DCS early warning method and system based on intelligent AI visual identification

Abstract

The invention provides a DCS early warning method and system based on intelligent AI visual identification, and relates to the technical field of DCS early warning, and the method comprises the steps: obtaining the real-time operation parameters of a DCS system, and obtaining the real-time operation parameter characteristics of a unified format; inputting the real-time operation parameter characteristics into a pre-constructed intelligent AI visual identification model, analyzing the operation state of the DCS system, identifying an abnormal mode of a system level, and comparing the abnormal mode with a normal operation mode in a preset normal operation mode library; a digital twinborn simulation technology based on a physical mechanism is adopted to predict a fault mode, a fault reason and an occurrence probability of the DCS system, early warning information of a corresponding level is automatically generated according to a prediction result in combination with a preset multi-level early warning decision matrix, and the early warning information is pushed to an intelligent operation and maintenance platform of the DCS system in real time. An operator is assisted in recognizing system risks in advance, and targeted preventive maintenance measures and suggestions are given.

Description

Technical Field

[0001] The present invention relates to DCS early warning technology, and in particular to a DCS early warning method and system based on intelligent AI visual recognition. Background Art

[0002] DCS (distributed control system) is widely used in industrial fields such as petroleum, chemical industry, and electric power to achieve real-time monitoring and control of production processes. However, DCS systems are large in scale, complex in structure, have a harsh operating environment, and have a high risk of failure. Once a failure occurs, it will cause significant economic losses and safety accidents. Traditional DCS system early warning methods mainly rely on manual experience and simple threshold judgments, which are difficult to cope with increasingly complex industrial scenarios, and the accuracy and real-time nature of early warnings cannot be guaranteed. In recent years, the development of artificial intelligence technology has provided a new way for intelligent early warning of DCS systems. Some studies have attempted to apply machine learning algorithms to fault diagnosis and prediction of DCS systems, such as support vector machines and neural networks, and have achieved certain results. However, these methods are mainly aimed at single devices or subsystems, lack macro analysis of the entire DCS system, and are difficult to discover abnormal patterns at the system level. In addition, the complexity and variability of industrial big data bring challenges to data processing and feature extraction for intelligent early warning.

[0003] Therefore, there is an urgent need for a new DCS system early warning method that can make full use of industrial big data and artificial intelligence technology, identify abnormal patterns at the system level, and combine digital twin simulation to predict failure risks, generate intelligent early warning information, and assist operation and maintenance personnel to make preventive maintenance decisions in a timely manner, thereby improving the safety and reliability of the DCS system. Summary of the invention

[0004] The embodiments of the present invention provide a DCS early warning method and system based on intelligent AI visual recognition, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provided is a DCS early warning method based on intelligent AI visual recognition, including:

[0007] Acquire the real-time operation parameters of the DCS system, wherein the real-time operation parameters include the operation status data, alarm information, operation log and video monitoring data of each subsystem of the DCS system, wherein the operation status data includes the real-time process parameters and equipment operation parameters of each subsystem, and pre-process the real-time operation parameters to obtain the real-time operation parameter features in a unified format;

[0008] Input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in a preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode;

[0009] If the similarity is lower than the preset abnormal threshold, it is determined that there is an abnormal risk in the current operation of the DCS system, and the digital twin simulation technology based on physical mechanisms is used to simulate the real-time operation scenario of the DCS system to predict the failure mode, failure cause and probability of occurrence of the DCS system. Based on the prediction results and the preset multi-level early warning decision matrix, the corresponding level of early warning information is automatically generated and pushed to the intelligent operation and maintenance platform of the DCS system in real time to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

[0010] In an optional embodiment,

[0011] The real-time operating parameter features are input into the pre-built intelligent AI visual recognition model to analyze the operating status of the DCS system and identify abnormal patterns at the system level, including:

[0012] A multimodal feature fusion network is designed in the intelligent AI visual recognition model. The alarm information in the real-time operation parameter features is mined by text to obtain the alarm information features. The operation log is mined by sequence patterns to obtain the operation log features. The video surveillance data is subjected to target detection, target tracking and behavior recognition to obtain the video surveillance features.

[0013] The extracted alarm information features, operation log features and video surveillance features are input into the multimodal feature fusion network, and each feature is mapped to the latent space of the same dimension through the feature mapping method;

[0014] The attention mechanism is used to calculate the similarity between features in the latent space, and weights are assigned to each feature. The features in the latent space are fused using a hybrid expert network to obtain fused features. The fused features are compressed using an autoencoder to obtain the final multimodal fused features.

[0015] A heterogeneous graph is constructed according to the multimodal fusion features, wherein the nodes in the heterogeneous graph include devices, parameters, and alarms, and the edges include physical connections, logical associations, and timing dependencies. A graph neural network is applied to the heterogeneous graph, and the hidden states of the nodes are iteratively updated through a message passing mechanism. In each propagation step, the node first aggregates messages from neighboring nodes, and then updates the hidden states of the nodes in combination with its own features and neighbor information.

[0016] For the local anomaly detection task, the graph decoder is used to map the node hidden state into anomaly metrics, and the graph neural network corresponding to the local anomaly detection task is trained by optimizing the cross entropy loss function between the anomaly metric and the true anomaly label of the node; for the global anomaly association task, the graph pooling method based on anomaly metrics is used to gradually aggregate node information to generate an abnormal region hierarchy, and the graph decoder is used to map the abnormal region features into global anomaly metrics, and the graph neural network corresponding to the global anomaly association task is trained by optimizing the cross entropy loss function between the global anomaly metric and the true anomaly label at the system level;

[0017] The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results.

[0018] In an optional embodiment,

[0019] The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results, including:

[0020] A joint optimization loss function is constructed according to the loss function of the local anomaly detection subtask and the loss function of the global anomaly association subtask. The training steps of the local anomaly detection and global anomaly association subtasks are repeatedly executed until the joint optimization loss function converges to obtain the abnormal pattern recognition result at the system level. The formula of the joint optimization loss function is as follows:

[0021]

[0022] Among them, L represents the joint optimization loss function, λ1 represents the weight of the local anomaly detection task, V represents the number of nodes, and α i represents the weight factor of node i, represents the predicted anomaly metric of node i, y i represents the true anomaly label of node i, γ represents the exponential parameter of anomaly measurement error, λ2 represents the weight of the global anomaly association task, C represents the number of operation log features, and α c represents the weight factor of neighbor node c, y c represents the true abnormal label of the neighbor node c, represents the predicted abnormality measure of neighbor node c, λ3 represents the weight of the regularization term, C i represents the set of neighbor nodes of node i, and p represents the exponential parameter of the regularization term.

[0023] In an optional embodiment,

[0024] The digital twin simulation technology based on physical mechanism is used to simulate the real-time operation scenario of the DCS system, and predict the failure mode, cause and probability of the DCS system, including:

[0025] Construct a multidisciplinary high-fidelity digital twin model of the DCS system, which includes a production process model, an equipment model, and a control logic model;

[0026] Extract key feature parameters of abnormal patterns, use feature dimension reduction methods to reduce the dimensions of key feature parameters, and use semantic mapping technology to map the reduced key feature parameters to the corresponding positions and object attributes of the digital twin model, triggering the digital twin model to perform simulation predictions under abnormal working conditions;

[0027] The simulation prediction starts from the abnormal point, adopts a multi-scenario parallel simulation strategy, sets different initial conditions, boundary conditions and random disturbances, generates multiple parallel simulation instances, and adopts model simplification and multi-scale modeling methods to selectively simplify the digital twin model at different time and space scales, and performs grid encryption and calculation intensification for key areas and key processes. At the same time, uncertainty quantification technology is introduced to adaptively adjust the time and space resolution and calculation accuracy of the simulation;

[0028] Based on the simulation of multi-scenario abnormal working conditions, the abnormal mode is taken as the top event, and the deductive reasoning and causal reverse tracing methods are used to automatically construct the fault tree, identify the key links and influencing factors of abnormal propagation, and on the basis of the constructed fault tree, the Monte Carlo simulation method is used to randomly sample the probability of occurrence of basic events in the fault tree and evaluate the abnormal mode;

[0029] According to the results of fault tree analysis and Monte Carlo simulation, a case-based reasoning fault diagnosis method is used to measure the similarity between abnormal patterns and historical fault cases, match fault patterns from the historical fault library, and finally output the predicted fault mode, fault cause and occurrence probability of the DCS system.

[0030] In an optional embodiment,

[0031] Adopting a multi-scenario parallel simulation strategy, setting different initial conditions, boundary conditions and random disturbances, multiple parallel simulation instances are generated, including:

[0032] Obtain the historical abnormal data of the DCS system for preprocessing, extract the multi-scale spatiotemporal features of the historical abnormal data, including abnormal features at different time scales and spatial scales, and construct a multi-scale spatiotemporal feature representation of the abnormal scene;

[0033] Inputting the multi-scale spatiotemporal feature representation of the abnormal scene into a multi-scale conditional generative adversarial network, wherein the multi-scale conditional generative adversarial network includes a generator and a discriminator, wherein the generator receives random noise and conditional variables as inputs, and the discriminator receives real abnormal scene features and abnormal scene features generated by the generator as inputs;

[0034] In the generator, multi-scale feature extraction is performed on random noise and conditional variables to obtain noise features and conditional features of different scales. The attention mechanism is used to adaptively fuse the noise features and conditional features of different scales to generate abnormal scene features that meet the conditional constraints. In the discriminator, multi-scale feature extraction is performed on the input real abnormal scene features and the generated abnormal scene features to obtain real features and generated features of different scales. The attention mechanism is used to adaptively fuse the real features and generated features of different scales.

[0035] Constructing the objective function of a multi-scale conditional generative adversarial network, the objective function includes the logarithmic probability of the real abnormal scene features of the discriminator and the false probability of the abnormal scene features generated by the generator, and optimizing the training by maximizing the objective function of the discriminator and minimizing the objective function of the generator, and finally obtaining a trained abnormal scene distribution model;

[0036] Setting an abnormal type and severity condition of an abnormal scenario, encoding the abnormal type and severity condition into a conditional vector, connecting the conditional vector with a random noise vector to form a conditional noise vector, inputting the conditional noise vector into a generator of an abnormal scenario distribution model, and generating an abnormal scenario that meets the abnormal type and severity conditions through the generator;

[0037] Taking the diversity, coverage and abnormality degree of the abnormal scenarios as the optimization objectives, a fitness function is constructed, an adaptive sampling strategy is used to optimize the sampling of abnormal scenarios, and the sampling parameters are adaptively adjusted to obtain an optimized set of abnormal scenarios. The optimized set of abnormal scenarios is input into the digital twin model for multi-scenario parallel simulation.

[0038] In an optional embodiment,

[0039] The calculation formula for the objective function of constructing a multi-scale conditional generative adversarial network is as follows:

[0040]

[0041] in, represents the objective function of the multi-scale conditional generative adversarial network, R(D, G) represents the value function between the generator G and the discriminator D, and E f [·] represents the expectation of the characteristics of the real abnormal scene, E z[·] represents the expectation of the generated synthetic abnormal scene features, J represents the temporal scale level, K represents the spatial scale level, and a (D) represents the attention weight in the discriminator, f (j，k) Multi-scale representation of the real abnormal scene characteristics, s (j，k) represents the multi-scale representation of the conditional variable, a (G) represents the attention weight in the generator, z (j，k) Multi-scale representation of random noise.

[0042] In an optional embodiment,

[0043] According to the prediction results combined with the preset multi-level warning decision matrix, the corresponding level of warning information is automatically generated, including:

[0044] Construct an abnormal risk warning domain ontology, define the relationship mapping between abnormal types, influencing factors, warning levels and response measures, and form an abnormal risk warning ontology model;

[0045] Integrate structured data, semi-structured data and unstructured data, realize the association of multi-source heterogeneous data through data integration, and use the knowledge extraction model of deep learning to extract entities, attributes and relationships from unstructured data to form an abnormal risk warning knowledge graph;

[0046] Take the abnormal type and influencing factors output by abnormal risk prediction as known conditions, match the corresponding entity nodes in the consistent abnormal risk warning knowledge graph, use the abnormal risk warning ontology reasoning rules and graph reasoning algorithms to infer the concepts, attributes and relationships associated with the known conditions, and generate abnormal risk warning reasoning explanations;

[0047] Using the knowledge subgraph obtained by the abnormal risk warning reasoning interpretation, an abnormal risk warning question and answer template is constructed to form a structured problem description. The problem description is input into the natural language generation model based on Seq2Seq to automatically generate the recommended text of abnormal risk response measures.

[0048] Based on the confidence, coverage and novelty indicators of abnormal risk response measures knowledge, a knowledge sorting algorithm is used to prioritize the suggested texts of abnormal risk response measures. The abnormal risk level is calculated according to the abnormal prediction results, and the abnormal risk response measures of the corresponding level are selected to automatically generate warning information of the corresponding abnormal risk level.

[0049] According to a second aspect of the embodiments of the present invention,

[0050] Provide a DCS early warning system based on intelligent AI visual recognition, including:

[0051] The first unit is used to obtain the real-time operation parameters of the DCS system, wherein the real-time operation parameters include the operation status data, alarm information, operation log and video monitoring data of each subsystem of the DCS system, wherein the operation status data includes the real-time process parameters and equipment operation parameters of each subsystem, and pre-process the real-time operation parameters to obtain the real-time operation parameter features in a unified format;

[0052] The second unit is used to input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in a preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode;

[0053] The third unit is used to determine that there is an abnormal risk in the current operation of the DCS system if the similarity is lower than the preset abnormal threshold, and use digital twin simulation technology based on physical mechanisms to simulate the real-time operation scenario of the DCS system, predict the failure mode, cause and probability of the DCS system, and automatically generate warning information of the corresponding level based on the prediction results combined with the preset multi-level warning decision matrix, and push the warning information to the intelligent operation and maintenance platform of the DCS system in real time, to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

[0054] According to a third aspect of the embodiments of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] According to a fourth aspect of the embodiments of the present invention,

[0060] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0061] In this embodiment, the abnormal mode is identified from the system level through the intelligent AI visual recognition model, which avoids the limitations of manual experience and threshold judgment in the traditional method, and realizes the comprehensive perception and analysis of the operation status of the DCS system. Combined with the digital twin simulation technology of physical mechanism, the identified abnormal risks are predicted and verified, and the failure mode, cause and probability of occurrence are given, providing a reliable basis for early warning decision-making, and improving the interpretability and credibility of early warning information. A multi-level early warning decision matrix is ​​introduced to automatically generate early warning information of corresponding levels according to the severity and urgency of abnormal risks, realize intelligent classification and dynamic adjustment of early warning information, avoid insufficient or excessive early warning, and improve the pertinence and effectiveness of early warning. Through real-time docking with the intelligent operation and maintenance platform, the early warning information is pushed to the operator, and preventive maintenance measures are recommended to realize the timely transmission and disposal of early warning information, shorten the duration of abnormal state of DCS system, and reduce the risk of failure and loss. The overall solution integrates emerging technologies such as industrial big data, artificial intelligence, and digital twins to build a complete set of intelligent early warning closed loop for DCS system, which has strong innovation and practicality, and is of great significance to improving the operation safety and maintenance efficiency of industrial control systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of a DCS early warning method based on intelligent AI visual recognition according to an embodiment of the present invention;

[0063] Figure 2 The figure is a schematic diagram of the structure of a DCS early warning system based on intelligent AI visual recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0066] Figure 1 FIG. 1 is a flow chart of a DCS early warning method based on intelligent AI visual recognition according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0067] S101. Acquire real-time operating parameters of the DCS system, wherein the real-time operating parameters include operating status data, alarm information, operation logs, and video monitoring data of each subsystem of the DCS system, wherein the operating status data includes real-time process parameters and equipment operating parameters of each subsystem, and pre-process the real-time operating parameters to obtain real-time operating parameter features in a unified format.

[0068] In addition to real-time operating parameters, DCS systems also include various types of heterogeneous data such as alarm information, operation logs, video surveillance, etc. In order to perform global abnormal pattern recognition, these heterogeneous data must first be preprocessed and feature extracted to form a unified feature representation.

[0069] Specifically, for alarm information, text mining techniques such as TF-IDF and Word2Vec can be used to extract keywords and semantic features. Specifically, the alarm information is segmented into a corpus, the frequency and inverse document frequency of each word are counted, the TF-IDF weight matrix is ​​generated, and the matrix is ​​reduced in dimension through methods such as SVD to obtain a low-dimensional semantic representation of the alarm information. At the same time, pre-trained word vector models such as BERT and GPT can also be used to map the alarm information to a high-dimensional semantic space to capture the contextual associations between words.

[0070] For operation logs, you can use sequence pattern mining techniques, such as N-gram and sequence association rules, to extract operation behavior features. Specifically, sort the operation logs by timestamp, slide a fixed-length time window, and generate operation sequence fragments. Count the frequency and confidence of various operation sequences, mine frequent operation patterns and association rules, and convert them into real-valued vectors using one-hot encoding or embedding methods.

[0071] For video surveillance, computer vision technologies such as target detection, target tracking, and behavior recognition can be used to extract the characteristics of the device's operating status. Specifically, pre-trained target detection models such as YOLO and Faster R-CNN are used to detect the location and category of the device in the video frame, and multi-target tracking algorithms such as DeepSORT and TrackletNet are used to track the movement trajectory of the device. Based on the trajectory, motion features such as speed, acceleration, and rotation angle, as well as behavioral features such as trajectory shape and movement frequency are extracted. At the same time, information such as optical flow and skeleton can also be used to analyze the device's operating status and posture changes.

[0072] Finally, the preprocessed different modal data are converted into the form of feature vectors, and they are spliced ​​or fused to obtain the real-time operation parameter feature representation in a unified format.

[0073] In this embodiment, by acquiring and preprocessing multi-source heterogeneous data of the DCS system, a solid data foundation is laid for realizing efficient and comprehensive system status monitoring, and the real-time operation status of the DCS system can be fully monitored, including process parameters, equipment status, alarm information, operation records, video monitoring and other dimensions, and the system operation status can be perceived in a panoramic manner. It breaks through the limitation that traditional methods can only process a single data source, and greatly improves the coverage and accuracy of status monitoring. It provides high-quality data support for subsequent advanced applications such as fault diagnosis, predictive maintenance and optimization decision-making, which is conducive to improving the operational reliability and efficiency of the DCS system.

[0074] S102. Input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in the preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode.

[0075] Optionally, the intelligent AI visual recognition model can adopt a deep learning algorithm, through the combined structure of a convolutional neural network (CNN) and a long short-term memory network (LSTM), to automatically extract the multi-dimensional and multi-scale key features contained in the feature vector of real-time operation parameters, explore the spatiotemporal correlation between parameters, and realize accurate identification of abnormal patterns.

[0076] The steps of building the model include: collecting a large amount of historical operation data of the DCS system, including normal working conditions and various abnormal working conditions, preprocessing it, and obtaining a sample set of parameter feature vectors. According to the characteristics of the DCS system and the complexity of the abnormal pattern, a combined structure of CNN+LSTM is designed. CNN is responsible for extracting the local features of the parameter feature vector, and LSTM is responsible for mining the temporal correlation between features. A supervised learning method is adopted, and the model is trained using sample data. The model parameters are updated through the back propagation algorithm to continuously improve the model's feature extraction and anomaly recognition capabilities. Optimization techniques such as attention mechanism and regularization are introduced to enhance the generalization performance of the model. The model's anomaly recognition accuracy, precision and recall are evaluated using methods such as holdout or cross-validation to verify the effectiveness and reliability of the model.

[0077] During operation, the real-time operating parameter feature vector is input into the trained intelligent AI visual recognition model, and the model automatically and comprehensively analyzes the operating status of each subsystem and identifies potential abnormal modes. In order to determine the severity of the abnormal mode, it is necessary to compare the identified abnormal mode with the mode in the preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode. The lower the similarity, the farther the current operating state deviates from the normal operating condition, and the greater the abnormal risk. When the similarity is lower than the preset abnormal threshold, it is determined that there is an abnormal risk in the current operation of the DCS system, and further early warning measures need to be taken.

[0078] In an optional embodiment,

[0079] The real-time operating parameter features are input into the pre-built intelligent AI visual recognition model to analyze the operating status of the DCS system and identify abnormal patterns at the system level, including:

[0080] A multimodal feature fusion network is designed in the intelligent AI visual recognition model. The alarm information in the real-time operation parameter features is mined by text to obtain the alarm information features. The operation log is mined by sequence patterns to obtain the operation log features. The video surveillance data is subjected to target detection, target tracking and behavior recognition to obtain the video surveillance features.

[0081] The extracted alarm information features, operation log features and video surveillance features are input into the multimodal feature fusion network, and each feature is mapped to the latent space of the same dimension through the feature mapping method;

[0082] The attention mechanism is used to calculate the similarity between features in the latent space, and weights are assigned to each feature. The features in the latent space are fused using a hybrid expert network to obtain fused features. The fused features are compressed using an autoencoder to obtain the final multimodal fused features.

[0083] A heterogeneous graph is constructed according to the multimodal fusion features, wherein the nodes in the heterogeneous graph include devices, parameters, and alarms, and the edges include physical connections, logical associations, and timing dependencies. A graph neural network is applied to the heterogeneous graph, and the hidden states of the nodes are iteratively updated through a message passing mechanism. In each propagation step, the node first aggregates messages from neighboring nodes, and then updates the hidden states of the nodes in combination with its own features and neighbor information.

[0084] For the local anomaly detection task, the graph decoder is used to map the node hidden state into anomaly metrics, and the graph neural network corresponding to the local anomaly detection task is trained by optimizing the cross entropy loss function between the anomaly metric and the true anomaly label of the node; for the global anomaly association task, the graph pooling method based on anomaly metrics is used to gradually aggregate node information to generate an abnormal region hierarchy, and the graph decoder is used to map the abnormal region features into global anomaly metrics, and the graph neural network corresponding to the global anomaly association task is trained by optimizing the cross entropy loss function between the global anomaly metric and the true anomaly label at the system level;

[0085] The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results.

[0086] Exemplarily, a multimodal feature fusion network is first designed. The network consists of three modules: feature mapping, feature interaction, and feature fusion. The feature mapping module uses a deep neural network to map the original features of different modalities to the latent space of the same dimension to achieve feature alignment and unified representation. The feature interaction module uses the attention mechanism and the gating mechanism to model the interaction and dependency between different modal features. Among them, the attention mechanism assigns weights to each feature by calculating the similarity between features, highlighting key features and suppressing noise features. The gating mechanism adaptively controls the flow and fusion of different modal information by learning the gating function, and dynamically adjusts the contribution of each modality. The feature fusion module uses a hybrid expert network and an autoencoder to achieve deep fusion and high-order abstraction of different interactive features. The hybrid expert network contains multiple subnetworks, each of which corresponds to an interaction mode. The weights of each subnetwork are adaptively adjusted according to task requirements and sample distribution, and the decision results of different interaction modes are integrated. The autoencoder compresses the fused features into a lower-dimensional latent space by reconstructing the task, and optimizes the compactness and distinguishability of the feature representation.

[0087] After obtaining the multimodal fusion features, the graph neural network is further used to model the topological structure of the system to which the heterogeneous data belongs, and to mine the complex associations and interaction patterns between entities such as devices, parameters, and alarms. Specifically, a heterogeneous graph is constructed with devices, parameters, and alarms as nodes and physical connections, logical associations, and temporal dependencies as edges. The graph neural network consists of two parts: a graph encoder and a graph decoder. The graph encoder uses a message passing mechanism to iteratively update the hidden state of the node. In each propagation step, the node first aggregates messages from neighboring nodes, and then updates the hidden state based on its own features and neighbor information. Through multiple rounds of message passing, the node can capture multi-hop relationships and long-range dependencies and learn more global and higher-order feature representations. The graph decoder uses a readout function to generate anomaly metrics based on the hidden state of the node, that is, the possibility that the node belongs to an abnormal pattern. By optimizing the loss function of the anomaly metric and the true label, the graph neural network can be trained end-to-end to adaptively mine the node interactions and propagation paths that are most relevant to the abnormal pattern. At the same time, a graph regularization term is introduced during the training process. Graph regularization guides the graph neural network to learn more generalized and robust abnormal pattern representations by modeling the similarities and differences between nodes. On the one hand, similar nodes are encouraged to learn similar representations to avoid overfitting to noise patterns. On the other hand, different nodes are penalized to learn the same representation, which improves the distinguishability and diversity of features.

[0088] Based on the graph neural network, a multi-task learning framework is further designed for the abnormal pattern recognition task. The framework includes two subtasks: local anomaly detection and global anomaly association. Through joint modeling and training, the multi-scale and multi-granularity characteristics of abnormal patterns are fully utilized to improve the accuracy and generalization performance of abnormal pattern recognition. First, for the local anomaly detection task, the node hidden state of the graph neural network is used to generate anomaly metrics. Specifically, by aggregating neighbor information and updating its own state through the graph encoder, the graph neural network can learn the high-order feature representation of each node and characterize the role and importance of the node in the graph. Then, the graph decoder is used to map the node hidden state to anomaly metrics, indicating the possibility that the node belongs to an abnormal pattern. The generation of anomaly metrics can use a simple fully connected layer or a more complex attention mechanism or gating mechanism to adaptively adjust the contribution of different nodes. The goal of the anomaly metric is to be as consistent as possible with the true anomaly label of the node. Therefore, the graph neural network corresponding to the local anomaly detection task is trained by minimizing the cross entropy loss function between the anomaly metric and the true label.

[0089] Secondly, for the global anomaly association task, an anomaly propagation mechanism based on graph pooling is designed. Graph pooling generates subgraph-level representations by aggregating node features, and gradually compresses the graph into a higher level of abstraction. In each pooling layer, the nodes are sorted according to the anomaly metric, and several nodes with the highest anomaly metric are selected as anomaly centers. Their neighbor nodes are aggregated to the anomaly center to form an anomaly subgraph. The features of the anomaly center summarize the information inside the anomaly subgraph through an aggregation function and pass it to the next layer. Through multiple iterations, the anomaly information can be propagated and diffused in the graph to form hierarchical anomaly regions. The feature representation of the anomaly region generates a global anomaly metric through a graph decoder to characterize the anomaly risk of the entire system. The learning goal of the global anomaly metric is to be consistent with the true anomaly label at the system level. Therefore, the graph neural network corresponding to the global anomaly association task is trained by minimizing the cross entropy loss function between the global anomaly metric and the true label.

[0090] Finally, in order to achieve the effective fusion of local anomaly detection and global anomaly association, the paradigm of multi-task learning is adopted to jointly optimize the two subtasks. Multi-task learning realizes information interaction and knowledge transfer between different tasks by sharing some network parameters, thereby improving the generalization performance and robustness of the model. Specifically, the two subtasks share the same graph encoder and independently use task-specific graph decoders. The graph encoder is responsible for learning general node feature representations to capture the commonalities of anomaly patterns. The graph decoder is responsible for adapting to the needs of different tasks and generating task-specific anomaly metrics. During joint training, the loss functions of the two subtasks are optimized in turn, and the shared graph encoder parameters are updated during back propagation. Through alternating training and parameter sharing, multi-task learning can explicitly balance local anomaly detection and global anomaly association, and explore the performance of anomaly patterns at different scales and granularities, thereby obtaining more accurate and comprehensive anomaly pattern recognition results.

[0091] In the inference phase, for a given test sample, the trained graph encoder is first used to generate node hidden states, and then the graph decoders for local anomaly detection and global anomaly association are used to generate node-level and system-level anomaly metrics, respectively. Finally, the prediction results of the two anomaly metrics are combined to generate the final anomaly pattern recognition result through weighted averaging or voting mechanism. The weights of the two anomaly metrics can be flexibly adjusted according to task requirements and domain knowledge to balance the impact of local and global anomalies.

[0092] In an optional embodiment,

[0093] The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results, including:

[0094] A joint optimization loss function is constructed according to the loss function of the local anomaly detection subtask and the loss function of the global anomaly association subtask. The training steps of the local anomaly detection and global anomaly association subtasks are repeatedly executed until the joint optimization loss function converges to obtain the abnormal pattern recognition result at the system level. The formula of the joint optimization loss function is as follows:

[0095]

[0096] Among them, L represents the joint optimization loss function, λ1 represents the weight of the local anomaly detection task, V represents the number of nodes, and α i represents the weight factor of node i, represents the predicted anomaly metric of node i, y i represents the true anomaly label of node i, γ represents the exponential parameter of anomaly measurement error, λ2 represents the weight of the global anomaly association task, C represents the number of operation log features, and α c represents the weight factor of neighbor node c, y c represents the true abnormal label of the neighbor node c, represents the predicted abnormality measure of neighbor node c, λ3 represents the weight of the regularization term, C i represents the set of neighbor nodes of node i, and p represents the exponential parameter of the regularization term.

[0097] In this embodiment, by designing a multimodal feature fusion network, different modal data such as text alarm information, time series operation logs and video surveillance data can be mapped to the same semantic space for fusion. By using the attention mechanism and the hybrid expert network, the weights of different modal features can be adaptively assigned to fully explore the correlation between different modalities. The fused multimodal features are constructed as a heterogeneous graph, which can depict the complex topological structure and mutual relationship between devices, parameters and alarms. Based on the message propagation mechanism of the graph neural network, local and global information can be efficiently aggregated on the heterogeneous graph to learn and reason about complex abnormal patterns. For local anomaly detection tasks, the graph neural network can accurately detect the abnormal situation of each parameter or device based on the abnormal measurement corresponding to the node hidden state output. At the same time, based on the graph pooling operation of the abnormal measurement, the graph neural network can also learn the global abnormal pattern and realize the abnormal correlation analysis at the system level. The two subtasks of local anomaly and global anomaly share the same graph encoder, and the joint optimization of the two tasks is achieved through alternating training and parameter sharing. This end-to-end optimization method avoids the error accumulation of traditional step-by-step processing and improves the overall performance of the model.

[0098] S103. If the similarity is lower than the preset abnormal threshold, it is determined that there is an abnormal risk in the current operation of the DCS system, and the digital twin simulation technology based on physical mechanisms is used to simulate the real-time operation scenario of the DCS system, predict the failure mode, failure cause and probability of the DCS system, and automatically generate warning information of the corresponding level based on the prediction results combined with the preset multi-level warning decision matrix, and push the warning information to the intelligent operation and maintenance platform of the DCS system in real time to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

[0099] Among them, according to the results of abnormal risk prediction, the system automatically matches the preset multi-level early warning decision matrix to generate early warning information of the corresponding level. Optionally, the early warning decision matrix can be pre-set based on historical experience and expert knowledge to associate the abnormal risk level with the early warning level, early warning method, and response measures. The generated early warning information is pushed to the operators of the DCS system in real time through the intelligent operation and maintenance platform, including the early warning level, abnormal cause, possible consequences, response measures, etc., to assist the operators to take targeted preventive maintenance measures in the first time to curb the further deterioration of abnormalities and avoid accidents.

[0100] Among them, the intelligent operation and maintenance platform adopts B / S architecture, supports Web and mobile access, and has a friendly visual interface, which is convenient for operators to monitor the system operation status anytime and anywhere and receive abnormal warning information. The main functional modules of the intelligent operation and maintenance platform include: Real-time data display: In the form of curve charts, bar charts, dashboards, etc., the operating parameters and KPI indicators of each subsystem of the DCS system are displayed in real time to intuitively reflect the overall operation status of the system. Abnormal warning push: abnormal warning information pops up in a prominent position, including warning level, abnormal cause, possible consequences, countermeasures, etc., and through sound and light alarms, SMS, emails and other methods, to ensure the timely delivery of warning information. Abnormal analysis and diagnosis: Provide visual analysis tools for abnormal patterns, such as fault tree diagrams, cause and effect diagrams, etc., to help operators quickly locate the root cause of abnormalities and formulate targeted response plans. Plan knowledge base: Establish an abnormal warning plan knowledge base, summarize the standard handling procedures and emergency plans for various abnormal situations, and provide decision-making references for operators. Mobile terminal APP: Develop an intelligent operation and maintenance mobile terminal APP to realize the real-time push of abnormal warning information on the mobile terminal, so that operators can grasp the system dynamics at any time and respond to abnormal situations in a timely manner.

[0101] In an optional embodiment,

[0102] The digital twin simulation technology based on physical mechanism is used to simulate the real-time operation scenario of the DCS system, and predict the failure mode, cause and probability of the DCS system, including:

[0103] Construct a multidisciplinary high-fidelity digital twin model of the DCS system, which includes a production process model, an equipment model, and a control logic model;

[0104] Extract key feature parameters of abnormal patterns, use feature dimension reduction methods to reduce the dimensions of key feature parameters, and use semantic mapping technology to map the reduced key feature parameters to the corresponding positions and object attributes of the digital twin model, triggering the digital twin model to perform simulation predictions under abnormal working conditions;

[0105] The simulation prediction starts from the abnormal point, adopts a multi-scenario parallel simulation strategy, sets different initial conditions, boundary conditions and random disturbances, generates multiple parallel simulation instances, and adopts model simplification and multi-scale modeling methods to selectively simplify the digital twin model at different time and space scales, and performs grid encryption and calculation intensification for key areas and key processes. At the same time, uncertainty quantification technology is introduced to adaptively adjust the time and space resolution and calculation accuracy of the simulation;

[0106] Based on the simulation of multi-scenario abnormal working conditions, the abnormal mode is taken as the top event, and the deductive reasoning and causal reverse tracing methods are used to automatically construct the fault tree, identify the key links and influencing factors of abnormal propagation, and on the basis of the constructed fault tree, the Monte Carlo simulation method is used to randomly sample the probability of occurrence of basic events in the fault tree and evaluate the abnormal mode;

[0107] According to the results of fault tree analysis and Monte Carlo simulation, a case-based reasoning fault diagnosis method is used to measure the similarity between abnormal patterns and historical fault cases, match fault patterns from the historical fault library, and finally output the predicted fault mode, fault cause and occurrence probability of the DCS system.

[0108] For example, a multidisciplinary high-fidelity digital twin model of a discrete control system was first constructed, including a production process model, an equipment model, and a control logic model. The production process model uses an object-oriented modeling method to abstract the production process into an object network consisting of equipment, pipelines, and valves, and defines the connection relationship and interaction interface between components. The equipment model uses a numerical simulation method, combined with the geometric parameters, material properties, and operating conditions of the equipment, to simulate the physical and chemical processes such as flow, heat transfer, mass transfer, and chemical reactions inside the equipment, while considering the mechanical characteristics and failure modes of the equipment, thereby improving the integrity and accuracy of the equipment model. The control logic model converts control strategies and logical relationships into executable code models, dynamically couples with the production process model and the equipment model, and realizes closed-loop control of the system. The operating parameters of the discrete control system are collected in real time through IoT devices and sensor networks, and mapped to the digital twin model, realizing the synchronous update of the physical world and the virtual world.

[0109] On the basis of building a digital twin model, an intelligent visual recognition model is used to detect abnormal working conditions of discrete control systems, extract key characteristic parameters of abnormal modes, and use feature dimension reduction methods such as principal component analysis and independent component analysis to reduce the dimension of key characteristic parameters. Principal component analysis transforms the original features into linearly independent new features, called principal components, through orthogonal transformation, and sorts them according to the size of variance, and selects the first few principal components with the largest variance as the features after dimension reduction. Independent component analysis assumes that the observed data is a linear combination of several statistically independent signals, and finds these independent signals by optimizing the objective function, thereby achieving feature dimension reduction. After feature dimension reduction, this paper uses semantic mapping technology to map the key characteristic parameters after dimension reduction to the corresponding positions and object attributes of the digital twin model, and reproduce abnormal working conditions in a fidelity virtual environment. Semantic mapping is to use knowledge representation methods such as ontology and semantic network to establish semantic associations between abnormal features and digital twin models, and realize seamless conversion of abnormal information between physical space and virtual space.

[0110] After reproducing the abnormal working conditions, the digital twin model is triggered to perform simulation prediction under abnormal working conditions. Among them, a multi-scenario parallel simulation strategy is adopted, and different initial conditions, boundary conditions and random disturbances are set to generate multiple parallel simulation instances. Multi-scenario parallel simulation can fully explore the evolution path of abnormal working conditions and improve the coverage and reliability of simulation results. In the process of parallel simulation, methods such as model simplification and multi-scale modeling are used to selectively simplify the digital twin model at different time and space scales. Model simplification is to remove minor factors that have little effect on the simulation and reduce the complexity of the model under the premise of ensuring key characteristics. Multi-scale modeling is to select appropriate time steps and spatial grids according to the characteristics of the simulation object to describe the system behavior at different scales. Grid encryption and calculation intensification are carried out for the key areas and key processes of abnormal propagation and diffusion, which improves the local simulation accuracy. At the same time, in order to quantify the uncertainty in the simulation process, uncertainty quantification technology is introduced to evaluate the uncertainty caused by model simplification and numerical discretization, and adaptively adjust the time and space resolution and calculation accuracy of the simulation to balance the simulation efficiency and simulation reliability.

[0111] Based on the simulation of abnormal working conditions in multiple scenarios, the abnormal mode is taken as the top event, and the deductive reasoning and causal reverse tracing methods are used to automatically construct the fault tree that may be caused by the abnormality. Deductive reasoning is to deduce the conclusion of individual cases based on general knowledge according to logical rules, while causal reverse tracing is to start from the abnormal results and reversely search for possible influencing factors and fault chains. Through the combination of these two methods, the key links and influencing factors of abnormal propagation are extracted from the digital twin model to form a complete fault tree. After constructing the fault tree, the Monte Carlo simulation method is used to randomly sample the probability of occurrence of basic events in the fault tree. Monte Carlo simulation simulates probability distribution through random numbers and estimates the characteristics of probability distribution or the numerical solution of function through a large number of random experiments. Through Monte Carlo simulation, the possibility of abnormal mode leading to system failure can be evaluated. For low-probability events in complex systems, improved Monte Carlo methods such as importance sampling and subset simulation are used to improve the efficiency and accuracy of probability estimation.

[0112] Finally, according to the results of fault tree analysis and Monte Carlo simulation, this paper adopts a fault diagnosis method based on case reasoning. By measuring the similarity between abnormal patterns and historical fault cases, the specific fault patterns that may be caused by abnormalities are matched from the historical fault library. Case reasoning is a common analogy reasoning method. It extracts the characteristics of cases, calculates the similarity between cases, finds the case most similar to the target problem, and uses the solution of the case to solve the target problem. When matching fault patterns, not only the characterization characteristics of the abnormal pattern are considered, but also the contextual information of the fault, such as equipment type, operating conditions, environmental conditions, etc., and finally outputs the predicted fault mode, fault cause and occurrence probability of the DCS system, which improves the accuracy of fault matching.

[0113] In this embodiment, through the abnormal risk prediction and early warning method based on digital twins, multidisciplinary simulation, intelligent feature extraction, fault tree analysis, Monte Carlo simulation, case reasoning and other technologies are comprehensively used to establish a complete abnormal diagnosis and risk assessment process. By building a high-fidelity digital twin model, real-time monitoring and prediction and early warning of abnormal conditions are achieved, providing strong support for the safe operation of the DCS system.

[0114] In an optional embodiment,

[0115] Adopting a multi-scenario parallel simulation strategy, setting different initial conditions, boundary conditions and random disturbances, multiple parallel simulation instances are generated, including:

[0116] Obtain the historical abnormal data of the DCS system for preprocessing, extract the multi-scale spatiotemporal features of the historical abnormal data, including abnormal features at different time scales and spatial scales, and construct a multi-scale spatiotemporal feature representation of the abnormal scene;

[0117] Inputting the multi-scale spatiotemporal feature representation of the abnormal scene into a multi-scale conditional generative adversarial network, wherein the multi-scale conditional generative adversarial network includes a generator and a discriminator, wherein the generator receives random noise and conditional variables as inputs, and the discriminator receives real abnormal scene features and abnormal scene features generated by the generator as inputs;

[0118] In the generator, multi-scale feature extraction is performed on random noise and conditional variables to obtain noise features and conditional features of different scales. The attention mechanism is used to adaptively fuse the noise features and conditional features of different scales to generate abnormal scene features that meet the conditional constraints. In the discriminator, multi-scale feature extraction is performed on the input real abnormal scene features and the generated abnormal scene features to obtain real features and generated features of different scales. The attention mechanism is used to adaptively fuse the real features and generated features of different scales.

[0119] Constructing the objective function of a multi-scale conditional generative adversarial network, the objective function includes the logarithmic probability of the real abnormal scene features of the discriminator and the false probability of the abnormal scene features generated by the generator, and optimizing the training by maximizing the objective function of the discriminator and minimizing the objective function of the generator, and finally obtaining a trained abnormal scene distribution model;

[0120] Setting an abnormal type and severity condition of an abnormal scenario, encoding the abnormal type and severity condition into a conditional vector, connecting the conditional vector with a random noise vector to form a conditional noise vector, inputting the conditional noise vector into a generator of an abnormal scenario distribution model, and generating an abnormal scenario that meets the abnormal type and severity conditions through the generator;

[0121] Taking the diversity, coverage and abnormality degree of the abnormal scenarios as the optimization objectives, a fitness function is constructed, an adaptive sampling strategy is used to optimize the sampling of abnormal scenarios, and the sampling parameters are adaptively adjusted to obtain an optimized set of abnormal scenarios. The optimized set of abnormal scenarios is input into the digital twin model for multi-scenario parallel simulation.

[0122] Exemplarily, first, the historical abnormal data of the DCS system is preprocessed by cleaning, denoising and normalization to improve data quality and consistency. Then, multi-scale spatiotemporal features are extracted from the preprocessed historical abnormal data. The time scale feature reflects the evolution of the abnormal scene in the time dimension, including the duration of the abnormality, the frequency of abnormal occurrence, the abnormal change trend, etc. The spatial scale feature reflects the distribution of the abnormal scene in the spatial dimension, including the abnormal impact range, the abnormal propagation path, the abnormal correlation relationship, etc. By extracting abnormal features at different time scales and spatial scales, a multi-scale spatiotemporal feature representation of the abnormal scene is constructed, which provides an information basis for the subsequent generation and optimization of abnormal scenes.

[0123] The generation of abnormal scenarios needs to fully consider the constraints of abnormal type, severity, etc., while ensuring the authenticity and diversity of the generated scenarios. To this end, the Multi-Scale Conditional Generative Adversarial Network (MS-CGAN) is introduced to model the distribution of abnormal scenarios. MS-CGAN consists of two parts: the generator and the discriminator. The generator is responsible for generating realistic abnormal scenarios according to the conditions, and the discriminator is responsible for judging the difference between the generated abnormal scenarios and the real abnormal scenarios.

[0124] In the generator of MS-CGAN, random noise and conditional variables are used as inputs, and multi-scale feature extraction is performed to obtain noise features and conditional features of different scales. The noise feature introduces randomness to increase the diversity of the generated scene; the conditional feature introduces constraint information to control the properties of the generated scene. Then, the attention mechanism is used to adaptively fuse the noise features and conditional features of different scales. The attention mechanism can dynamically adjust the weight of the noise feature according to the conditional information, so that the generated abnormal scene is more in line with the conditional constraints. The fused features are gradually restored to the original abnormal scene scale through upsampling operations such as deconvolution, and finally generate abnormal scene features that meet the conditional constraints.

[0125] In the discriminator of MS-CGAN, the real abnormal scene features and the abnormal scene features generated by the generator are used as inputs, and are respectively subjected to multi-scale feature extraction to obtain real features and generated features of different scales. Then, the attention mechanism is used to adaptively fuse the real features and generated features of different scales. The attention mechanism can dynamically adjust the weights of features of different scales according to the differences between the real features and the generated features, thereby improving the discriminative ability of the discriminator. The fused features pass through the fully connected layer to output the probability values ​​of the real abnormal scenes and the generated abnormal scenes.

[0126] The training process of MS-CGAN is a minimum-maximum game optimization problem. The goal of the discriminator is to maximize the difference between the log probability of the real abnormal scene features and the false probability of the generated abnormal scene features, that is, the ability to distinguish real abnormal scenes and generate abnormal scenes. The goal of the generator is to minimize the false probability of the generated abnormal scene features, that is, to deceive the discriminator into misjudging the generated abnormal scenes as real abnormal scenes. By alternately training the generator and the discriminator and continuously updating their parameters, we finally get a generator that can generate realistic abnormal scenes, that is, the abnormal scene distribution model.

[0127] After the abnormal scene distribution model is established, the conditions of the abnormal scene, such as abnormal type, severity, etc., can be set as needed. The abnormal type and severity conditions are encoded into a conditional vector of fixed length and connected with the randomly sampled noise vector to form a conditional noise vector. The conditional noise vector is input into the generator of the abnormal scene distribution model, and after multi-scale feature extraction, attention fusion and upsampling, the abnormal scene features that meet the conditional constraints are generated. The abnormal type condition controls the abnormal category of the generated scene, such as equipment failure, parameter overrun, environmental abnormality, etc.; the severity condition controls the abnormal degree of the generated scene, such as slight abnormality, moderate abnormality, severe abnormality, etc. By setting different condition combinations, a variety of abnormal scenes can be generated, covering different abnormal types and severities.

[0128] Although the abnormal scenarios generated by the abnormal scenario distribution model meet the conditional constraints, there may be problems such as duplication, redundancy or insufficient abnormality, and the generated abnormal scenarios need to be sampled optimally. The fitness function is constructed with the diversity, coverage and abnormality of the abnormal scenarios as optimization goals. The diversity goal measures the differences between abnormal scenarios to avoid generating repeated or similar scenarios; the coverage goal measures the coverage of abnormal scenarios for different abnormal types and severity levels to ensure that the generated scenarios are representative; the abnormality level goal measures the abnormality level of the abnormal scenarios, giving priority to scenarios with higher abnormality levels.

[0129] In the abnormal scene sampling process, an adaptive sampling strategy is adopted. In the initial stage, sampling is performed uniformly in the abnormal scene space to ensure the randomness and extensiveness of sampling. As sampling proceeds, the quality of the sampling scene is evaluated according to the fitness function, and the sampling distribution is adaptively adjusted. For abnormal scene areas with high fitness, the sampling probability is increased, and for abnormal scene areas with low fitness, the sampling probability is reduced. At the same time, the sampling parameters such as sampling step size and acceptance probability are dynamically adjusted to balance exploration and utilization. Through the adaptive sampling strategy, the abnormal scene collection is continuously optimized to improve the quality and efficiency of abnormal scenes.

[0130] In order to evaluate the impact and consequences of abnormal scenarios, the optimized set of abnormal scenarios needs to be input into the digital twin model for simulation. The digital twin model is a virtual mapping of the physical system that can simulate the behavior and response of the system. By injecting abnormal scenarios into the digital twin model, the operating status and fault evolution process of the system under abnormal conditions can be observed. Considering that the set of abnormal scenarios is large and the efficiency of serial simulation is low, multi-scenario parallel simulation is adopted. The set of abnormal scenarios is divided into multiple subsets, each subset is assigned to a different computing node, and simulated simultaneously in multiple copies of the digital twin model. Parallel simulation can make full use of computing resources and accelerate the evaluation and analysis of abnormal scenarios. After the simulation results are summarized, the impact range, severity and evolution trend of different abnormal scenarios can be comprehensively evaluated to provide support for abnormal warning and decision-making.

[0131] In an optional embodiment,

[0132] The calculation formula for the objective function of constructing a multi-scale conditional generative adversarial network is as follows:

[0133]

[0134] in, represents the objective function of the multi-scale conditional generative adversarial network, R(D, G) represents the value function between the generator G and the discriminator D, and E f [·] represents the expectation of the characteristics of the real abnormal scene, E z [·] represents the expectation of the generated synthetic abnormal scene features, J represents the temporal scale level, K represents the spatial scale level, and a (D) represents the attention weight in the discriminator, f (j，k) Multi-scale representation of the real abnormal scene characteristics, s (j，k) represents the multi-scale representation of the conditional variable, a (G) represents the attention weight in the generator, z (j，k) Multi-scale representation of random noise.

[0135] In this embodiment, the historical abnormal data of the DCS system is fully utilized to extract multi-scale spatiotemporal features and construct the information basis of abnormal scenarios. A multi-scale conditional generative adversarial network is introduced to establish an abnormal scenario distribution model to generate realistic, diverse, and conditionally constrained abnormal scenarios. An adaptive sampling strategy is used to optimize the generated abnormal scenarios to improve the quality and efficiency of the abnormal scenarios. Through multi-scenario parallel simulation, the evaluation and analysis of abnormal scenarios are accelerated to provide support for abnormal warning and decision-making. The overall method realizes the intelligent generation and optimization of abnormal scenarios, improves the ability of abnormal warning and abnormal location, reduces the cost of manually designed abnormal scenarios, and has strong practicality and operability.

[0136] In an optional embodiment,

[0137] According to the prediction results combined with the preset multi-level warning decision matrix, the corresponding level of warning information is automatically generated, including:

[0138] Construct an abnormal risk warning domain ontology, define the relationship mapping between abnormal types, influencing factors, warning levels and response measures, and form an abnormal risk warning ontology model;

[0139] Integrate structured data, semi-structured data and unstructured data, realize the association of multi-source heterogeneous data through data integration, and use the knowledge extraction model of deep learning to extract entities, attributes and relationships from unstructured data to form an abnormal risk warning knowledge graph;

[0140] Take the abnormal type and influencing factors output by abnormal risk prediction as known conditions, match the corresponding entity nodes in the consistent abnormal risk warning knowledge graph, use the abnormal risk warning ontology reasoning rules and graph reasoning algorithms to infer the concepts, attributes and relationships associated with the known conditions, and generate abnormal risk warning reasoning explanations;

[0141] Using the knowledge subgraph obtained by the abnormal risk warning reasoning interpretation, an abnormal risk warning question and answer template is constructed to form a structured problem description. The problem description is input into the natural language generation model based on Seq2Seq to automatically generate the recommended text of abnormal risk response measures.

[0142] Based on the confidence, coverage and novelty indicators of abnormal risk response measures knowledge, a knowledge sorting algorithm is used to prioritize the suggested texts of abnormal risk response measures. The abnormal risk level is calculated according to the abnormal prediction results, and the abnormal risk response measures of the corresponding level are selected to automatically generate warning information of the corresponding abnormal risk level.

[0143] For example, first, through in-depth analysis of the field of abnormal risk warning, core concepts such as abnormal type, influencing factors, warning level, countermeasures, etc. are identified. Then, hierarchical and non-hierarchical relationships between concepts are defined, such as the causal relationship between abnormal type and influencing factors, the corresponding relationship between warning level and countermeasures, etc. Next, attributes are added to the concepts, such as the severity attribute of the abnormal type, the weight attribute of the influencing factor, etc. Finally, based on domain knowledge and expert experience, a series of ontology reasoning rules and constraints are constructed, such as a certain combination of abnormal type and influencing factor will lead to a specific warning level, and corresponding countermeasures need to be taken for a specific warning level. Through the above steps, an abnormal risk warning ontology model is formed, which provides a semantic basis for the subsequent knowledge graph construction and reasoning.

[0144] Abnormal risk warning knowledge comes from various types of data, including structured data, semi-structured data and unstructured data. Structured data includes abnormal record tables in relational databases, semi-structured data includes abnormal reports in XML format, and unstructured data includes fault maintenance logs, expert experience documents, etc. In order to build a comprehensive and accurate abnormal risk warning knowledge graph, it is necessary to fuse and extract knowledge from multi-source heterogeneous data. First, through data integration technologies such as ETL and data virtualization, data from different sources and formats are cleaned, converted and associated to form a unified data view. Then, deep learning knowledge extraction models such as BERT and CNN-LSTM are used to identify and extract entities, attributes and relationships related to abnormal risk warning from unstructured data. Deep learning models can automatically learn the semantic features of text, overcoming the limitations of traditional rules and template methods. The extracted knowledge elements are fused with structured data, mapped to corresponding concepts, attributes and relationships based on the abnormal risk warning ontology model, and finally form an abnormal risk warning knowledge graph rich in semantic information.

[0145] When the abnormal risk prediction module outputs the abnormal type and influencing factors, it is necessary to infer the corresponding warning level and countermeasures based on these known conditions. The abnormal risk warning knowledge graph provides complete domain knowledge, including the association information between abnormal types, influencing factors, warning levels and countermeasures, which can be used to support this reasoning process. First, the abnormal type and influencing factors are used as query conditions to match consistent entity nodes in the abnormal risk warning knowledge graph. The matching process considers semantic similarity, including not only literal matching, but also synonym matching and hyponymy matching. Then, taking the matched entity nodes as the starting point, the abnormal risk warning ontology reasoning rules and graph reasoning algorithms are used to infer the concepts, attributes and relationships associated with the known conditions. The ontology reasoning rules are based on ontology axioms and domain constraints, and use formal methods such as description logic to deduce and reason about knowledge. Graph reasoning algorithms such as the shortest path algorithm and the random walk algorithm discover implicit associations by analyzing the topological structure of the knowledge graph and the semantics of the edges. Finally, the associated concepts, attributes and relationships obtained by reasoning are organized into knowledge subgraphs to form an abnormal risk warning reasoning explanation, providing knowledge support for the subsequent generation of countermeasures.

[0146] Based on the knowledge subgraph obtained by abnormal risk warning reasoning, an abnormal risk warning question and answer template is constructed. The question and answer template defines the structured representation of the problem, including elements such as abnormal type, influencing factors, warning level, and the logical relationship between these elements. The entities, attributes, and relationships in the knowledge subgraph are mapped to the question and answer template to form a structured problem description. Then, the structured problem description is converted into natural language text using a natural language generation model based on Seq2Seq. The Seq2Seq model consists of an encoder and a decoder. The encoder encodes the problem description into a semantic vector, and the decoder generates natural language countermeasures based on the semantic vector. The model is trained on the abnormal risk countermeasure corpus to learn the mapping relationship between the problem description and the countermeasures. In this way, the recommended text of the abnormal risk countermeasures is automatically generated, providing highly readable and complete abnormal handling guidance.

[0147] The generated text is evaluated based on the confidence, coverage and novelty indicators of the abnormal risk countermeasures knowledge. Confidence reflects the credibility of the countermeasures knowledge, coverage reflects the completeness of the countermeasures knowledge, and novelty reflects the innovation of the countermeasures knowledge. Using knowledge sorting algorithms such as PageRank and HITS, the countermeasures suggestion texts are prioritized according to these indicators, and the texts with high confidence, wide coverage and strong novelty are ranked first. After sorting, the countermeasures suggestion texts that match the abnormal risk prediction results are selected as the candidate set. Finally, according to the abnormal degree in the abnormal risk prediction results, the abnormal risk level is calculated, and the countermeasures suggestion texts corresponding to the abnormal risk level in the candidate set are selected to automatically generate the corresponding warning information. The warning information includes abnormal risk level, abnormal description, impact range, countermeasures and other contents, which are presented in a concise and clear manner to provide decision support for relevant personnel.

[0148] In this embodiment, an abnormal risk warning domain ontology is constructed to formally represent the semantic relationship between abnormal types, influencing factors, warning levels and countermeasures, providing a unified conceptual framework for knowledge graph construction. Multi-source heterogeneous data is integrated, and the knowledge extraction model of deep learning is used to extract abnormal risk warning knowledge from structured, semi-structured and unstructured data to build a comprehensive and accurate abnormal risk warning knowledge graph. Reasoning is performed based on the abnormal risk warning knowledge graph, and the ontology reasoning rules and graph reasoning algorithms are used to infer the warning level and countermeasures from the known abnormal types and influencing factors, and generate abnormal risk warning reasoning explanations. The Seq2Seq natural language generation model is used to convert the structured problem description into a suggested text of abnormal risk countermeasures, providing highly readable and complete abnormal handling guidance. Evaluate and optimize abnormal risk countermeasures, sort based on confidence, coverage and novelty indicators, select measures that match the abnormal risk level, and generate concise and clear warning information. The automated process from abnormal risk prediction to countermeasure generation is realized, the timeliness and accuracy of abnormal risk warning are improved, and intelligent support is provided for abnormal risk management, which has strong practical value.

[0149] Figure 2 FIG. 1 is a schematic diagram of the structure of a DCS early warning system based on intelligent AI visual recognition according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0150] The first unit is used to obtain the real-time operation parameters of the DCS system, wherein the real-time operation parameters include the operation status data, alarm information, operation log and video monitoring data of each subsystem of the DCS system, wherein the operation status data includes the real-time process parameters and equipment operation parameters of each subsystem, and pre-process the real-time operation parameters to obtain the real-time operation parameter features in a unified format;

[0151] The second unit is used to input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in a preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode;

[0152] The third unit is used to determine that there is an abnormal risk in the current operation of the DCS system if the similarity is lower than the preset abnormal threshold, and use digital twin simulation technology based on physical mechanisms to simulate the real-time operation scenario of the DCS system, predict the failure mode, cause and probability of the DCS system, and automatically generate warning information of the corresponding level based on the prediction results combined with the preset multi-level warning decision matrix, and push the warning information to the intelligent operation and maintenance platform of the DCS system in real time, to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

[0153] According to a third aspect of the embodiments of the present invention,

[0154] An electronic device is provided, comprising:

[0155] processor;

[0156] a memory for storing processor-executable instructions;

[0157] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0158] According to a fourth aspect of the embodiments of the present invention,

[0159] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0160] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A DCS early warning method based on intelligent AI visual recognition, characterized in that: include: Acquire the real-time operation parameters of the DCS system, wherein the real-time operation parameters include the operation status data, alarm information, operation log and video monitoring data of each subsystem of the DCS system, wherein the operation status data includes the real-time process parameters and equipment operation parameters of each subsystem, and pre-process the real-time operation parameters to obtain the real-time operation parameter features in a unified format; Input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in a preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode; If the similarity is lower than the preset abnormal threshold, it is determined that there is an abnormal risk in the current operation of the DCS system, and the digital twin simulation technology based on physical mechanisms is used to simulate the real-time operation scenario of the DCS system to predict the failure mode, failure cause and probability of occurrence of the DCS system. Based on the prediction results and the preset multi-level early warning decision matrix, the corresponding level of early warning information is automatically generated and pushed to the intelligent operation and maintenance platform of the DCS system in real time to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

2. The method according to claim 1, characterized in that The real-time operating parameter features are input into the pre-built intelligent AI visual recognition model to analyze the operating status of the DCS system and identify abnormal patterns at the system level, including: A multimodal feature fusion network is designed in the intelligent AI visual recognition model. The alarm information in the real-time operation parameter features is mined by text to obtain the alarm information features. The operation log is mined by sequence patterns to obtain the operation log features. The video surveillance data is subjected to target detection, target tracking and behavior recognition to obtain the video surveillance features. The extracted alarm information features, operation log features and video surveillance features are input into the multimodal feature fusion network, and each feature is mapped to the latent space of the same dimension through the feature mapping method; The attention mechanism is used to calculate the similarity between features in the latent space, and weights are assigned to each feature. The features in the latent space are fused using a hybrid expert network to obtain fused features. The fused features are compressed using an autoencoder to obtain the final multimodal fused features. A heterogeneous graph is constructed according to the multimodal fusion features, wherein the nodes in the heterogeneous graph include devices, parameters, and alarms, and the edges include physical connections, logical associations, and timing dependencies. A graph neural network is applied to the heterogeneous graph, and the hidden states of the nodes are iteratively updated through a message passing mechanism. In each propagation step, the node first aggregates messages from neighboring nodes, and then updates the hidden states of the nodes in combination with its own features and neighbor information. For the local anomaly detection task, the graph decoder is used to map the node hidden state into anomaly metrics, and the graph neural network corresponding to the local anomaly detection task is trained by optimizing the cross entropy loss function between the anomaly metric and the true anomaly label of the node; for the global anomaly association task, the graph pooling method based on anomaly metrics is used to gradually aggregate node information to generate an abnormal region hierarchy, and the graph decoder is used to map the abnormal region features into global anomaly metrics, and the graph neural network corresponding to the global anomaly association task is trained by optimizing the cross entropy loss function between the global anomaly metric and the true anomaly label at the system level; The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results.

3. The method according to claim 2, characterized in that The two subtasks of local anomaly detection and global anomaly association share the same graph encoder, and achieve joint optimization through alternating training and parameter sharing. The outputs of the two subtasks are combined to finally output the system-level anomaly pattern recognition results, including: A joint optimization loss function is constructed according to the loss function of the local anomaly detection subtask and the loss function of the global anomaly association subtask. The training steps of the local anomaly detection and global anomaly association subtasks are repeatedly executed until the joint optimization loss function converges to obtain the abnormal pattern recognition result at the system level. The formula of the joint optimization loss function is as follows: Among them, L represents the joint optimization loss function, λ1 represents the weight of the local anomaly detection task, V represents the number of nodes, and α i represents the weight factor of node i, represents the predicted anomaly metric of node i, y i represents the true anomaly label of node i, γ represents the exponential parameter of anomaly measurement error, λ2 represents the weight of the global anomaly association task, C represents the number of operation log features, and α c represents the weight factor of neighbor node c, y c represents the true abnormal label of the neighbor node c, represents the predicted abnormality measure of neighbor node c, λ3 represents the weight of the regularization term, C i represents the set of neighbor nodes of node i, and p represents the exponential parameter of the regularization term.

4. The method according to claim 1, characterized in that: The digital twin simulation technology based on physical mechanism is used to simulate the real-time operation scenario of the DCS system, and predict the failure mode, cause and probability of the DCS system, including: Construct a multidisciplinary high-fidelity digital twin model of the DCS system, which includes a production process model, an equipment model, and a control logic model; Extract key feature parameters of abnormal patterns, use feature dimension reduction methods to reduce the dimensions of key feature parameters, and use semantic mapping technology to map the reduced key feature parameters to the corresponding positions and object attributes of the digital twin model, triggering the digital twin model to perform simulation predictions under abnormal working conditions; The simulation prediction starts from the abnormal point, adopts a multi-scenario parallel simulation strategy, sets different initial conditions, boundary conditions and random disturbances, generates multiple parallel simulation instances, and adopts model simplification and multi-scale modeling methods to selectively simplify the digital twin model at different time and space scales, and performs grid encryption and calculation intensification for key areas and key processes. At the same time, uncertainty quantification technology is introduced to adaptively adjust the time and space resolution and calculation accuracy of the simulation; Based on the simulation of multi-scenario abnormal working conditions, the abnormal mode is taken as the top event, and the deductive reasoning and causal reverse tracing methods are used to automatically construct the fault tree, identify the key links and influencing factors of abnormal propagation, and on the basis of the constructed fault tree, the Monte Carlo simulation method is used to randomly sample the probability of occurrence of basic events in the fault tree and evaluate the abnormal mode; According to the results of fault tree analysis and Monte Carlo simulation, a case-based reasoning fault diagnosis method is used to measure the similarity between abnormal patterns and historical fault cases, match fault patterns from the historical fault library, and finally output the predicted fault mode, fault cause and occurrence probability of the DCS system.

5. The method according to claim 4, characterized in that Adopting a multi-scenario parallel simulation strategy, setting different initial conditions, boundary conditions and random disturbances, multiple parallel simulation instances are generated, including: Obtain the historical abnormal data of the DCS system for preprocessing, extract the multi-scale spatiotemporal features of the historical abnormal data, including abnormal features at different time scales and spatial scales, and construct a multi-scale spatiotemporal feature representation of the abnormal scene; Inputting the multi-scale spatiotemporal feature representation of the abnormal scene into a multi-scale conditional generative adversarial network, wherein the multi-scale conditional generative adversarial network includes a generator and a discriminator, wherein the generator receives random noise and conditional variables as inputs, and the discriminator receives real abnormal scene features and abnormal scene features generated by the generator as inputs; In the generator, multi-scale feature extraction is performed on random noise and conditional variables to obtain noise features and conditional features of different scales. The attention mechanism is used to adaptively fuse the noise features and conditional features of different scales to generate abnormal scene features that meet the conditional constraints. In the discriminator, multi-scale feature extraction is performed on the input real abnormal scene features and the generated abnormal scene features to obtain real features and generated features of different scales. The attention mechanism is used to adaptively fuse the real features and generated features of different scales. Constructing the objective function of a multi-scale conditional generative adversarial network, the objective function includes the logarithmic probability of the real abnormal scene features of the discriminator and the false probability of the abnormal scene features generated by the generator, and optimizing the training by maximizing the objective function of the discriminator and minimizing the objective function of the generator, and finally obtaining a trained abnormal scene distribution model; Setting an abnormal type and severity condition of an abnormal scenario, encoding the abnormal type and severity condition into a conditional vector, connecting the conditional vector with a random noise vector to form a conditional noise vector, inputting the conditional noise vector into a generator of an abnormal scenario distribution model, and generating an abnormal scenario that meets the abnormal type and severity conditions through the generator; Taking the diversity, coverage and abnormality degree of the abnormal scenarios as the optimization objectives, a fitness function is constructed, an adaptive sampling strategy is used to optimize the sampling of abnormal scenarios, and the sampling parameters are adaptively adjusted to obtain an optimized set of abnormal scenarios. The optimized set of abnormal scenarios is input into the digital twin model for multi-scenario parallel simulation.

6. The method according to claim 5, characterized in that The calculation formula for the objective function of constructing a multi-scale conditional generative adversarial network is as follows: in, represents the objective function of the multi-scale conditional generative adversarial network, R(D, G) represents the value function between the generator G and the discriminator D, and E f [·] represents the expectation of the characteristics of the real abnormal scene, E z [·] represents the expectation of the generated synthetic abnormal scene features, J represents the temporal scale level, K represents the spatial scale level, and a (D) represents the attention weight in the discriminator, f (j，k) Multi-scale representation of the real abnormal scene characteristics, s (j，k) represents the multi-scale representation of the conditional variable, a (G) represents the attention weight in the generator, z (j，k) Multi-scale representation of random noise.

7. The method according to claim 1, characterized in that According to the prediction results combined with the preset multi-level warning decision matrix, the corresponding level of warning information is automatically generated, including: Construct an abnormal risk warning domain ontology, define the relationship mapping between abnormal types, influencing factors, warning levels and response measures, and form an abnormal risk warning ontology model; Integrate structured data, semi-structured data and unstructured data, realize the association of multi-source heterogeneous data through data integration, and use the knowledge extraction model of deep learning to extract entities, attributes and relationships from unstructured data to form an abnormal risk warning knowledge graph; Take the abnormal type and influencing factors output by abnormal risk prediction as known conditions, match the corresponding entity nodes in the consistent abnormal risk warning knowledge graph, use the abnormal risk warning ontology reasoning rules and graph reasoning algorithms to infer the concepts, attributes and relationships associated with the known conditions, and generate abnormal risk warning reasoning explanations; Using the knowledge subgraph obtained by the abnormal risk warning reasoning interpretation, an abnormal risk warning question and answer template is constructed to form a structured problem description. The problem description is input into the natural language generation model based on Seq2Seq to automatically generate the recommended text of abnormal risk response measures. Based on the confidence, coverage and novelty indicators of abnormal risk response measures knowledge, a knowledge sorting algorithm is used to prioritize the suggested texts of abnormal risk response measures. The abnormal risk level is calculated according to the abnormal prediction results, and the abnormal risk response measures of the corresponding level are selected to automatically generate warning information of the corresponding abnormal risk level.

8. A DCS early warning system based on intelligent AI visual recognition, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain the real-time operation parameters of the DCS system, wherein the real-time operation parameters include the operation status data, alarm information, operation log and video monitoring data of each subsystem of the DCS system, wherein the operation status data includes the real-time process parameters and equipment operation parameters of each subsystem, and pre-process the real-time operation parameters to obtain the real-time operation parameter features in a unified format; The second unit is used to input the real-time operating parameter features into a pre-built intelligent AI visual recognition model, analyze the operating status of the DCS system, identify the abnormal mode at the system level, compare the abnormal mode with the normal operating mode in a preset normal operating mode library, and calculate the similarity between the abnormal mode and the normal operating mode; The third unit is used to determine that there is an abnormal risk in the current operation of the DCS system if the similarity is lower than the preset abnormal threshold, and use digital twin simulation technology based on physical mechanisms to simulate the real-time operation scenario of the DCS system, predict the failure mode, cause and probability of the DCS system, and automatically generate warning information of the corresponding level based on the prediction results combined with the preset multi-level warning decision matrix, and push the warning information to the intelligent operation and maintenance platform of the DCS system in real time, to assist operators in identifying system risks in advance and provide targeted preventive maintenance measures.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images