-VAE model and performing event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports, the method also includes: identifying unknown events in the event analysis results and marking the identified unknown events; visualizing the latent space trajectory of the event diagnosis report to obtain causal graph deviation analysis to determine the visualized event diagnosis report.

[0076] Figure 5 An event reasoning and diagnosis flowchart provided in an embodiment of the present application.

[0077] Figure 6 A crash event identification diagram provided in an embodiment of the present application.

[0078] Figure 7 A continuous downtime event identification diagram provided in an embodiment of the present application.

[0079] Figure 8 An unsuccessful downtime event identification diagram provided in an embodiment of the present application.

[0080] In one embodiment, when a trained CT-VAE model is deployed, for each new input data window , perform the following reasoning and diagnosis steps: First, reconstruct the score Calculate, calculate the reconstruction error ,High reconstruction error indicates that the current data pattern is not seen by the model in normal data, which is a traditional abnormal signal.

[0081] Potential bias Calculate, calculate the potential vector corresponding to the current data window The distance from the center of the normal data distribution (usually the origin). The larger the distance, the more the system state deviates from normal at the model's abstract level of understanding.

[0082] Causal violation score Calculation, i.e., causal consistency loss term The value in the current data window. A high score indicates that the system's current behavior violates known physical causal laws and can capture some situations where the reconstruction error may not be obvious but is physically and logically incorrect.

[0083] The above three scores can be weighted and summed to get a total anomaly score. The combination of different scores can provide more detailed insights, for example, high But low May indicate a very subtle, early-stage fault that violates physical logic Then, when the total anomaly score exceeds a preset threshold that is adjusted by the validation set, the system determines that an event has occurred. to classify.

[0084] For known event classification, if If the position of is close to the cluster center of a known event category learned in the fine-tuning phase (such as "unsuccessful machine downtime"), it is classified as that specific event.

[0085] For unknown event recognition, if If the location of an event is significantly far away from the center of the normal data cluster but not close to any known event cluster, it will be marked as a "new or unknown anomaly" and a prompt will be given that human expert intervention is needed for analysis.

[0086] For diagnostic report generation, reconstruction error analysis is first performed. By visualizing the point-by-point reconstruction error of each sensor channel, the system can immediately highlight the variable with the largest deviation from the normal mode, providing the first clue for problem location.

[0087] Draw the potential vector representing the state of the system during the event The motion trajectory in two-dimensional or three-dimensional reduced-dimensional space can be used to judge the dynamic characteristics of the event by observing the trajectory shape (such as sharp deviation, slow drift, or oscillation between different states) and compare it with the characteristic trajectory of known events.

[0088] Then, we use feature attribution methods such as SHAP (SHapley Additive exPlanations) or input gradients to analyze the source of the total anomaly score.

[0089] Finally, a causal graph deviation analysis is performed to convert the dynamic causal relationships inferred by the model within the event window (i.e., the adjacency matrix ) and the benchmark prior causal graph Comparisons are performed and significant deviations are highlighted for visualizing event diagnostic reports.

[0090] Furthermore, these deviations include: Broken Links, where the expected causal relationship fails to materialize. For example, the diagnostic report indicates that fan B has reached its rated power, but this does not result in the expected increase in air volume. Spurious Links, where new, unexpected causal relationships emerge. For example, ambient temperature begins to significantly affect bearing temperature, a relationship that does not exist under normal operating conditions.

[0091] The dotted line indicates that the fan equipment is not running, the solid line indicates that it is in operation, and the gray area indicates the detected part.

[0092] The above is an embodiment of the method proposed in this application. Based on the same inventive concept, this application embodiment also provides an industrial time series event analysis device based on causal regularization, whose structure is as follows: Figure 9 shown.

[0093] Figure 9 This is a schematic diagram of the internal structure of an industrial time series event analysis device based on causal regularization provided in an embodiment of the present application. Figure 9 As shown, the equipment includes: at least one processor 901; and a memory 902 communicatively connected to the at least one processor; The memory 902 stores instructions that can be executed by at least one processor. The instructions are executed by the at least one processor 901 to enable the at least one processor 901 to: Acquire multivariate time series data and perform multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; reconstruct the latent vector of the probability distribution of the time series feature data to determine the reconstruction loss and distance metric loss; construct a priori causal graph, and based on the prior causal graph, obtain the causal consistency loss function through acyclic constraint balance analysis; perform hyperparameter consistency processing on the causal consistency loss function according to the reconstruction loss and distance metric loss to determine the overall loss function, and perform unsupervised training on the overall loss function to minimize the overall loss; obtain latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tune the model with a small number of samples until the model converges, and determine the CT-VAE model with implicit labels; input the data window into the CT-VAE model with implicit labels, and perform event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports.

[0094] Some embodiments of the present application provide corresponding Figure 1 A non-volatile computer storage medium for industrial time series event analysis based on causal regularization stores computer executable instructions, wherein the computer executable instructions are set as follows: Acquire multivariate time series data and perform multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; reconstruct the latent vector of the probability distribution of the time series feature data to determine the reconstruction loss and distance metric loss; construct a priori causal graph, and based on the prior causal graph, obtain the causal consistency loss function through acyclic constraint balance analysis; perform hyperparameter consistency processing on the causal consistency loss function according to the reconstruction loss and distance metric loss to determine the overall loss function, and perform unsupervised training on the overall loss function to minimize the overall loss; obtain latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tune the model with a small number of samples until the model converges, and determine the CT-VAE model with implicit labels; input the data window into the CT-VAE model with implicit labels, and perform event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports.

[0095] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences from the other embodiments. In particular, the IoT device and media embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.

[0096] The system and medium provided in the embodiments of the present application correspond one-to-one to the method. Therefore, the system and medium also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.

[0097] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0099] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0101] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0102] Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0103] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can be implemented using any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0104] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0105] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for analyzing industrial time series events based on causal regularization, characterized in that: The method comprises: Acquiring multivariate time series data, and performing multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; Reconstructing a potential vector of a probability distribution of the time series feature data to determine a reconstruction loss and a distance metric loss; Constructing a priori causal graph, and based on the priori causal graph, obtaining a causal consistency loss function through acyclic constraint balance analysis; According to the reconstruction loss and the distance metric loss, performing hyperparameter consistency processing on the causal consistency loss function to determine an overall loss function, and performing unsupervised training on the overall loss function to minimize the overall loss; Obtaining latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tuning the model with a small number of samples until the model converges, thereby determining the CT-VAE model of the implicit label; The data window is input into the implicitly labeled CT-VAE model, and event diagnosis reasoning is performed on the data window to obtain event analysis results and event diagnosis reports.

2. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: Performing multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data specifically includes: performing information extraction on the multivariate time series data to obtain dense information time series feature data; Performing a short-scale causal dilation convolution on the dense information time series feature data to determine transient time series feature data; Performing a long-scale causal dilation convolution on the dense information time series feature data to determine continuous time series feature data; The time series feature data is obtained based on the transient time series feature data and the continuous time series feature data.

3. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: Reconstructing the potential vector of the probability distribution of the time series feature data to determine the reconstruction loss and the distance metric loss, specifically including: Performing Gaussian distribution latent variable prediction on the time series feature data to obtain a vector corresponding to the latent variable; wherein the vector corresponding to the latent variable includes: a mean vector and a logarithmic variance vector; Re-sampling the vector corresponding to the latent variable to determine a latent vector; Reconstructing the latent vector into an original input data window to obtain reconstructed time series feature data; Determine the reconstruction loss by calculating a mean square error function based on the time series feature data and the reconstructed time series feature data; The distance metric loss is determined by performing distribution difference analysis based on the time series feature data and the reconstructed time series feature data.

4. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: Based on the prior causal graph, a causal consistency loss function is obtained through acyclic constraint balance analysis, specifically including: performing knowledge encoding on the priori causal graph to determine mandatory edges of the priori causal graph; Based on the mandatory edges of the prior causal graph, an updated documented causal graph is obtained through temporal causal discovery analysis; Integrating the prior causal graph with the updated documented causal graph to determine a total causal graph; According to the total causal graph, a difference loss term is obtained through implicit causal graph difference analysis; The difference loss term is subjected to acyclic constraint balancing to obtain the causal consistency loss function.

5. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: According to the reconstruction loss and the distance metric loss, the causal consistency loss function is subjected to hyperparameter consistency processing to determine an overall loss function, specifically including: Determining a first hyperparameter based on the distance metric loss; Determining a second hyperparameter according to the causal consistency loss function; The first hyperparameter and the second hyperparameter are configured consistently to determine the overall loss function.

6. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: Based on the potential representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tuning the model with a small number of samples until the model converges is performed to determine the CT-VAE model with implicit labels, specifically including: Performing confidence sample discrimination on the potential representation data to determine a high-confidence event sample set; wherein the confidence sample discrimination includes: sample discrimination, confidence discrimination, and sample confidence threshold filtering; Based on the high-confidence event sample set, abnormal event data clusters are deduced in the latent space to obtain abnormal cluster clustering; According to the abnormal cluster clustering, the CT-VAE model is trained until the model converges, and the CT-VAE model of the implicit label is determined.

7. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: Perform event diagnosis reasoning on the data window to obtain event analysis results and event diagnosis reports, specifically including: Performing reconstruction score calculation on the data window to obtain a reconstruction error coefficient; Performing a potential deviation calculation on the data window to determine a deviation distance between a potential vector corresponding to the data window and a normal data distribution center; Calculating a causal violation score for the data window to obtain a score of a causal consistency loss item on the data window; Performing a weighted summation of the scores of the reconstruction error coefficient, the deviation distance, and the causal consistency loss term on the data window to determine a total anomaly score; Based on the total anomaly score, the event analysis result and the event diagnosis report are obtained through event detection and classification.

8. The industrial time series event analysis method based on causal regularization according to claim 1 is characterized in that: After inputting the data window into the implicitly labeled CT-VAE model and performing event diagnosis reasoning on the data window to obtain an event analysis result and an event diagnosis report, the method further includes: Identifying unknown events based on the event analysis results and marking the identified unknown events; Performing latent space trajectory visualization on the event diagnosis report to obtain a causal graph deviation analysis to determine a visualized event diagnosis report.

9. A method and device for analyzing industrial time series events based on causal regularization, characterized in that: The device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Acquiring multivariate time series data, and performing multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; Reconstructing a potential vector of a probability distribution of the time series feature data to determine a reconstruction loss and a distance metric loss; Constructing a priori causal graph, and based on the priori causal graph, obtaining a causal consistency loss function through acyclic constraint balance analysis; According to the reconstruction loss and the distance metric loss, performing hyperparameter consistency processing on the causal consistency loss function to determine an overall loss function, and performing unsupervised training on the overall loss function to minimize the overall loss; Obtaining latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tuning the model with a small number of samples until the model converges, thereby determining the CT-VAE model of the implicit label; The data window is input into the implicitly labeled CT-VAE model, and event diagnosis reasoning is performed on the data window to obtain event analysis results and event diagnosis reports.

10. A non-volatile computer storage medium storing computer executable instructions for an industrial time series event analysis method based on causal regularization, characterized in that: The computer executable instructions are configured to: Acquiring multivariate time series data, and performing multi-scale time series feature encoding on the multivariate time series data to obtain time series feature data; Reconstructing a potential vector of a probability distribution of the time series feature data to determine a reconstruction loss and a distance metric loss; Constructing a priori causal graph, and based on the priori causal graph, obtaining a causal consistency loss function through acyclic constraint balance analysis; According to the reconstruction loss and the distance metric loss, performing hyperparameter consistency processing on the causal consistency loss function to determine an overall loss function, and performing unsupervised training on the overall loss function to minimize the overall loss; Obtaining latent representation data, and based on the latent representation data and the CT-VAE model corresponding to the minimized overall loss, fine-tuning the model with a small number of samples until the model converges, thereby determining the CT-VAE model of the implicit label; The data window is input into the implicitly labeled CT-VAE model, and event diagnosis reasoning is performed on the data window to obtain event analysis results and event diagnosis reports.

Images