System status detection method, system, equipment and medium of industrial Internet system

Abstract

The present application provides a method, system, device and medium for detecting the system status of an industrial Internet system. It relates to the field of industrial Internet technology. The method comprises: obtaining system data of an industrial Internet system; detecting system data using a system status detection model to obtain the system status of the industrial Internet system; the system status detection model comprises a shallow feature extraction module and a plurality of deep feature detection modules connected in sequence, and the system status detection model is obtained by training a dynamic neural network model using a training data set, and the training data set comprises historical system data of the industrial Internet system; the system status comprises a normal system and an abnormal system; when the system status is an abnormal system, the system status of the industrial Internet system is output. The method of the present application reduces the amount of data calculation of the model and improves the detection efficiency while ensuring the accuracy of the detection results.

Description

Technical Field

[0001] The present application relates to the field of industrial Internet technology, and in particular to a system status detection method, system, equipment and medium for an industrial Internet system. Background Art

[0002] With the rapid development of the manufacturing industry and the diversification of market demand, the Industrial Internet System has been widely used in modern industrial production. The Industrial Internet System is a system that combines traditional industrial equipment with Internet technology, and realizes the intelligence, interconnection and automation of industrial equipment through sensors, communication technology, big data analysis, cloud computing and other means.

[0003] However, the industrial Internet system often encounters various abnormal situations during the production process, such as equipment failure, process deviation, material shortage, etc. If these abnormal situations cannot be discovered and handled in time, they will not only affect production efficiency, but may also lead to product quality problems and even safety accidents. Therefore, it is of great significance to effectively monitor the system status of the industrial Internet system.

[0004] In recent years, with the widespread application of deep learning, system status detection of industrial Internet systems based on deep learning has attracted more and more attention. However, due to the increasing depth of neural network models, the status detection efficiency of existing intelligent industrial Internet systems is low. Summary of the invention

[0005] The present application provides a system status detection method, system, device and medium for an industrial Internet system, so as to solve the problem of low status detection efficiency of existing industrial Internet systems and improve the detection efficiency of industrial Internet systems.

[0006] In a first aspect, the present application provides a system status detection method for an industrial Internet system, comprising:

[0007] Obtain system data of industrial Internet systems;

[0008] The system data is detected by using a system status detection model to obtain the system status of the industrial Internet system; wherein the system status detection model includes a shallow feature extraction module and a plurality of deep feature detection modules connected in sequence, and the system status detection model is obtained by training a dynamic neural network model using a training data set, and the training data set includes historical system data of the industrial Internet system; the system status includes a normal system and an abnormal system;

[0009] When the system status is system abnormality, the system status of the industrial Internet system is output.

[0010] In an optional implementation, using a system status detection model to detect the system data to obtain the system status of the industrial Internet system includes:

[0011] Inputting the system data into the shallow feature extraction module to obtain corresponding shallow feature extraction data;

[0012] For any level of deep feature detection module except the last level of deep feature detection module, the shallow feature data output by the shallow feature extraction module, or the deep feature detection module of the previous level, is input into the deep feature detection module of the current level to perform deep feature extraction, and the extracted deep feature data is subjected to state detection to obtain the corresponding system state, and the confidence corresponding to the system state is obtained;

[0013] If the confidence does not meet the preset confidence threshold condition, outputting the deep feature data of the current level as shallow feature data of the deep feature detection module of the next level;

[0014] If the confidence level satisfies the confidence level threshold condition, the detection status of the current level is output as the system status corresponding to the industrial Internet system.

[0015] In an optional implementation, any deep feature detection module at any level except the last level includes a deep feature extraction and state detection submodule, and a detection result review submodule; wherein,

[0016] The deep feature extraction and state detection submodule is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous level, and perform state detection on the extracted deep feature data to obtain the corresponding system state;

[0017] The detection result review submodule obtains the confidence level corresponding to the system status; if the confidence level does not meet the preset confidence threshold condition, the deep feature data of the current layer is used as the shallow feature data of the next layer of deep feature detection module; if the confidence level meets the confidence threshold condition, the system status corresponding to the industrial Internet system is obtained.

[0018] In an optional implementation, the deep feature extraction and state detection submodule includes a deep feature extraction layer and a state detection layer;

[0019] The deep feature extraction layer is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous layer to obtain deep feature data;

[0020] The state detection layer is used to perform state detection on the extracted deep feature data to obtain the corresponding system state.

[0021] In an optional implementation, the confidence threshold condition includes a first threshold and a second threshold; the first threshold is less than the second threshold;

[0022] The confidence level not meeting the confidence level threshold condition includes: the confidence level is less than the first threshold level, or the confidence level is greater than the second threshold level;

[0023] The confidence level satisfies the confidence level threshold condition, including: the confidence level is greater than the first threshold level, and the confidence level is less than the second threshold level.

[0024] In an optional implementation, the method further includes:

[0025] The constructed dynamic neural network model is trained using a training data set to obtain the system state detection model; the loss functions used in the training include: regression loss, association difference loss and feature disentanglement loss; the association difference loss includes prior association loss and sequence association loss; wherein,

[0026] The regression loss represents the difference between the reconstruction result of the historical system data and the historical system data output by the dynamic neural network model during the training process;

[0027] The prior association loss represents the attention weight distribution of the corresponding row of the historical system data in the attention matrix;

[0028] The sequence association loss represents the association between the deep feature data corresponding to the current data point and the domain data point in the historical system data;

[0029] The feature disentanglement loss represents the degree of correlation between the deep feature data corresponding to different data points in the historical system data.

[0030] In a second aspect, the present application provides a system status detection system for an industrial Internet system, including: a system data acquisition subsystem, a system status detection subsystem, and a system status alarm subsystem; wherein,

[0031] The system data acquisition subsystem is used to collect system data of the industrial Internet system and transmit the system data to the system status detection subsystem;

[0032] The system status detection subsystem is used to execute the system status detection method of the industrial Internet system described in any one of the first aspects;

[0033] The system status alarm subsystem is used to alarm the system status output by the system status detection subsystem.

[0034] In a third aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0035] The memory stores computer-executable instructions;

[0036] The processor executes the computer-executable instructions stored in the memory to implement the method according to the first aspect.

[0037] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect.

[0038] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0039] The system status detection method of the industrial Internet system provided in the present application obtains the system data of the industrial Internet system; uses the system status detection model to detect the system data to obtain the system status of the industrial Internet system; wherein the system status detection model includes a shallow feature extraction module and a plurality of deep feature detection modules connected in sequence, and the system status detection model is obtained by training a dynamic neural network model with a training data set, and the training data set includes historical system data of the industrial Internet system; the system status includes a normal system and an abnormal system; when the system status is an abnormal system, the system status of the industrial Internet system is output; in the above scheme, shallow feature extraction is performed on the collected system data, and in the subsequent When multiple deep feature detection modules connected in sequence are used for state detection, a shallower deep feature detection module is first used to perform deep feature extraction and state detection, and the detection is terminated directly when the detection result is satisfactory; secondly, a deeper detection module is used to continue the detection when the detection result is unsatisfactory; in this way, the detection is terminated when a satisfactory detection result can be obtained by using a shallower deep feature detection module, so as to reduce the use of a full-depth detection module for state detection, reduce the high computational complexity of the model, and realize the deployment of the model for state detection under the condition of insufficient computing power on the edge side, and realize the state detection model for state detection of the industrial Internet system, thereby realizing efficient detection of the network state of the industrial Internet. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0041] Figure 1 A flowchart of a system status detection method for an industrial Internet system provided in an embodiment of the present application;

[0042] Figure 2 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 1 ;

[0043] Figure 3 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 2 ;

[0044] Figure 4 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 3 ;

[0045] Figure 5 A schematic diagram of the structure of a system status detection system for an industrial Internet system provided in an embodiment of the present application;

[0046] Figure 6 It is a block diagram of an electronic device provided in an embodiment of the present application.

[0047] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0048] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0049] In modern industrial production, various abnormal situations are often encountered, such as equipment failure, process deviation, material shortage, etc. If these abnormal situations cannot be discovered and handled in time, they will not only affect production efficiency, but may also lead to product quality problems and even safety accidents. Therefore, it is of great significance to effectively monitor the system status of the Industrial Internet system.

[0050] Among the system status detection methods corresponding to the industrial Internet system, a large number of traditional methods mainly include clustering-based, density-based, isolation-based, and distance-based methods. In recent years, with the widespread application of deep learning, supervised, semi-supervised, and unsupervised detection methods based on deep learning have received more and more attention.

[0051] In some optional implementations, for supervised deep learning detection methods, Potluri et al. proposed a detection method based on convolutional neural network (CNN), first converting the feature vector into an image of 8*8 pixels, and then constructing a three-layer CNN to classify the attack. Yang et al. proposed an iFinger fingerprint method, using CNN to predict network traffic, and judging whether the industrial Internet is attacked based on the difference between the predicted traffic and the actual traffic. Li et al. proposed a deep learning anomaly detection method based on multi-convolutional neural network (multi-CNN) fusion to fully utilize weakly correlated feature data samples to establish a classification model, thereby identifying abnormal data of the industrial Internet. Zhou et al. proposed a variational long short-term memory network (LSTM) learning model, in which a variational reparameterized compression (encoding and decoding) network and an estimation network for low-dimensional feature representation are constructed to solve the data distribution imbalance problem in high-dimensional feature anomaly detection for intelligent industrial applications of the industrial Internet. Chen et al. used the CNN-BiLSTM network structure to extract cross-domain sample features to further improve the feature expression capability.

[0052] In other optional implementations, for unsupervised deep learning detection methods, Zhang et al. extracted features based on sparse autoencoders, sampled unbalanced data sets using synthetic minority oversampling techniques (SMOTE), and used XGBoost models to detect attacks. Sparse autoencoders introduce sparse constraints in the original autoencoders to enhance the detection capability of unknown samples. The SMOTE algorithm oversamples the minority class and divides the majority class into many subclasses so that each class is balanced. Kravchik et al. use an improved undercomplete autoencoder (UAE) to detect attacks based on both time domain and frequency domain features. Liu et al. introduce a regularized sparse deep belief network (RSDBN) model to model abnormal behavior. RSDBN uses an unsupervised restricted Boltzmann machine to train the model layer by layer for potential attack detection of information systems, effectively achieving high detection accuracy and low false alarm rate. Adversarial learning methods can improve the detection accuracy of small data sets. Zhang et al. used GAN for data enhancement and added the generated data to the training set to detect variants of the attack. Chen et al. proposed a GAN-based adversarial attack defense method for covert network attacks such as injection attacks, function code attacks, and reconnaissance attacks. Deep reinforcement learning can combine the perception ability of deep learning with the decision-making ability of reinforcement learning. Qiu et al. developed a dueling deep Q-learning method to deal with the network threats faced by secure communications in the industrial Internet.

[0053] In some other optional implementations, for the detection method of semi-supervised deep learning, Aouedi et al. first train an autoencoder AE on each device using unlabeled local private data to learn low-dimensional features, then the cloud server aggregates these models into the global AE using federated learning, and finally the cloud server forms a supervised neural network by adding a fully connected layer to the global AE encoder, and trains the final model using publicly available labeled data. Hassan et al. proposed a semi-supervised industrial Internet network attack detection model based on deep learning. The model is divided into two layers. First, the unsupervised layer uses a stacked restricted Boltzmann network to extract features of labeled and unlabeled data, and then the supervised layer uses support vector machines and random forest methods to train elastic classifier models to detect unlabeled attacks. This method uses unsupervised learning to use unlabeled data to learn the rapid changes of unknown attack models, so it has strong adaptability to emerging network attacks and their dynamic characteristics. Xu et al. found that the self-attention weight distribution of each time point can reflect the association with the entire sequence. Due to the scarcity of outliers, the correlation between the outlier points and the entire sequence is low, which reflects the inherent difference between normal points and outliers. They then proposed an Anomaly-Transformer structure with an Anomaly-Attention mechanism to calculate the association difference. Qiu et al. proposed a fast and stable multivariate time series unsupervised anomaly detection (USAD) method based on reverse training autoencoders. Its autoencoder structure enables it to learn in an unsupervised manner, and the use of adversarial training and its architecture allows it to detect attacks while providing fast training.

[0054] However, current deep learning-based detection methods basically have the problem of high model complexity, which greatly increases the security overhead of the system's equipment when the Industrial Internet system performs system status detection. This also puts forward certain requirements on the computing power of the system's equipment, making it difficult for the Industrial Internet system to be deployed under conditions of insufficient computing power on the edge side. That is, there are problems such as low efficiency of Industrial Internet system status detection and impact on industrial production efficiency.

[0055] The system status detection method of the industrial Internet system provided in this application is intended to solve the above technical problems of the prior art. Specifically, a shallower detection module among multiple sequentially connected deep feature detection modules is first used to perform detection, and the detection is terminated when the detection result is satisfactory. When the detection result is unsatisfactory, a deeper detection module is used to continue the detection, so as to reduce the number of detections using the full-depth detection module, thereby reducing the amount of calculation of the model to a certain extent and improving the detection efficiency. In other words, if the detection results of the shallower deep feature detection module hardly contain system anomalies or are almost completely occupied by system anomalies, there is no need to use a deeper deep feature detection module to further extract features and perform detection; when the output results of the deep detection module include both system anomalies and system normality, a deeper detection module can be used to extract deeper features and perform detection to obtain more accurate detection results, thereby minimizing the amount of data calculation while ensuring prediction accuracy, thereby solving the problem of low efficiency of state detection of existing Industrial Internet systems, and realizing the state detection model of the Industrial Internet system by deploying a model for state detection under the condition of insufficient computing power on the edge side, thereby achieving efficient detection of the network status of the Industrial Internet.

[0056] The industrial Internet system is a system that combines traditional industrial equipment with Internet technology. Its application range is very wide, including manufacturing, energy, transportation, medical and other fields. Its main advantages are to improve production efficiency, reduce operating costs, improve product quality and realize intelligent management. The application scenario of the technical solution of this application can be applied to the scenario of status detection of industrial Internet systems deployed in any field to ensure smooth production. For any scenario, this solution includes data acquisition equipment, data processing equipment and status alarm equipment.

[0057] During the production process, first, the Industrial Internet system converts the received orders into task orders and generates a production schedule corresponding to the task orders. On this basis, the Industrial Internet system controls the stacker to accurately grab the pallets containing raw materials according to the production schedule, and puts the pallets into the outfeed conveyor line, and then controls the material conveying equipment to pick up the materials, and controls the Automated Guided Vehicle (AGV) to automatically deliver the sorted materials. Specifically, the AGV automatically transports the raw material pallets to the processing machine for material assembly. The type of AGV is not limited in this application, and an example may be a belt line composite AGV.

[0058] Secondly, the Industrial Internet system also controls the collaborative robotic arms to flexibly and automatically assemble materials delivered to the processing machines; optionally, the Industrial Internet system can control the robotic arms to install fixtures corresponding to different materials to achieve the clamping state of different materials, ultimately realizing the intelligent assembly of products, and controlling the automatic transportation of the assembled finished products to the intelligent packaging equipment. The three-axis module in the intelligent packaging equipment first clamps the ground cover and places it on the packaging station, then clamps the product and places it in the ground cover, ensuring that the top cover is clamped and accurately snapped in after being placed smoothly, and the labeling machine affixes the QR code label printed with product information to the packaging box.

[0059] Finally, the industrial Internet system controls the product packaging equipment to place the packaged finished products into the finished product warehouse, thereby completing the delivery of the received orders.

[0060] In the above-mentioned production process, the industrial Internet system includes multiple production equipment and control systems corresponding to each production equipment; in the real-time production operation process of each production equipment, operation data corresponding to each production equipment will be generated; the data acquisition equipment includes multiple sensing devices, and each sensing device is used to sense the operation data of each production equipment in the industrial Internet system during the production operation process, that is, to obtain the system data of the industrial Internet system in the production process, and transmit the above system data to the data processing equipment.

[0061] Optionally, data can be transmitted through a variety of communication networks in the Industrial Internet system; exemplary ones include Ethernet, wireless networks (such as Wi-Fi, 5G), industrial buses (such as Modbus, Profibus), etc.

[0062] When receiving system data, the data processing device first performs shallow feature extraction on it to facilitate subsequent deep feature extraction and status detection; then, a shallower detection module among multiple deep detection modules connected in sequence is used to perform detection. When the detection result is satisfactory, the detection is terminated. When the detection result is unsatisfactory, a deeper detection module is used to continue the detection, so as to reduce the use of full-depth detection modules for detection, reduce the high computational complexity of the model, and realize the deployment of models for status detection under the condition of insufficient computing power on the edge side. The status detection model for performing status detection on the industrial Internet system can realize efficient detection of the network status of the industrial Internet.

[0063] When the data processing equipment determines that the status detection result of the industrial Internet system is a system abnormality, it outputs the alarm signal corresponding to the abnormal status to the alarm device, and uses the alarm device to indicate and remind the staff that the industrial Internet system has an abnormality, so that production personnel can quickly respond to the system abnormality alarm and maintain the current Industrial Internet system.

[0064] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0065] Figure 1 A flowchart of a method for detecting the system status of an industrial Internet system provided in an embodiment of the present application. The method can be executed by a system status detection device of an industrial Internet system, which can be deployed on a server or an electronic device. The following description will be made using the deployment on an electronic device as an example. The method in this embodiment can be implemented by software, hardware, or a combination of software and hardware, such as Figure 1 As shown, the method comprises the following steps:

[0066] S110. Obtain system data of the industrial Internet system.

[0067] In this application, the Industrial Internet system can be interpreted as a system that combines traditional industrial equipment with Internet technology. Depending on the application field of the Industrial Internet system, different devices and corresponding control systems can be configured in the Industrial Internet system to control product production through the Industrial Internet system.

[0068] Next, this application takes the industrial Internet system in any field as an example to exemplify this solution. Specifically, in the current industrial Internet system, multiple sensing devices can be pre-installed according to each device in the industrial Internet system, and each sensing device is connected to each device configured in the system to collect the device data generated by each device during the production process; each device data includes but is not limited to the device type data corresponding to each device and the operation status data generated by each device during operation.

[0069] During the production process, each device can be monitored and analyzed by a visual device to obtain device type data corresponding to the device. Optionally, the visual device can include an image acquisition device and a detection device, and the detection device detects the image acquired by the image acquisition device to obtain the component type.

[0070] Optionally, some operating status data can also be obtained directly by reading digital signals in the control systems corresponding to each production equipment in the Industrial Internet system, that is, by collecting data in a Supervisory Control and Data Acquisition (SCADA) system.

[0071] In an exemplary embodiment of the present application, the various devices of the industrial Internet system may include a storage device for storing materials, an assembly device for distributing materials, a robotic arm for assembling materials, and the like. For a storage device, the device data includes the operating status and the type of discharging corresponding to the discharging conveyor line in the storage device; exemplarily, the operating status may be represented by a numerical value, where 0 represents stationary and 1 represents in transit. For a distribution device, the device data may be the coordinate data and the type of distribution corresponding to the distribution trajectory. For a robotic arm, the device data includes the type of clamp corresponding to when the robotic arm clamps the material and the spatial coordinate data of the robotic arm relative to the base when clamping the material. In some other embodiments, the industrial Internet system may further include a packaging device, and the device data corresponding to the device may include, but is not limited to, the type of packaging material clamped by a three-axis module, and the like.

[0072] In this way, the system data of the industrial Internet system can be obtained according to the device data generated by each device during the production process. In other words, the system data in this application includes the device data corresponding to each device in the industrial Internet system.

[0073] The collected system data can be used for system status detection on the one hand, and can also be stored in the server to form big data on the other hand, which is convenient for later industrial data mining, industrial data analysis, industrial data statistics and other big data mining and analysis.

[0074] S120. Use the system status detection model to detect system data and obtain the system status of the industrial Internet system.

[0075] Among them, the system state detection model includes a shallow feature extraction module and multiple deep feature detection modules connected in sequence.

[0076] In this application, the process of system status detection can be carried out in the Industrial Internet system, or in other terminal devices or servers that communicate with the Industrial Internet system; it should be noted that whether the detection is carried out in the Industrial Internet system or in other terminal devices or servers, the system status detection model can be trained in advance to improve the detection efficiency during the detection process. Optionally, if it is carried out in other terminal devices or servers, the data needs to be transmitted to the computer or server for detection when the system data is obtained.

[0077] During the detection process, the pre-trained system status detection model is called up, and the system data is input into the system status detection model for anomaly detection to obtain the system status corresponding to the industrial Internet system.

[0078] The system state detection model includes a shallow feature extraction module and a deep feature detection module; the shallow feature extraction module is used to extract shallow features of the system data input into the model, which is convenient for subsequent deep feature extraction and state detection processing. In this way, compared with the situation where deep features are directly extracted through one feature extraction module, two feature extractions are performed through two feature extraction modules, which can speed up the feature extraction speed and ensure the improvement of detection efficiency.

[0079] In the present application, the deep feature detection modules included in the model are multiple deep feature detection modules connected in sequence, and each deep feature detection module contains the same network structure; in some detection implementation methods, a deeper network structure is used to extract deeper features, and the information contained in the extracted features is also richer, which is convenient for improving the accuracy of subsequent detection results; however, before performing state detection on each set of data, all deep feature detection modules are used to extract the deepest features, which undoubtedly increases the security overhead of the system where the model is located. In order to reduce unnecessary security overhead as much as possible while ensuring prediction accuracy, in the system state detection model provided in the present application, if the shallower deep feature detection If the detection results of the detection module contain almost no system anomalies, or almost all of them are system anomalies, there is no need to use a deeper deep feature detection module to further extract features and detect them; when the output results of the shallower deep feature detection module contain both system anomalies and system normality, a deeper deep feature detection module is used to extract deeper features and detect them. This processing method can use fewer deep feature detection modules in the model for processing, thereby reducing the calculation amount of the model, and realizing the deployment of the model for status detection under the condition of insufficient computing power on the edge side. The status detection model for status detection of the industrial Internet system can achieve efficient detection of the network status of the industrial Internet.

[0080] S130. When the system status is system abnormality, output the system status of the industrial Internet system.

[0081] In this application, the model's detection results for the system status include normal system and abnormal system. In order to reduce the amount of calculation of the system where the model is located, and to reduce the ineffective attention of the staff to the system, when the detection result output by the model is normal, it can be saved only in the preset storage location inside the system, and the system status is not output to the outside, that is, the system is in normal production when there is no output result by default; on the contrary, when the output system status is detected, it indicates that there is an abnormality in the industrial Internet system, so that the staff can quickly pay attention to the abnormal state of the industrial Internet system, realize rapid maintenance, and reduce production costs and safety risks.

[0082] Optionally, in order to reduce the amount of data stored in the model, the normal detection results of the system stored in the model can be cleaned up regularly.

[0083] In this application, the abnormal system status can be output to the alarm device. Specifically, the alarm device may include a dashboard and an alarm set in the industrial Internet system; wherein the dashboard and the alarm may be connected to the industrial Internet system via a wired or wireless network alarm; the dashboard display content includes early warning information, task information, and notification information, and the system status is displayed based on the collected data. When the system is abnormal, the alarm information is displayed through the dashboard, and the alarm is sounded through the alarm to remind the staff that the system is abnormal, and to quickly troubleshoot and handle the abnormality.

[0084] The system status detection method of the industrial Internet system provided by the present application obtains the system data of the industrial Internet system; uses the system status detection model to detect the system data to obtain the system status of the industrial Internet system; wherein the system status detection model includes a shallow feature extraction module and a plurality of deep feature detection modules connected in sequence, and the system status detection model is obtained by training a dynamic neural network model with a training data set, and the training data set includes the historical system data of the industrial Internet system; the system status includes the system being normal and the system being abnormal; when the system status is the system being abnormal, the system status of the industrial Internet system is output; in the above scheme, shallow feature extraction is performed on the collected system data, and then When multiple deep feature detection modules connected in sequence are used for state detection, a shallower deep feature detection module is first used to perform deep feature extraction and state detection, and the detection is terminated directly when the detection result is satisfactory; secondly, a deeper detection module is used to continue the detection when the detection result is unsatisfactory; in this way, the detection is terminated when a satisfactory detection result can be obtained by using a shallower deep feature detection module, so as to reduce the use of a full-depth detection module for state detection, reduce the high computational complexity of the model, and realize the deployment of the model for state detection under the condition of insufficient computing power on the edge side, and realize the state detection model for state detection of the industrial Internet system, thereby realizing efficient detection of the network state of the industrial Internet.

[0085] Next, the model structure of the system status detection model used in the detection is introduced in detail. Figure 2 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 1 See also Figure 2In this model: the input end of the shallow feature extraction module is the input end of the system state detection model, and the output end of the shallow feature extraction module is connected to the input end of the first deep feature detection module; the first output end and the second output end of the deep feature detection module of the first level are connected to the input end of the deep feature detection module of the next level; the third output end of the deep feature detection module of the first level is the output end of the system state detection model;

[0086] Furthermore, for any intermediate-level deep feature detection module, its input end is connected to the first output end and the second output end of the deep feature detection module of the previous level, and the first output end and the second output end of the deep feature detection module of the current level are also commonly connected to the input end of the deep feature detection module of the next level; the third output end of the deep feature detection module of the current level can also be the output end of the system state detection model.

[0087] It should be understood that, for the deep feature detection module of the last level, its input end is also connected to the first output end and the second output end of the deep feature detection module of the previous level. However, since there is no deep feature detection module of the next level, the deep feature detection module of the last level does not have the first output end and the second output end, and only uses the third output end as the output end of the system state detection model; in other words, its first output end and the second output end do not output data, and only outputs the final detection result of the model through the third output end.

[0088] In the present application, a system state detection model is used to detect system data to obtain the system state of the industrial Internet system, including: inputting the shallow features corresponding to the system data into a shallow feature extraction module to obtain the corresponding shallow feature extraction data; for any level of deep feature detection module except the last level of deep feature detection module, the shallow feature extraction module, or the shallow feature data output by the deep feature detection module of the previous level, is input into the deep feature detection module of the current level for deep feature extraction, and the extracted deep feature data is subjected to state detection to obtain the corresponding system state, and the confidence corresponding to the system state is obtained; if the confidence does not meet the preset confidence threshold condition, the deep feature data of the current level is output as the shallow feature data of the deep feature detection module of the next level; if the confidence meets the confidence threshold condition, the detection state of the current level is output as the system state corresponding to the industrial Internet system.

[0089] Specifically, the output result of the shallow feature extraction module is the shallow feature data corresponding to the system data.

[0090] For any level of deep feature detection module except the last level of deep feature detection module, the shallow feature data input into the deep feature extraction module of the current level is subjected to deep feature extraction and state detection processing. During the processing, the first output end outputs the deep feature data corresponding to the shallow feature data in the deep feature extraction module of the current level, the second output end is the output channel corresponding to the detection result corresponding to the deep feature data when the preset condition is not satisfied, and the third output end is the output channel corresponding to the detection result corresponding to the deep feature data when the preset condition is satisfied.

[0091] In other words, if the processing result of the deep feature detection module in the current level meets the preset conditions, the detection result output by it, that is, the system state is output as the output result of the system state detection model; conversely, if the processing result of the deep feature detection module in the current level does not meet the preset conditions, the corresponding detection result and the deep feature data extracted at the current level are input into the deep feature detection module at a deeper level to continue the feature extraction and detection processing until the processing result of the deep feature detection module at the deeper level meets the preset conditions, then the detection result is output and the detection is stopped.

[0092] Figure 3 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 2 . See Figure 3 On the basis of the above, in the model: the deep feature detection module of any level except the deep feature detection module of the last level includes a deep feature extraction and state detection submodule, and a detection result verification submodule; wherein, the input end of the deep feature extraction and state detection submodule is the input end of the deep feature detection module; the first output end of the deep feature extraction and state detection submodule is connected to the input end of the detection result verification submodule; the second output end of the deep feature extraction and state detection submodule and the first output end of the detection result verification submodule serve as the first output end and the second output end of the deep feature detection module respectively; the second output end of the detection result verification submodule is the third output end of the deep feature detection module.

[0093] In the present application, the deep feature extraction and state detection submodule is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module, or the deep feature detection module of the previous level, and perform state detection on the extracted deep feature data to obtain the corresponding system state; the detection result review submodule obtains the confidence level corresponding to the system state; wherein, if the confidence level does not meet the preset confidence threshold condition, the deep feature data of the current layer is used as the shallow feature data of the deep feature detection module of the next layer; if the confidence level meets the confidence threshold condition, the system state corresponding to the industrial Internet system is obtained.

[0094] Specifically, the deep feature extraction and state detection submodule performs deep feature extraction on the feature data output by the deep feature detection module of the previous level, or the shallow feature extraction module, to obtain the deep feature data of the current level, and implements state detection based on the deep feature data to obtain the detection result output by the current level.

[0095] The detection result review submodule determines whether the detection result output by the current level meets the preset threshold; optionally, if it does, the detection result output by the current level is directly used as the final detection result of the model, and the detection result is output through the second output end of the detection result review submodule; on the contrary, if it does not meet the threshold, the deep feature data and the corresponding detection result output by the deep feature extraction and state detection submodule are output through the first output end of the detection result review submodule, so that the deep feature detection module of the next level can continue processing.

[0096] Figure 4 A schematic diagram of a system status detection model provided in an embodiment of the present application Figure 3 . See Figure 4 On the basis of the above, in the model: the deep feature extraction and state detection submodule includes a deep feature extraction layer and a state detection layer; the input end of the deep feature extraction layer is the input end of the deep feature extraction and state detection submodule, and the first output end of the deep feature extraction layer is connected to the input end of the state detection layer; the second output end of the deep feature extraction layer is the second output end of the deep feature extraction and state detection submodule; the output end of the state detection layer is the first output end of the deep feature extraction and state detection submodule.

[0097] In the present application, the deep feature extraction layer is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module, or the deep feature detection module of the previous level, to obtain deep feature data; the state detection layer is used to perform state detection on the extracted deep feature data to obtain the corresponding system state.

[0098] Specifically, the deep feature extraction module performs deep feature extraction on the input data to obtain deep feature data. The deep feature data is used to perform state detection in the deep feature detection module of the current level. On the other hand, when the detection result output by the deep feature detection module of the current level does not meet the preset conditions, it is used as basic data and input into the deep feature detection module of the next level to continue feature extraction and state detection.

[0099] The state detection layer performs state detection processing based on the deep feature data extracted by the deep feature extraction layer to obtain the detection result. The detection result is output to the detection result verification submodule in the deep feature detection module of the current level for judgment to determine whether the detection result meets the preset conditions, and then determine whether the state detection of the current group of data can output the result.

[0100] The above network structure of the model, on the one hand, can control the deep feature data generated at the current level to be output to the deep feature detection module at the next level when the detection result output by the deep feature detection module at the current level does not meet the preset conditions, so as to further perform deep feature extraction and detection processing and improve the accuracy of the detection results. On the other hand, when the detection results that meet the preset conditions are obtained, they can be directly output as the detection results of the model, thereby reducing unnecessary security overhead as much as possible while ensuring the prediction accuracy, thereby improving the detection efficiency of the system and reducing production costs and safety risks.

[0101] In the present application, the above-mentioned system status detection model can be a machine learning module built based on a dynamic neural network. Through the "early exit" mechanism in the dynamic neural network, it can decide whether to end the task in advance according to the model's satisfaction with the output results, avoid deeper processing, and avoid redundant calculations by selectively skipping some structures. While ensuring the accuracy of the output results, the amount of calculation of the system where the model is located is reduced, and the model can be deployed for status detection under the condition of insufficient computing power on the edge side. A status detection model for performing status detection on the industrial Internet system can be implemented, thereby achieving efficient detection of the network status of the industrial Internet.

[0102] In this application, based on the initial model built based on the dynamic neural network, a training data set can be used to train the constructed dynamic neural network model to obtain a system state detection model. In this application, the training data set can be, and the training data set includes historical system data of the industrial Internet system.

[0103] Since training multiple deep feature detection modules simultaneously may cause interference between them, in order to avoid interference, the multiple deep feature detection modules included in the model can be trained one by one based on the training data set.

[0104] It should be noted that in the actual production process, since there are fewer abnormal data in the industrial Internet system and the industrial Internet data is difficult to label, the dynamic neural network model of the present invention adopts unsupervised learning to obtain historical device data of each device in the system under normal working conditions to construct a training data set, and based on the training data set, the shallow feature extraction module and each deep feature detection module of the dynamic neural network model are trained one by one to obtain a trained dynamic neural network model, that is, a system status detection model.

[0105] During the training process, first, the shallow feature extraction module and the first deep feature detection module can be trained based on the training data set, and the loss is calculated according to the output of the first deep feature detection module. Then, the parameters of the shallow feature extraction module and the first deep feature detection module are updated by back propagation until the training requirements are met; secondly, the parameters of the shallow feature extraction module and the first deep feature detection module are fixed, and the second deep feature detection module is trained based on the training data set, and the loss is calculated according to the output of the second deep feature detection module; then, the parameters of the second deep feature detection module are updated by back propagation until the training requirements are met, and so on, until all the deep feature detection modules are trained to obtain a trained system state detection model.

[0106] Next, we will take the training of any level of deep feature detection module as an example to illustrate the training process of the model.

[0107] During the training process, historical system data in the training data set is obtained. In this application, the historical system data is multi-dimensional time series data. The historical system data can be expressed as In the form of; wherein, L represents the data length, K represents the data dimension; it can be understood that the data dimension represents the type of data.

[0108] On the one hand, in order to realize the network structure of the deep feature detection module in the matching model, on the other hand, in order to obtain richer features in the data, the data feature extraction module can be used to extract the shallow features corresponding to the historical system data.

[0109] In the process of shallow feature extraction, the data X is first processed by a one-dimensional convolution kernel to transform the dimension of the historical system data from dimension K to dimension d. model , and obtain data TE(X).

[0110] Secondly, in order to provide the position information of different time steps in the historical system data, the position encoding mechanism is used to perform position encoding processing on the data X to obtain the data PE(X).

[0111] Exemplarily, the following expression is used to obtain the position code corresponding to the historical system data at each time point:

[0112] PE pos,2i =sin(pos / 5000 2i / d )

[0113] PE pos,2i+1 =cos(pos / 5000 2i / d ),i=1,…,d / 2

[0114] Among them, d represents the dimension after dimensionality reduction, pos represents the position of the time point in the sample, and PE pos，2i Indicates the value of the even-numbered bit in the position code, PE pos，2i+1 Represents the value of the odd-numbered bits in the position encoding.

[0115] Finally, the position encoding and dimension transformation results are fused to obtain the shallow feature data Z 0 .

[0116] For example, the process of data fusion can refer to the following expression:

[0117]

[0118] In this application, the shallow feature data output by the shallow feature extraction module is input into the deep feature detection module for deeper feature extraction processing to obtain the deep feature extraction feature Z l On the one hand, the deep feature data is output to the deep feature detection module of the next level, and on the other hand, the state detection is performed in the state detection layer of the current level, and the obtained detection results are reviewed in the detection result review submodule.

[0119] In this application, deep feature extraction in the deep feature detection module can be implemented by the Anomaly Attention mechanism. Exemplarily, in this mechanism, deep feature data can be obtained by the following expression:

[0120]

[0121]

[0122]

[0123] Among them, Softmax(·) represents the Softmax function, Z l represents the deep feature data output by the deep feature detection module at the lth level, d represents the dimension of the deep feature data, represents the Q weight matrix of the deep feature detection module at the lth level, represents the K weight matrix of the deep feature detection module at the lth level, V represents the weight matrix of the deep feature detection module at the lth level. Q, K and V represent the Query matrix, Keys matrix and Values ​​matrix required for calculating attention, respectively. S l Represents the sequence association between the deep feature data corresponding to the historical system data.

[0124] On this basis, the prior correlation P between deep feature data can also be calculated lFor example, the astaxanthin expression can be used to calculate the prior association:

[0125]

[0126] Among them, P 1 represents the prior association between the deep feature data in the deep feature detection module at the lth level, σ represents the scale parameter matrix, σ i represents the scale parameter corresponding to the i-th historical system data, L represents the number of data points of the training sample, and Rescale(·) represents the Rescale function.

[0127] Specifically, through serial association and prior association, the temporal dependencies in historical system data can be mined, so that the model can adaptively capture the most effective associations, facilitate accurate extraction of data features, and provide a basis for subsequent detection.

[0128] It should be understood that the above-mentioned implementation modes are only optional implementation modes exemplarily introduced in the technical solution of the present application, and are not intended to limit the technical solution of the present application. The present application can also adopt any open source method for deep feature extraction.

[0129] On the basis of obtaining the deep feature data, the system status of the industrial Internet system is detected according to the deep feature data to obtain the predicted detection results output by the deep feature detection module at the current level.

[0130] Specifically, the predicted test results can be obtained by the following expression:

[0131]

[0132] Among them, X l It represents the predicted detection result output by the deep feature detection module at the lth level. Layer-Normal(·) represents layer normalization, and Feed-Forward(·) represents the feedforward unit.

[0133] On this basis, the loss function in the training process is constructed according to the above feature data, and the shallow feature extraction module and the l-th level deep feature detection module are trained through the loss function.

[0134] The loss functions used in training in this application include: regression loss, association difference loss and feature disentanglement loss; the association difference loss includes a priori association loss and serial association loss; among them, the regression loss represents the difference between the predicted system state and the historical system state of the historical system data output by the dynamic neural network model during the training process; the a priori association loss represents the attention weight distribution of the corresponding rows of the historical system data in the attention matrix; the serial association loss represents the association between the deep feature data corresponding to the current data point and the domain data point in the historical system data; the feature disentanglement loss represents the degree of correlation between the deep feature data corresponding to different data points in the historical system data.

[0135] Specifically, the expression of the loss function can be obtained based on the following expression:

[0136]

[0137] in, represents the loss function of the model, represents the regression loss of the model; X represents the training data set, that is, the historical system data, represents the reconstruction result output by the deep feature detection module at the lth level, ||·||F represents the F-norm, and ||·||1 represents the 1-norm; AssDis l represents the associated discrepancy loss of the model; represents the feature disentanglement loss, and λ represents the parameter.

[0138] It should be understood that, compared with normal data, it is difficult for abnormal data to establish a strong correlation with the entire sequence dominated by the normal pattern, and they tend to pay more attention to the adjacent areas (due to continuity). Therefore, this difference in correlation with the overall sequence and the adjacent priors provides a natural and highly discriminative criterion for anomaly detection.

[0139] Based on this, in this application, the smaller the associated difference loss AssDis is, the greater the probability that the detection result is a system abnormality; the larger the reconstruction error is, the greater the probability that the detection result is a system abnormality.

[0140] Specifically, the association difference loss in this application includes a priori association and sequence association. Specifically, the association difference loss can be obtained based on the following expression:

[0141]

[0142] in, Represents the prior association of the i-th data in the l-th level deep feature detection module; represents the sequence association of the i-th data in the deep feature detection module of the l-th level, L represents the number of data points of the training sample, and KL(·||·) represents the KL distance.

[0143] It should also be understood that the prior association uses a learnable Gaussian kernel function, centered on the index of the corresponding time point. This design can take advantage of the unimodal characteristics of the Gaussian distribution to make the data pay more attention to nearby points. At the same time, in order to make the prior association adaptable to different time series patterns, the Gaussian kernel function contains a learnable scale σ parameter. The sequence association is obtained by the attention calculation in the standard Transformer. The sequence association of a point is the distribution of the attention weights of the corresponding rows of the point in the attention matrix. This branch is to mine the associations in the original sequence and allow the model to adaptively capture the most effective associations.

[0144] The correlation difference is a criterion for measuring anomalies. The prior correlation adopts a normal distribution with a unimodal characteristic. When an anomaly occurs, the adjacent points around the current point are very likely to have an anomaly. Therefore, both the prior correlation and the sequence correlation focus on local information, so there is a small difference between the two correlations. On a normal sequence, the attention map mechanism has a global focus characteristic and will be distributed on non-adjacent points, such as a periodic sequence. Compared with the unimodal prior correlation, there is a big difference between the two correlations.

[0145] It should also be understood that time series data includes multiple dimensions, and the features of some dimensions may be highly correlated. The features of related dimensions usually contain redundant information. This situation of containing redundant information is generally called feature entanglement.

[0146] In this application, when the model maps deep feature data to the reconstructed output, information redundancy may cause deviations in the anomaly detection results. In order to solve the problem that this situation may cause poor model results, the technical solution provided in this application also designs a loss function L for characterizing feature disentanglement during model training. DAE , this loss function is only calculated in the training phase, so it does not increase the computational complexity of the model in the application phase.

[0147] Specifically, to achieve feature untangle operation, we can first calculate the Pearson correlation coefficient between any two feature dimensions, so that the correlation between the two variables can be measured between -1 and 1, and then calculate the second-order moment S corresponding to the correlation coefficient of the same dimension. main The second-order moment S corresponding to the correlation coefficient of different dimensions other , in order to reduce the overall correlation between different feature dimensions and avoid the influence of different numbers of summed elements.

[0148] Exemplarily, the following expression can be used in this application to calculate the feature disentanglement loss:

[0149]

[0150]

[0151]

[0152] Among them, ρ(·,·) represents the Pearson correlation coefficient, and d represents the number of dimensions of the feature data; represents the feature data of the i-th dimension extracted by the deep feature detection module at the l-th level, Represents the features of the j-th dimension extracted by the deep feature detection module at the l-th level.

[0153] In this application, by calculating the feature disentanglement loss, the gap between the correlation of the selected feature dimension itself and its correlation with other feature dimensions can be expanded, thereby reducing the entangled distribution of statistical features, achieving more accurate reconstruction, and then improving the sensitivity to feature changes caused by different types of attacks.

[0154] On the basis of the system status detection model obtained after training, the system status detection model is used to detect the system data to obtain the system status corresponding to the industrial Internet system. The process may specifically include: extracting shallow features of the system data to obtain shallow feature data; executing the following feature extraction and status detection steps until a system status that meets preset conditions is obtained, and then determining the system status corresponding to the industrial Internet system; performing deep feature extraction on the shallow feature data of the current layer to obtain the deep feature data corresponding to the current layer; the shallow features include the shallow feature data obtained after the shallow feature extraction of the system data, or the deep feature data of the previous layer; performing status detection on the deep feature data of the current layer to obtain the corresponding system status, and the confidence corresponding to the system status; if the confidence does not meet the preset confidence threshold condition, the deep feature data of the current layer is used as the shallow feature data of the next layer; if the confidence meets the confidence threshold condition, the system status corresponding to the industrial Internet system is obtained.

[0155] Specifically, the shallow feature extraction module in the model is used to perform shallow feature extraction on the input system data to obtain shallow feature data; and the shallow feature data is input into the deep feature detection module for deep feature extraction and state detection processing.

[0156] Specifically, deep-level feature extraction is performed on shallow feature data to obtain deep feature data, and state detection is performed based on the deep feature data to obtain a state detection result; further, if the state detection result meets the preset conditions, it is output as the final output result of the model; conversely, if the state detection result does not meet the preset conditions, the detection result and the deep feature data are output to the next level of deep feature detection module to continue feature extraction and detection processing until the detection result is output.

[0157] In the present application, the confidence level corresponding to the detection result can be calculated, and whether the detection result meets the preset conditions can be determined by using a preset confidence threshold and the confidence level corresponding to the detection result.

[0158] Specifically, the anomaly score corresponding to the detection result can be calculated in advance, and the confidence can be obtained by calculating the anomaly score. It should be noted that the anomaly score is a vector with a length of L, and each element represents the probability that the corresponding data point in the sample is an anomaly point.

[0159] Furthermore, each element in the Anomaly Score vector is compared with the anomaly threshold. If it is greater than or equal to the anomaly threshold, it is marked as an abnormal data point, otherwise it is marked as a normal data point. Then the confidence is obtained according to the proportion of abnormal data points, that is, the confidence is obtained by calculating the ratio of the number of abnormal data points to the number of elements in the anomaly score. Considering the sparsity of abnormal data in the anomaly detection task, the evaluation performance of the model will be more significantly improved if all abnormal points are found as much as possible. Based on this, a simple strategy is to count the proportion of abnormal points in the predicted sequence. If the proportion of predicted abnormal points exceeds the exit threshold, exit and output the current anomaly detection result, otherwise enter the next layer of processing. However, doing so will cause a large number of samples without abnormal points to enter deeper models, which actually leads to a waste of computing resources.

[0160] In order to simultaneously increase the recall of the model for outliers and reduce the proportion of outliers as much as possible, we proposed a dynamic exit mechanism based on bilateral thresholds. The confidence level is based on the bilateral threshold to determine whether to exit early.

[0161] In the present application, the confidence threshold condition includes a first threshold and a second threshold; the first threshold is less than the second threshold; the confidence does not meet the confidence threshold condition including: the confidence is less than the first threshold, or the confidence is greater than the second threshold; the confidence meets the confidence threshold condition including: the confidence is greater than the first threshold, and the confidence is less than the second threshold.

[0162] Exemplarily, the first threshold and the second threshold are 0.2 and 0.8 respectively; optionally, if the confidence level corresponding to the detection result is 0.1, it means that the detection result meets the threshold condition; optionally, if the confidence level corresponding to the detection result is 0.6, it means that the detection result does not meet the preset condition; optionally, if the confidence level corresponding to the detection result is 0.9, it means that the detection result meets the preset condition.

[0163] In this application, the collected system data is input into the system state detection model. After the detection result is output in the deep feature detection module of the model, the abnormal score of the detection result is calculated, and then the confidence is calculated to determine whether to exit early according to the confidence, that is, output the detection result; optionally, for the confidence between the bilateral thresholds, it is considered to be complex data, so the next deep feature detection module can be used to continue monitoring, that is, not exit early; on the contrary, the judgment result is credible, so it exits early and obtains the detection result; in this way, it is possible to reduce the security overhead of the system while ensuring the detection efficiency. If the detection process is propagated to the last deep feature detection module, the abnormal score output by the last deep feature detection module is the detection result; thereby solving the problem of low state detection efficiency of the existing industrial Internet system and affecting production efficiency, improving the detection efficiency of the system, and reducing production costs and safety risks.

[0164] Figure 5 A schematic diagram of the structure of a system status detection system for an industrial Internet system provided in an embodiment of the present application. Figure 5 The system 50 includes: a system data acquisition subsystem 510, a system status detection subsystem 520 and a system status alarm subsystem 530; wherein,

[0165] The system data acquisition subsystem 510 is used to collect system data of the industrial Internet system and transmit the system data to the system status detection subsystem;

[0166] The system status detection subsystem 520 is used to execute the system status detection method of the industrial Internet system in the above-mentioned embodiment;

[0167] The system status alarm subsystem 530 is used to alarm the system status output by the system status detection subsystem.

[0168] In an optional implementation, the system status detection subsystem 520 detects system data using a system status detection model to obtain the system status of the industrial Internet system, specifically for:

[0169] Inputting the shallow features corresponding to the system data into the shallow feature extraction module to obtain the corresponding shallow feature extraction data;

[0170] For any level of deep feature detection module except the last level of deep feature detection module, the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous level is input into the deep feature detection module of the current level to extract deep features, and the extracted deep feature data is subjected to state detection to obtain the corresponding system state, and the confidence corresponding to the system state is obtained;

[0171] If the confidence does not meet the preset confidence threshold condition, the deep feature data of the current level is output as the shallow feature data of the deep feature detection module of the next level;

[0172] If the confidence meets the confidence threshold condition, the detection status of the current level is output as the system status corresponding to the industrial Internet system.

[0173] In an optional implementation, any level of deep feature detection module except the last level of deep feature detection module includes deep feature extraction and state detection submodules, and a detection result review submodule; wherein,

[0174] The deep feature extraction and state detection submodule is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous level, and perform state detection on the extracted deep feature data to obtain the corresponding system state;

[0175] The detection result review submodule obtains the confidence level corresponding to the system status; if the confidence level does not meet the preset confidence threshold condition, the deep feature data of the current layer is used as the shallow feature data of the next layer of deep feature detection module; if the confidence level meets the confidence threshold condition, the system status corresponding to the industrial Internet system is obtained.

[0176] In an optional implementation, the deep feature extraction layer is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous layer to obtain deep feature data;

[0177] The state detection layer is used to perform state detection on the extracted deep feature data to obtain the corresponding system state.

[0178] In an optional implementation, the confidence threshold condition includes a first threshold and a second threshold; the first threshold is less than the second threshold;

[0179] The confidence level not meeting the confidence level threshold condition includes: the confidence level is less than a first threshold level, or the confidence level is greater than a second threshold level;

[0180] The confidence level satisfies the confidence level threshold condition, including: the confidence level is greater than a first threshold level, and the confidence level is less than a second threshold level.

[0181] In an optional implementation, the system state detection subsystem 520 is further used to: use a training data set to train the constructed dynamic neural network model to obtain a system state detection model; the loss function used in the training includes: regression loss, association difference loss and feature disentanglement loss; the association difference loss includes a priori association loss and sequence association loss; wherein,

[0182] The regression loss represents the difference between the reconstruction result of the historical system data and the historical system data output by the dynamic neural network model during the training process;

[0183] The prior association loss represents the attention weight distribution of the corresponding rows of the historical system data in the attention matrix;

[0184] The sequence association loss characterizes the association between the deep feature data corresponding to the current data point and the domain data point in the historical system data;

[0185] Feature disentanglement loss characterizes the degree of correlation between the deep feature data corresponding to different data points in the historical system data.

[0186] Figure 6 It is a block diagram of an electronic device provided in an embodiment of the present application, and the device may be a computer, a digital broadcast terminal, etc.

[0187] See also Figure 6 , the device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output interface 812 , a sensor component 814 , and a communication component 816 .

[0188] The processing component 802 generally controls the overall operation of the device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0189] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0190] The power supply component 806 provides power to the various components of the device 800. The power supply component 806 can include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 800.

[0191] The multimedia component 808 includes a screen that provides an output interface between the device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0192] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the device 800 is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0193] The input / output interface 812 provides an interface between the processing component 802 and the peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0194] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the device 800, and the sensor assembly 814 can also detect the position change of the device 800 or a component of the device 800, the presence or absence of user contact with the device 800, the orientation or acceleration / deceleration of the device 800, and the temperature change of the device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a solid image (Complementary Metal Oxide Semiconductor, CMOS) sensor or a semiconductor image (Charge-coupled Device, CCD) sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0195] The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and other devices. The device 800 can access a wireless network based on a communication standard, such as WiFi, 4G or 5G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0196] In an exemplary embodiment, the device 800 may be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0197] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by a processor 820 of the device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0198] A non-temporary computer-readable storage medium, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to execute the system status detection method of the above-mentioned industrial Internet system.

[0199] An embodiment of the present application also provides a chip for running instructions, which is used to execute the technical solution of the system status detection method of the industrial Internet system in the above embodiment.

[0200] An embodiment of the present application also provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed on a computer, the computer executes the technical solution of the system status detection method of the industrial Internet system in the above embodiment.

[0201] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the system status detection method of the industrial Internet system in the above embodiment.

[0202] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0203] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

[0204] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0205] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A system status detection method for an industrial Internet system, characterized in that: The method comprises: Obtain system data of industrial Internet systems; The system data is detected by using a system status detection model to obtain the system status of the industrial Internet system; wherein the system status detection model includes a shallow feature extraction module and a plurality of deep feature detection modules connected in sequence, and the system status detection model is obtained by training a dynamic neural network model using a training data set, and the training data set includes historical system data of the industrial Internet system; the system status includes a normal system and an abnormal system; When the system status is a system abnormality, outputting the system status of the industrial Internet system; Wherein, using the system status detection model to detect the system data to obtain the system status of the industrial Internet system includes: Inputting the system data into the shallow feature extraction module to obtain corresponding shallow feature extraction data; For any level of deep feature detection module except the last level of deep feature detection module, the shallow feature data output by the shallow feature extraction module, or the deep feature detection module of the previous level, is input into the deep feature detection module of the current level to perform deep feature extraction, and the extracted deep feature data is subjected to state detection to obtain the corresponding system state, and the confidence corresponding to the system state is obtained; If the confidence does not meet the preset confidence threshold condition, outputting the deep feature data of the current level as shallow feature data of the deep feature detection module of the next level; If the confidence level satisfies the confidence level threshold condition, outputting the detection status of the current level as the system status corresponding to the industrial Internet system; Among them, the deep feature detection module of any level except the deep feature detection module of the last level includes deep feature extraction and state detection submodules, and a detection result review submodule; wherein, The deep feature extraction and state detection submodule is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous level, and perform state detection on the extracted deep feature data to obtain the corresponding system state; The detection result review submodule obtains the confidence level corresponding to the system status; if the confidence level does not meet the preset confidence threshold condition, the deep feature data of the current layer is used as the shallow feature data of the next layer of deep feature detection module; if the confidence level meets the confidence threshold condition, the system status corresponding to the industrial Internet system is obtained.

2. The method according to claim 1, characterized in that: The deep feature extraction and state detection submodule includes a deep feature extraction layer and a state detection layer; The deep feature extraction layer is used to perform deep feature extraction on the shallow feature data output by the shallow feature extraction module or the deep feature detection module of the previous layer to obtain deep feature data; The state detection layer is used to perform state detection on the extracted deep feature data to obtain the corresponding system state.

3. The method according to claim 1 or 2, characterized in that: The confidence threshold condition includes a first threshold and a second threshold; the first threshold is less than the second threshold; The confidence level not meeting the confidence level threshold condition includes: the confidence level is less than the first threshold level, or the confidence level is greater than the second threshold level; The confidence level satisfies the confidence level threshold condition, including: the confidence level is greater than the first threshold level, and the confidence level is less than the second threshold level.

4. The method according to claim 1 or 2, characterized in that: The method further comprises: The constructed dynamic neural network model is trained using a training data set to obtain the system state detection model; the loss functions used in the training include: regression loss, association difference loss and feature disentanglement loss; the association difference loss includes prior association loss and sequence association loss; wherein, The regression loss represents the difference between the reconstruction result of the historical system data and the historical system data output by the dynamic neural network model during the training process; The prior association loss represents the attention weight distribution of the corresponding row of the historical system data in the attention matrix; The sequence association loss represents the association between the deep feature data corresponding to the current data point and the domain data point in the historical system data; The feature disentanglement loss represents the degree of correlation between the deep feature data corresponding to different data points in the historical system data.

5. A system status detection system for an industrial Internet system, characterized in that: The system includes: a system data acquisition subsystem, a system status detection subsystem and a system status alarm subsystem; wherein, The system data acquisition subsystem is used to collect system data of the industrial Internet system and transmit the system data to the system status detection subsystem; The system status detection subsystem is used to execute the system status detection method of the industrial Internet system described in any one of claims 1 to 4; The system status alarm subsystem is used to alarm the system status output by the system status detection subsystem.

6. An electronic device, characterized in that: include: A processor and a memory communicatively connected to the processor; The memory stores computer-executable instructions; When executing the computer execution instruction, the processor is used to implement the system status detection method of the industrial Internet system as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer execution instructions, which, when executed by a processor, are used to implement the system status detection method for an industrial Internet system as described in any one of claims 1 to 4.

8. A computer program product, characterized in that The computer program product includes a computer program, which, when executed by a processor, implements the system status detection method of the industrial Internet system as described in any one of claims 1 to 4.

Images