CN115659135A - Anomaly detection method for multi-source heterogeneous industrial sensor data - Google Patents
Anomaly detection method for multi-source heterogeneous industrial sensor data Download PDFInfo
- Publication number
- CN115659135A CN115659135A CN202211414174.3A CN202211414174A CN115659135A CN 115659135 A CN115659135 A CN 115659135A CN 202211414174 A CN202211414174 A CN 202211414174A CN 115659135 A CN115659135 A CN 115659135A
- Authority
- CN
- China
- Prior art keywords
- anomaly detection
- data
- samples
- encoder
- generator
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses an anomaly detection method for multisource heterogeneous industrial sensor data, which comprises the following steps: acquiring data, namely acquiring initial signal data from a sensor; the initial signal data respectively enters a stacked self-encoder model to be trained to obtain a first anomaly detection result, and a meta-learning method for unsupervised anomaly detection is adopted to carry out anomaly detection to obtain a second anomaly detection result; combining the two detection results to obtain a final abnormal result; and inputting the detection result output by the encoder model to a time Transformer model to predict the sensor abnormity. The invention can not only detect the abnormal of the sensing data, but also improve the detection accuracy by adopting various methods, can predict the sensor fault, and take certain measures to the prediction result, thereby greatly reducing the abnormal condition of the sensor data.
Description
Technical Field
The invention belongs to the technical field of industrial sensor detection, and particularly relates to an anomaly detection method for multi-source heterogeneous industrial sensor data.
Background
With the rapid development of information technology, sensor-based monitoring networks are widely used in various industrial monitoring and control.
At present, the anomaly detection methods of multi-source heterogeneous industrial sensor data are mainly divided into three types: the first method is an anomaly detection method based on a deep learning algorithm, and combines the self time sequence characteristics of an industrial sensor, wherein the most common method is a non-linear time sequence 'ARIMA + GARCH' mixed model of unsupervised learning; the second method is an anomaly detection method based on multilayer network flow, which is layered in combination with the industrial process of an industrial control system and is based on the analysis of effective load and the analysis of message header data; the third method is an anomaly detection method of multi-layer network traffic coupled with a deep learning algorithm.
However, these methods can only detect when the sensor is abnormal, but cannot predict the sensor failure, and cannot determine the detection accuracy well.
Disclosure of Invention
In order to solve the problems, the invention provides an anomaly detection method for multisource heterogeneous industrial sensor data, which can not only detect the anomaly of the sensor data, but also improve the detection accuracy by adopting various methods, can predict the sensor fault, and take certain measures for the prediction result, thereby greatly reducing the abnormal conditions of the sensor data.
In order to achieve the purpose, the invention adopts the technical scheme that: an anomaly detection method for multi-source heterogeneous industrial sensor data comprises the following steps:
s10, acquiring data, and acquiring initial signal data from a sensor;
s20, respectively entering initial signal data into a stacked self-encoder model for training to obtain a first anomaly detection result, and performing anomaly detection by adopting a meta-learning method for unsupervised anomaly detection to obtain a second anomaly detection result; combining the two detection results to obtain a final abnormal result;
and S30, inputting the detection result output by the stacked self-encoder model into a time Transformer model to predict the sensor abnormity.
Further, after initial signal data are preprocessed and extracted, the initial signal data are used for training a stacking automatic encoder;
the initial signal data is preprocessed by a data enhancement method based on a generation countermeasure network, based on countermeasure learning and cooperative learning, a generator generates abnormal state samples with two characteristics, namely real and distinguishable samples, the generated samples are added into actual samples and serve as balanced training data, data extraction is completed, and the window size is set.
Further, the generation countermeasure network includes a generator G that will generate a manufactured sample G (z) from the initial signal data and a discriminator D that will distinguish the source of the input sample from the actual data x r Or generating a manufactured sample G (z), and approximating the potential distribution P (X) of the actual sample during acquisition by learning antagonism to result in the distribution of newly generated samples r )；
A process state tag is attached to the input of the generator and discriminator which enables the user to determine the number of samples generated during an abnormal state of the collection process to generate a balanced sample.
Further, the goal in the discriminator is to maximize the generation of the adversarial network objective function through adversarial learning with the generator; regularizing, by the discriminator, a gradient of the discriminator by a gradient penalty; and provides the actual sample x to the discriminator r And the process state y of the error flag m Constituent additional inputs, discriminators learn to distinguish inputs (x) by minimization r ,y m ) Not actual but rather generated samples.
Further, the goal in the generator is to generate the competing network objective function by minimizing, by learning the actual sample distribution P (X) of the acquisition process r ) Generating a sample; the generator has an additional item in the objective function related to the classifier, and a cooperative relationship is established between the generator and the classifier, contrary to the confrontation relationship of the discriminator, so as to generate samples which are easy to distinguish between process states in the acquisition process, provide the classification loss of the labels of the generated samples to the objective function of the generator, and finally train the generator by minimizing the objective function thereof;
the target function of the classifier consists of real samples in the acquisition process and classification losses of generated samples, and the samples of the generator supplement the real samples in the acquisition process so as to generate balanced training data in the classifier; by minimizing the equation, the classifier is optimized to minimize the classification loss from the actual and generated samples.
Further, training is performed by stacking the auto-encoder models, including individually training the auto-encoder models, combining the output of each individually trained auto-encoder model using a stacked auto-encoder;
the middle layer of the auto-encoder is used to represent the feature vectors of the input of the rear model, and this process is repeated for each model until the last layer, which uses the features constructed by the previous layer as the input of the last layer to reproduce the input layer; the original input is recreated at the last level using the most refined feature form, and anomalies are identified by the feature vectors of the input and output of the final model, which are then subjected to error analysis.
Further, abnormality detection is performed by employing a meta-learning method for unsupervised abnormality detection, and a meta-learner for abnormality detection and a tool for collecting metadata for automatic supervised abnormality detection are cited for abnormality detection.
Further, the meta-learner for anomaly detection is optimized for automatic supervised learning that establishes anomaly detection using the set of meta-data based on the existing set of data provided as the set of meta-data, and a new set of data is detected by the meta-learner for anomaly detection.
Furthermore, the output of the stacked automatic encoder is used as the input of the time Transformer model, the QoS position encoding is used for describing a dimension vector generated by each time step of position information in an input sequence, then the dimension vector is transmitted to an encoder module of the time Transformer, the encoder module firstly obtains a QoS attention score and finally feeds the QoS attention score to a feed-forward network for predicting QoS, and the fault prediction of the sensor is completed so as to improve the accuracy of the abnormal detection.
The beneficial effects of the technical scheme are as follows:
the present invention introduces a machine learning method to model normal working operations and detect anomalies. The method extracts key features from known normal operation signals through a self-encoder to simulate machine behaviors and automatically identify abnormalities, and improves the accuracy of abnormality detection by combining a meta-learning method for unsupervised abnormality detection. And the output data of the self-encoder is used as input, and a time Transformer model is adopted to predict the sensor fault, so that the accuracy of the abnormal detection can be improved to a certain extent.
In order to better realize the abnormal detection of the data of the multi-source heterogeneous sensor, the invention adopts a method for enhancing data based on a generation countermeasure network to balance training data, and adopts two methods, namely a stacked self-encoder and a meta-learning method for unsupervised abnormal detection to carry out abnormal detection on the data of the multi-source heterogeneous sensor, and combines the detection results of the two methods to improve the accuracy of the abnormal detection. Meanwhile, a time Transformer model is adopted to predict QoS and also predict sensor faults, so that the accuracy and efficiency of abnormality detection are improved to a certain degree.
In order to highlight the occurrence of abnormal behaviors, the invention utilizes a stacked automatic encoder to extract features from the acquired original signals and uses the features to identify the abnormality. Stacked autoencoders allow more complex models to be created by stacking the autoencoders because the models force the data dimensions to be reduced, thereby determining more complex features in the data. Such a multi-level model may provide more accurate results by learning and predicting anomalous data by modeling only normal data. Meanwhile, a meta-learning method for unsupervised anomaly detection is used in combination with stacked automatic coding, so that the accuracy of anomaly detection is improved.
In order to better detect the abnormality of the sensor data, the data finally output by the stacked automatic encoder is sent to a time Transformer model as input data to predict the QoS of the sensor after the abnormality detection is finished, and the sensor fault is also predicted, so that the accuracy of the abnormality detection can be improved to a certain extent.
The invention rarely occurs in abnormal conditions because the number of sensor data samples collected from normal conditions far exceeds the number of data samples collected from abnormal conditions, while the sensor is normally in normal conditions. It results in data set imbalances between normal and abnormal process states, and an abnormal detection method trained with an unbalanced data set provides inaccurate prediction results. In order to solve the problem of data imbalance, a data enhancement method based on a generation countermeasure network is utilized, so that the effectiveness and the accuracy of anomaly detection are improved.
Drawings
FIG. 1 is a schematic diagram of the anomaly detection method for multi-source heterogeneous industrial sensor data according to the invention;
FIG. 2 is a schematic diagram of a stacked self-encoder in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram of a time Transformer model according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described with reference to the accompanying drawings.
In this embodiment, referring to fig. 1, the present invention provides an anomaly detection method for multi-source heterogeneous industrial sensor data, including the steps of:
s10, acquiring data, and acquiring initial signal data from a sensor;
s20, respectively entering initial signal data into a stacked self-encoder model for training to obtain a first anomaly detection result, and performing anomaly detection by adopting a meta-learning method for unsupervised anomaly detection to obtain a second anomaly detection result; combining the two detection results to obtain a final abnormal result;
and S30, inputting the detection result output by the encoder model to the time Transformer model to predict the sensor abnormity.
As an optimization scheme of the above embodiment, the initial signal data is preprocessed to extract data, and then used for training a stack automatic encoder;
the initial signal data is preprocessed by a data enhancement method based on a generation countermeasure network, based on countermeasure learning and cooperative learning, a generator generates abnormal state samples with two characteristics, namely real and distinguishable samples, the generated samples are added into actual samples and serve as balanced training data, data extraction is completed, and the window size is set. The problem of data unbalance is solved.
The signal is divided into different window sizes and used to simulate normal behavior and detect anomalies. The number of hidden node parameters grows exponentially according to the initial window size. A larger window will allow more data to be processed, but scalability will be affected accordingly. Thus, anomaly detection is performed by applying multiple windows of different sizes to the data.
By inputting each window through a machine learning model and automatically extracting features, abnormal behavior can be effectively detected by using a stacked auto-encoder applied by the window model.
The generation countermeasure network comprises a generator G and a discriminator D, the generator generates a manufactured sample G (z) from initial signal data, the discriminator distinguishes whether the source of the input sample is from actual data xr or the generated manufactured sample G (z), the distribution of the newly generated sample is caused by the countermeasure learning, and the potential distribution P (X) of the actual sample in the acquisition process is close to the potential distribution P (X) of the actual sample r )；
A process state label is attached to the generator and discriminator input that enables the user to determine the number of samples generated during an abnormal state of the collection process to generate a balanced sample.
The goal in the discriminator is to maximize the generation of the competing network objective function through competing learning with the generator; regularizing, by the discriminator, a gradient of the discriminator by a gradient penalty; and provides the actual sample x to the discriminator r And the process state y of the error flag m Constituent additional inputs, discriminators learn to distinguish inputs (x) by minimization r ,y m ) Not actual but rather generated samples. Since the input provides an additional learning task for the discriminator, it prevents the discriminator from well distinguishing between actual samples and generated samples before the generator approaches the actual sample distribution of the manufacturing process; otherwise, the discriminator does not provide information gradients for generator learning. Two additional conditions are provided for the stable learning process to overcome the instability problem of training GAN and the difficulty in optimizing antagonistic learning and convergence.
The goal in the generator is to generate the competing network objective function by minimization, by learning the actual sample distribution P (X) of the acquisition process r ) Generating a sample; the generator has an additional item in the objective function related to the classifier, and a cooperative relationship is established between the generator and the classifier, contrary to the confrontation relationship of the discriminator, so as to generate samples which are easy to distinguish between process states in the acquisition process, provide the classification loss of the labels of the generated samples to the objective function of the generator, and finally train the generator by minimizing the objective function thereof;
the target function of the classifier consists of real samples in the acquisition process and classification losses of generated samples, and the samples of the generator supplement the real samples in the acquisition process so as to generate balanced training data in the classifier; by minimizing the equation, the classifier is optimized to minimize the classification loss from the actual and generated samples.
As an optimization of the above embodiments, training is performed by stacked self-encoder models, including individually training the auto-encoder models, combining the output of each individually trained auto-encoder model using a stacked auto-encoder;
as shown in fig. 2. The middle layer of the auto-encoder is used to represent the feature vectors of the input of the rear model, and this process is repeated for each model until the last layer, which uses the features constructed by the previous layer as the input of the last layer to reproduce the input layer; the original input is recreated at the last level using the most refined feature form, and anomalies are identified by the feature vectors of the input and output of the final model, which are then subjected to error analysis.
Self-encoder models are trained using windows of different sizes. The first model uses a 500 size viewing window. In this model, the hidden layer contains 300 neurons, which ensures that enough structure remains from the original data and that enough features can be extracted. The output of the hidden layer is connected to an output layer containing 500 neurons. The model is trained to approximate the input on the output layer. Training is ended until the optimal reconstruction error is reached. When the training in the model is complete, the hidden layer (feature) is used as input to the second self-encoder, which is trained again to approximate the input on the output layer. In the second model, the hidden layer is reduced to 200 neurons. This configuration continues until the final model contains 70 hidden neurons that, after training, constitute our final feature set (70 features).
As an optimization scheme of the above embodiment, the abnormality detection is performed by adopting a meta-learning method for unsupervised abnormality detection, and a meta-learner for abnormality detection and a tool for collecting metadata for automatically supervised abnormality detection are cited for abnormality detection.
A meta-learner for anomaly detection is optimized from an existing data set provided as a meta-data set using automatic supervised learning to establish anomaly detection with the meta-data set, and a new data set is detected by the meta-learner for anomaly detection.
The meta learning method has two parts: first, there is a transformation function applied to the data set and this function is called, which applies to the given data set; in a second stage, dataset similarity is calculated based on the distance metric ψ. The anomaly detection task is performed by capturing the similarity between data sets.
As an optimization scheme of the above embodiment, as shown in fig. 3, the output of the stacked automatic encoder is used as an input of a time Transformer model, a dimension vector generated by each time step of describing position information in an input sequence through QoS position encoding is then passed to an encoder module of a time Transformer, the encoder module first obtains a QoS attention score, and finally feeds the QoS attention score into a feed-forward network to predict QoS, so as to complete fault prediction of a sensor to improve accuracy of anomaly detection.
Position coding is a dimensional vector generated for each time step describing position information in an input sequence. In this step, sinusoidal position coding is employed, since the position coding provided by the method is fixed for each time step and no additional weights need to be trained. The position code of each input sequence is added to the output of the input layer by position and then passed to the encoder block of the time transformer.
The encoder module consists of a stack of encoders, all of which are structurally identical. The input to the encoder is first passed to a multi-headed attention module that looks at the QoS values in the input sequence. It then provides an attention score between these two QoS values and proceeds to process other QoS in all other input sequences in the same manner. These attention scores are forwarded to a normalization layer. These layers are used to stabilize the hidden state of the network and reduce training time. Finally, the output of the normalization layer is fed into a feed-forward network.
The invention can not only detect the abnormality of the sensing data, but also detect the abnormality of the multi-source heterogeneous sensor data by adopting two methods, and combines the detection results of the two methods, thereby improving the accuracy of the abnormality detection. Meanwhile, the sensor fault is predicted, and the accuracy and efficiency of the abnormal detection are improved to a certain degree. The present invention utilizes a stacked self-encoder approach for anomaly detection. The method comprises the steps of extracting key features through a self-encoder in training data generated by a data enhancement method based on a generation countermeasure network to simulate machine behaviors and automatically identify abnormalities, and improving the accuracy of abnormality detection by combining a meta-learning method for unsupervised abnormality detection. And the output data of the stacked self-encoder is used as input, and a time Transformer model is adopted to predict the sensor fault, so that the accuracy and efficiency of the abnormal detection can be improved to a certain extent.
The foregoing shows and describes the general principles and features of the present invention, together with the advantages thereof. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are given by way of illustration of the principles of the present invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, and such changes and modifications are within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (9)
1. An anomaly detection method for multi-source heterogeneous industrial sensor data is characterized by comprising the following steps:
s10, acquiring data, and acquiring initial signal data from a sensor;
s20, respectively entering initial signal data into a stacked self-encoder model for training to obtain a first anomaly detection result, and performing anomaly detection by adopting a meta-learning method for unsupervised anomaly detection to obtain a second anomaly detection result; combining the two detection results to obtain a final abnormal result;
and S30, inputting the detection result output by the stacked self-encoder model into a time Transformer model to predict the sensor abnormity.
2. The anomaly detection method for the multisource heterogeneous industrial sensor data according to claim 1, characterized in that initial signal data is preprocessed to extract data and then used for stacking automatic encoder training;
the initial signal data is preprocessed by a method of data enhancement based on a generation confrontation network, based on confrontation learning and cooperative learning, a generator generates abnormal state samples with two characteristics, namely real and distinguishable samples, the generated samples are added into actual samples and used as balanced training data to finish data extraction and set window size.
3. The anomaly detection method oriented to multi-source heterogeneous industrial sensor data according to claim 2, characterized in that the generation countermeasure network comprises a generator G and a discriminator D, the generator is used for generating a manufactured sample G (z) from initial signal data, and the discriminator is used for distinguishing the source of an input sample from actual data x r Or generating a manufactured sample G (z), and approximating the potential distribution P (X) of the actual sample during acquisition by learning antagonism to result in the distribution of newly generated samples r )；
A process state label is attached to the generator and discriminator input that enables the user to determine the number of samples generated during an abnormal state of the collection process to generate a balanced sample.
4. The anomaly detection method oriented to the multi-source heterogeneous industrial sensor data is characterized in that the goal in the discriminator is to maximize the generation of a confrontation network objective function through the confrontation learning with a generator; regularizing, by the discriminator, a gradient of the discriminator by a gradient penalty; and provides the actual sample x to the discriminator r And the process state y of the error flag m Composed additional inputs, discriminators learn to distinguish inputs (x) by minimization r ,y m ) Not actual but rather generated samples.
5. The anomaly detection method oriented to multi-source heterogeneous industrial sensor data according to claim 3, characterized in that the goal in the generator is to generate a countermeasure network objective function through minimization and learn an actual sample distribution P (X) of an acquisition process r ) Generating a sample; the generator has an additional item in an objective function related to the classifier, and a cooperative relationship is established between the generator and the classifier, contrary to the confrontation relationship of the discriminator, so as to generate samples which are easy to distinguish between process states in the acquisition process, provide the classification loss of labels of the generated samples to the objective function of the generator, and finally train the generator by minimizing the objective function thereof;
the target function of the classifier consists of real samples in the acquisition process and classification losses of generated samples, and the samples of the generator supplement the real samples in the acquisition process so as to generate balanced training data in the classifier; by minimizing the equation, the classifier is optimized to minimize the classification loss from the actual and generated samples.
6. The anomaly detection method for the multi-source heterogeneous industrial sensor data according to claim 1, characterized in that training is performed by stacking self-encoder models, including individually training the auto-encoder models, combining the output of each individually trained auto-encoder model using a stacked auto-encoder;
the middle layer of the auto-encoder is used to represent the feature vectors of the input of the rear model, and this process is repeated for each model until the last layer, which uses the features constructed by the previous layer as the input of the last layer to reproduce the input layer; the original input is recreated at the last level using the most refined feature form, and anomalies are identified by the feature vectors of the input and output of the final model, which are then subjected to error analysis.
7. The anomaly detection method oriented to the multi-source heterogeneous industrial sensor data according to claim 1, characterized in that the anomaly detection is carried out by adopting a meta-learning method for unsupervised anomaly detection, and a meta-learner for anomaly detection and a tool for collecting metadata for automatically supervised anomaly detection are introduced for anomaly detection.
8. The anomaly detection method oriented to the multi-source heterogeneous industrial sensor data according to claim 7, wherein an automatic supervised learning for establishing anomaly detection by using a metadata set is used for optimizing a meta-learner for anomaly detection according to an existing data set provided as the metadata set, and a new data set is detected by the meta-learner for anomaly detection.
9. The anomaly detection method oriented to the multisource heterogeneous industrial sensor data is characterized in that the output of a stacked automatic encoder is used as the input of a time Transformer model, a QoS position code is used for describing a dimension vector generated by each time step of position information in an input sequence, then the dimension vector is transmitted to an encoder module of the time Transformer, the encoder module firstly obtains a QoS attention score and finally feeds the QoS attention score into a feed-forward network to predict QoS, and the fault prediction of a sensor is completed to improve the accuracy of anomaly detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414174.3A CN115659135A (en) | 2022-11-11 | 2022-11-11 | Anomaly detection method for multi-source heterogeneous industrial sensor data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414174.3A CN115659135A (en) | 2022-11-11 | 2022-11-11 | Anomaly detection method for multi-source heterogeneous industrial sensor data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115659135A true CN115659135A (en) | 2023-01-31 |
Family
ID=85020779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211414174.3A Pending CN115659135A (en) | 2022-11-11 | 2022-11-11 | Anomaly detection method for multi-source heterogeneous industrial sensor data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115659135A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117421386A (en) * | 2023-12-19 | 2024-01-19 | 成都市灵奇空间软件有限公司 | GIS-based spatial data processing method and system |
-
2022
- 2022-11-11 CN CN202211414174.3A patent/CN115659135A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117421386A (en) * | 2023-12-19 | 2024-01-19 | 成都市灵奇空间软件有限公司 | GIS-based spatial data processing method and system |
| CN117421386B (en) * | 2023-12-19 | 2024-04-16 | 成都市灵奇空间软件有限公司 | GIS-based spatial data processing method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914873A (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN109583092B (en) | Intelligent mechanical system fault diagnosis method based on multi-level and multi-mode feature extraction | |
| CN112491796B (en) | Intrusion detection and semantic decision tree quantitative interpretation method based on convolutional neural network | |
| CN111914883B (en) | Spindle bearing state evaluation method and device based on deep fusion network | |
| CN111598179B (en) | Power monitoring system user abnormal behavior analysis method, storage medium and equipment | |
| CN105893208A (en) | Cloud computing platform system fault prediction method based on hidden semi-Markov models | |
| CN112087442B (en) | Time sequence related network intrusion detection method based on attention mechanism | |
| KR102281819B1 (en) | Auto Encoder Ensemble Based Anomaly Detection Method and System | |
| CN112738014A (en) | Industrial control flow abnormity detection method and system based on convolution time sequence network | |
| CN114615093A (en) | Anonymous network traffic identification method and device based on traffic reconstruction and inheritance learning | |
| CN117113262B (en) | Network traffic identification method and system | |
| CN113687972B (en) | Processing method, device, equipment and storage medium for abnormal data of business system | |
| CN112906764B (en) | Communication safety equipment intelligent diagnosis method and system based on improved BP neural network | |
| CN113485863B (en) | Method for generating heterogeneous imbalance fault samples based on improved generation of countermeasure network | |
| CN114528755A (en) | Power equipment fault detection model based on attention mechanism combined with GRU | |
| CN115811440B (en) | Real-time flow detection method based on network situation awareness | |
| CN115643115B (en) | Industrial control network security situation prediction method and system based on big data | |
| CN117220920A (en) | Firewall policy management method based on artificial intelligence | |
| CN111949480A (en) | Log anomaly detection method based on component perception | |
| CN115659135A (en) | Anomaly detection method for multi-source heterogeneous industrial sensor data | |
| CN115310837A (en) | Complex electromechanical system fault detection method based on causal graph attention neural network | |
| CN116633601A (en) | Detection method based on network traffic situation awareness | |
| CN114785548A (en) | Virtual flow anomaly detection method and system based on weighted adaptive ensemble learning and intelligent flow monitoring platform | |
| Jiang et al. | Recognizing control chart patterns with neural network and numerical fitting | |
| CN117349595A (en) | Industrial scene-oriented extremely few sample fault diagnosis method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |