Influence evaluation method and device for non-planned shutdown event, medium and product

Abstract

The invention relates to the technical field of computers, in particular to an influence assessment method and device for an unplanned shutdown event, a medium and a product. According to the method, the feature matrix of the refinery flow chart structure is determined through the real-time production data of the target refinery system, and the feature matrix is input into the trained influence prediction model to obtain the non-planned shutdown influence prediction value of each node, so that the evaluation accuracy and efficiency of the non-planned shutdown influence of the refinery plant production device are improved, and the evaluation efficiency of the non-planned shutdown influence of the refinery plant production device is improved. And the risk prevention and control and emergency scheduling capabilities of the enterprise under the non-planned shutdown event are effectively improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, medium and product for assessing the impact of unplanned shutdown events. Background Technology

[0002] In the complex petrochemical production environment, refining and chemical units are susceptible to unplanned shutdowns due to equipment failures, raw material fluctuations, and unforeseen external factors. This can lead to obstructed material flow, coordinated shutdowns of upstream and downstream units, and imbalanced inventory structures, ultimately resulting in severe economic losses and safety risks. Therefore, accurately assessing the impact of unplanned shutdowns on the production process is crucial for improving the risk control capabilities and emergency response levels of petrochemical enterprises.

[0003] However, traditional assessment methods often rely too heavily on predefined rules or correlations between events, making it difficult to accurately characterize the multi-level relationships and global topological characteristics between devices, devices and tanks, and tanks. For example, traditional deductive methods based on failure mode and effects analysis rely excessively on predefined rules and empirical knowledge, making it difficult to depict the dynamic propagation effects under multi-device coupling, which limits the accuracy and applicability of the assessment results. While inductive methods based on causal chains and probabilistic reasoning can reveal statistical correlations between events, they also face challenges of insufficient accuracy and limited generalization ability under high-dimensional topological structures and multi-node coupling conditions.

[0004] Therefore, there is an urgent need for an impact assessment method for unplanned shutdown events to improve the accuracy and efficiency of assessing the impact of unplanned shutdowns of refinery production units. Summary of the Invention

[0005] This invention provides a method, apparatus, storage medium, and program product for assessing the impact of unplanned shutdown events, thereby improving the accuracy and efficiency of assessing the impact of unplanned shutdowns of refinery production units.

[0006] Firstly, this application provides a method for assessing the impact of unplanned work stoppage events, the method comprising:

[0007] Based on the real-time production data of the target refining system, the feature matrix of the refining process flow diagram structure is determined; the refining process flow diagram structure includes multiple processing unit nodes and storage tank nodes, and each node is connected by material flow edges. The feature matrix includes the attribute features of each node and the flow characteristics of each material flow edge.

[0008] The feature matrix is ​​input into the trained impact prediction model to obtain the predicted value of the unplanned shutdown impact of each node. The impact prediction model is based on graph neural network and semi-supervised training.

[0009] Optionally, the impact prediction model is obtained based on the following steps:

[0010] Based on the refining and chemical production process information of the target refining and chemical system, the refining and chemical process diagram structure is constructed.

[0011] Based on the historical data generated by the target refining system, a historical feature matrix of the refining flow chart structure is determined.

[0012] Based on preset material balance constraints and neighborhood propagation strategies, the refining process flow diagram structure is marked with influence labels to obtain multiple labeled nodes with influence label values.

[0013] Based on the refining process flow chart structure, the feature matrix, and multiple labeled nodes, a pre-defined graph neural network model is trained using semi-supervised learning to obtain the influence prediction model.

[0014] Optionally, constructing the refining process flow diagram structure based on the refining production process information of the target refining system includes:

[0015] Based on the device information in the refining and chemical production process information, each processing device node and storage tank node is determined.

[0016] Based on the material transport relationships between nodes, the material flow direction edge is determined to construct the refining process flow diagram structure.

[0017] Optionally, based on preset material balance constraints and neighborhood propagation strategies, the refining process flow diagram structure is labeled with influence to obtain multiple labeled nodes with influence label values, including:

[0018] The simulated shutdown node is determined from the refining process flow diagram structure, and based on the material balance constraints, the influence label value corresponding to the one-hop neighbor node directly connected to the simulated shutdown node is calculated; the influence label value is determined based on the proportional relationship between the flow magnitude of the corresponding connection edge and the node attribute value of the target node;

[0019] Based on the influence label value of the first-hop neighbor node, the influence label value of the second-hop neighbor node is recursively calculated.

[0020] Optionally, the step of semi-supervised learning training of a preset graph neural network model based on the refining process flow chart structure, the feature matrix, and the multiple labeled nodes to obtain the impact prediction model includes:

[0021] The refining process flow diagram structure and the feature matrix are input into the graph neural network model to obtain the predicted value of the impact of unplanned shutdowns at each node.

[0022] The loss value of the graph neural network model is determined based on the difference between the impact label value of each labeled node and the predicted impact value of unplanned work stoppage.

[0023] Based on the loss value, the model parameters of the graph neural network model are iteratively updated until the preset iteration termination condition is met, and the trained impact prediction model is output.

[0024] Optionally, the feature matrix further includes an adjacency matrix and a transpose matrix. The adjacency matrix represents the forward connections between nodes, and the transpose matrix represents the reverse connections between nodes. Then, inputting the refining process flow chart structure and the feature matrix into the graph neural network model to obtain the predicted value of the unplanned shutdown impact of each node includes:

[0025] Based on the forward propagation path corresponding to the adjacency matrix, neighborhood feature convolution processing is performed to obtain the forward convolution result;

[0026] Based on the reverse propagation path corresponding to the transpose matrix, neighborhood feature convolution processing is performed to obtain the reverse convolution result;

[0027] The forward convolution result and the inverse convolution result are fused to obtain the predicted value of the unplanned shutdown impact of each node.

[0028] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the unplanned shutdown events described in the first aspect above for impact assessment.

[0029] Fourthly, this application provides a computer storage medium storing computer program instructions, which are executed by a processor using any of the unplanned shutdown event impact assessment methods described in the first aspect above.

[0030] Fifthly, an embodiment of this application provides a computer program product including computer program instructions, which, when executed by a processor, implement any one of the impact assessment methods for unplanned shutdown events described in the first aspect above.

[0031] The beneficial effects of this invention are as follows:

[0032] This application provides a method, system, equipment, medium, and product for assessing the impact of unplanned shutdown events. The method uses real-time production data of the target refining system to determine the feature matrix of the refining process flow diagram structure, and inputs the feature matrix into a trained impact prediction model to obtain the predicted value of the unplanned shutdown impact of each node. This improves the accuracy and efficiency of assessing the impact of unplanned shutdowns on refinery production units and effectively enhances the enterprise's risk prevention and emergency dispatch capabilities under unplanned shutdown events. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0034] Figure 1 A schematic diagram illustrating the training process of an impact prediction model provided in an embodiment of this application;

[0035] Figure 2 This is a schematic diagram of the structure of an impact prediction model provided in an embodiment of this application;

[0036] Figure 3 A schematic diagram of loss curve comparison provided for an embodiment of this application;

[0037] Figure 4 A schematic diagram of a comparison table of model prediction results and labeled values ​​provided in an embodiment of this application;

[0038] Figure 5 A flowchart illustrating an impact assessment method for unplanned work stoppage events provided in this application embodiment;

[0039] Figure 6 A schematic diagram illustrating the training and application process of an impact prediction model provided in an embodiment of this application;

[0040] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0042] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and this application does not impose limitations.

[0043] The term "and / or" in the embodiments of this application is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0044] It is understood that the following specific embodiments of this application involve data related to refinery production, etc. When the various embodiments of this application are applied to specific products or technologies, relevant licenses or consents are required, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, relevant volunteers can be recruited and agreements can be signed to authorize their data, thereby enabling the implementation using the data of these volunteers; or, implementation can be carried out within an authorized organization, using data from members of the organization to implement the following implementation methods for data management; or, the relevant data used in the specific implementation may be simulated data, such as simulated data generated in a virtual scene.

[0045] The design concept of the embodiments of this application will be briefly introduced below.

[0046] As a vital pillar of the national economy, the petrochemical industry is characterized by highly continuous, tightly coupled, and large-scale production processes, playing a crucial supporting role in energy supply and downstream industrial chains. However, in complex production environments, refining and chemical units are susceptible to unplanned shutdowns due to equipment failures, raw material fluctuations, and unforeseen external factors. Such events often lead to obstructed material flows, coordinated shutdowns of upstream and downstream units, and imbalanced inventory structures, resulting in severe economic losses and safety risks. Therefore, effectively assessing the impact of unplanned unit shutdowns on the production system is of great significance for improving the risk control capabilities and emergency response levels of petrochemical enterprises.

[0047] Unplanned shutdowns have a profound impact on refining and chemical production systems. Accurate impact assessment methods can provide petrochemical enterprises with a scientific basis for decision-making, supporting efficient emergency response and resource allocation. However, traditional assessment methods have significant limitations in complex refining and chemical production systems, struggling to accurately characterize the multi-level relationships and global topological characteristics between units, storage tanks, and storage tanks. For example, deductive methods based on failure mode and effects analysis rely heavily on predefined rules and empirical knowledge, making it difficult to depict the dynamic propagation effects under multi-unit coupling, thus limiting the accuracy and applicability of the assessment results. While inductive methods based on causal chains and probabilistic reasoning can reveal statistical correlations between events, they also face challenges of insufficient accuracy and limited generalization ability under high-dimensional topological structures and multi-node coupling conditions.

[0048] In view of the above problems, this application provides an impact assessment method and impact prediction model for unplanned shutdown events. The method determines the feature matrix of the refining process flow diagram structure through real-time production data of the target refining system, and inputs the feature matrix into a trained impact prediction model to obtain the predicted value of the unplanned shutdown impact of each node, so as to improve the accuracy and efficiency of the assessment of the impact of unplanned shutdown of refinery production units and effectively enhance the enterprise's risk prevention and emergency dispatch capabilities under unplanned shutdown events.

[0049] Furthermore, this application models the refining and chemical production process as a graph structure and constructs a corresponding adjacency matrix and its transpose to capture forward and backward propagation relationships. Based on the collected production data, after Z-Score standardization, impact labeling and node feature construction are performed to complete the construction of the training dataset. Simultaneously, a dual-channel graph neural network impact prediction model is designed and trained using a semi-supervised learning strategy. After training convergence, this model is used to assess the impact of unplanned shutdown events, providing decision support for emergency dispatch.

[0050] The following describes the methods provided by exemplary embodiments of this application in conjunction with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the application scenarios described above are only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.

[0051] In this embodiment of the application, an impact prediction model can be used to predict the impact of unplanned shutdowns of the above-mentioned refining and chemical system. Before the impact prediction model is put into use, it needs to be trained in advance to make the impact prediction model converge. Next, in order to facilitate the description of the model application process, the training process of the impact prediction model will be introduced first.

[0052] Please refer to Figure 1 The diagram below illustrates a training process for an influence prediction model, as provided in an embodiment of this application. The specific implementation flow of this training process is as follows:

[0053] Step 101: Based on the refining and chemical production process information of the target refining and chemical system, construct a refining and chemical process flow diagram structure.

[0054] In this embodiment, a graph structure model is constructed based on the actual refining and chemical production process of the refinery and refining system. Refining and chemical units and storage tanks are uniformly abstracted as graph nodes, and the flow relationships of materials between the units and storage tanks are abstracted as graph edges, thus constructing a directed graph that can represent material transfer relationships. In this way, the refining and chemical process graph structure constructed in this application can include multiple processing unit nodes and storage tank nodes, and the nodes are connected by material flow edges, thereby abstracting the complex physical production system into a mathematical model that can be processed by a computer.

[0055] Specifically, the refining process flow diagram structure in this application can be a directed graph. Each production unit and each storage tank unit is abstracted as a node in the graph, and the physical flow relationship of materials between the units and storage tanks is abstracted as directed edges connecting the nodes, with the direction of the edges pointing from the upstream node to the downstream node. Thus, through the formal expression of the graph structure, this application transforms the concrete, highly coupled refining production system into an abstract model that can be recognized and processed by a computer, laying the foundation for subsequent analysis and calculation. In particular, by introducing the adjacency matrix and its transpose, the potential positive and negative influence propagation channels in the system are fully characterized, providing an accurate topological framework for accurately simulating the bidirectional dynamic propagation of influences.

[0056] In one possible implementation, the embodiments of this application will determine each processing unit node and storage tank node through refining and chemical production process information, and determine the material flow direction edge according to the material conveying relationship between each node, so as to construct a refining and chemical process diagram structure.

[0057] Specifically, this application can use refining and chemical processing units and storage tank units as graph nodes, and the material transport relationship between nodes as directed edges, wherein the direction of the edges is from the upstream node of the material flow to the downstream node.

[0058] Step 102: Based on the historical production data of the target refining system, determine the historical feature matrix of the refining diagram structure.

[0059] In this embodiment of the application, the feature vectors of each node and material flow edge in the refining diagram structure will be determined based on the historical production data of the target refining system, thereby determining the corresponding feature matrix for subsequent model training.

[0060] In one possible implementation, the feature matrix in this application includes the attribute features of the nodes themselves, covering the node state indicating whether the node is in a shutdown state, the node type distinguishing whether the node is a production unit or a storage tank, and the operating parameters corresponding to the node type. For example, a production unit node is its normalized processing load, and a storage tank node is its normalized inventory level. In addition, there are flow association features between nodes and connected edges. For each material flow edge connected to a node in the graph structure, a direction marker indicating the edge relative to the node must be recorded to indicate whether the edge is a feed edge or a discharge edge of the node, and the normalized material flow rate on that edge must also be recorded. The feature vectors of all nodes and edges together constitute the feature matrix of the model input. Thus, through the above feature vectors, the model can learn how to distinguish different types of nodes and their current operating states, and perceive the specific connection relationships and flow rates of nodes in the material network. This provides rich, standardized, and physically meaningful input information for the model to learn the propagation patterns of unplanned shutdowns, enhancing the model's feature representation capabilities and the accuracy of subsequent predictions.

[0061] Specifically, in order to enable the model to effectively learn the propagation laws of influence and accurately distinguish the characteristic differences between shut-down units, normally operating units, and storage tanks, the production data of the target refining system is collected and node features are designed to form the input representation required for subsequent model training. In this embodiment, each node feature may include a node shutdown flag, a node type flag, a node operating load / node inventory level, a material flow direction flag, and a material flow rate. The node operating load applies to unit nodes, and the inventory level applies to storage tank nodes. The corresponding node features are represented as follows:

[0062]

[0063] in, Indicates whether the node is in a stopped state;

[0064] Indicates whether it is a device node;

[0065] This represents the normalized processing load of the device;

[0066] Indicate whether it is a storage tank node;

[0067] This indicates the normalized size of the storage tank inventory;

[0068] For each edge , ,1 represents for The incoming edge (feed), -1 indicates for The edge (material output), 0 indicates and irrelevant

[0069] Representing an edge The normalized flow rate.

[0070] In one possible implementation, the feature matrix of the refining process flow diagram structure in this application embodiment may further include an adjacency matrix and a transpose matrix. The adjacency matrix represents the forward connection relationship between each node, and the transpose matrix represents the reverse connection relationship between each node.

[0071] Specifically, this application embodiment considers that the impact of unplanned shutdowns of refining and chemical plants has a bidirectional propagation characteristic, meaning it not only spreads downstream to downstream units and storage tanks along the material flow direction but also propagates upstream in the reverse direction. This application embodiment will determine the adjacency matrix and transpose matrix based on the refining and chemical graph structure, where the adjacency matrix represents the forward connection relationship between nodes, and the transpose matrix represents the reverse connection relationship between nodes. Thus, this application, by simultaneously constructing the adjacency matrix and its transpose matrix during graph structure modeling, comprehensively depicts the forward and reverse propagation paths, providing a structured foundation for subsequent impact prediction model construction and impact analysis based on graph neural networks.

[0072] Specifically, to mathematically accurately describe the topological relationships of the graph structure, this application generates a corresponding adjacency matrix based on each directed edge. This adjacency matrix represents whether there is a direct material transport relationship between any two nodes. Furthermore, considering that the impact of unplanned shutdowns can propagate downstream along the main material flow direction and may also affect upstream processes, this application generates a transpose of the adjacency matrix to describe the reverse propagation path. In this way, this application transforms the propagation path of unplanned shutdown impacts between devices into the connection relationships of information on a specific graph structure, laying a structured foundation for subsequent modeling based on graph neural networks.

[0073] Specifically, this application will abstract refining and chemical equipment and storage tanks as graph nodes based on the actual refining and chemical production process, and construct a directed graph that reflects the material transfer relationship by using the material transfer relationship between the equipment and storage tanks as edges:

[0074]

[0075] Among them, the node set This represents refining and chemical equipment and storage tanks, with a side assembly. Adjacency matrix represents the material transfer relationship between the equipment and the storage tank. Used to represent the connection relationship between nodes, satisfying:

[0076]

[0077] Meanwhile, considering that the impact of unplanned shutdowns has a two-way propagation characteristic—that is, the impact can spread downstream along the material flow and may also back upstream—the transpose of the adjacency matrix is ​​introduced in the modeling process.

[0078]

[0079] Among them, matrix Used to describe the forward propagation path, matrix Used to describe the reverse propagation path. and The joint characterization can fully depict the potential impact channels of the refining system in both the forward and reverse directions at the topological level, providing a structured basis for the subsequent construction of bidirectional graph convolution modules and the assessment of the impact of unplanned shutdowns.

[0080] In one possible implementation, this application embodiment also considers that there are significant differences in the processing load range of different devices and the upper and lower limits of inventory and capacity of different storage tanks. Directly using the raw data may easily lead to non-convergence in the model training process. The collected production data will be normalized using the Z-Score method to achieve a unified numerical scale, thereby eliminating the differences caused by different physical dimensions and numerical scales and ensuring the stability of model training.

[0081] Specifically, this application will perform Z-Score normalization on the collected production operation data (such as unit load, tank inventory, and side flow rate), that is, for any feature value... The normalization result is:

[0082]

[0083] in, This represents the mean of the feature. This represents its standard deviation. This method can eliminate the influence of different units and scales, ensuring that various numerical features are trained on the same scale.

[0084] Step 103: Based on the preset material balance constraints and neighborhood propagation strategy, mark the impact on the refining process flow diagram structure to obtain the marked nodes.

[0085] In this embodiment, the material balance constraint means that for any node in the production system, the sum of the material flow on all incoming edges should equal the sum of the material flow on all outgoing edges. When a node experiences an unplanned shutdown, the flow on both its incoming and outgoing edges becomes zero. The neighborhood propagation strategy specifies that the influence derivation and labeling are performed only within the two-hop neighborhood of the shutdown node, balancing computational complexity and labeling accuracy. This embodiment calculates the influence label values ​​of upstream and downstream nodes directly connected to the shutdown node within its one-hop neighborhood based on the material balance constraint. Using the labeled one-hop neighborhood nodes as intermediaries, it calculates the influence label values ​​of nodes within their two-hop neighborhood. This combines material balance with neighborhood propagation, automatically generating locally labeled data with clear physical meaning and high reliability without requiring detailed labeling of all nodes. This reduces the model's reliance on large amounts of manually labeled data, providing data support for subsequent semi-supervised learning while ensuring the physical rationality of the generated training data.

[0086] In one possible implementation, embodiments of this application will determine simulated shutdown nodes from the graph structure, and based on material balance constraints, determine the influence label value of the one-hop neighbor node by considering the flow rate of the connection edge between the simulated shutdown node and its directly connected one-hop neighbor node, and the proportional relationship between the node attribute values ​​of the target node. Furthermore, based on the influence label value of the one-hop neighbor node, the influence label values ​​of the two-hop neighbor nodes will be recursively calculated.

[0087] Specifically, this embodiment of the application considers that the impact of unplanned shutdown events will spread along the material flow direction in the refining and chemical production system. It performs constraint reasoning on the input and output paths of the shutdown unit based on material balance constraints, and combines a 2-Hops neighborhood propagation mechanism to deduce the potential impact range on upstream and downstream nodes. Based on this, impact labels are generated for some nodes, forming a labeled dataset with partial supervision signals. This not only reduces the workload of manual data labeling but also effectively enhances the model's generalization ability in complex topological environments, providing reliable support for subsequent semi-supervised training.

[0088] Specifically, in the embodiments of this application, based on material balance constraints, for any node The input and output paths are constrained and inferred to satisfy:

[0089]

[0090] in and Representing nodes respectively The set of incoming edges and the set of outgoing edges, the node where work is stopped. Then it satisfies:

[0091]

[0092] Due to the complex structure of refining and chemical production processes and the highly coupled material flow relationships between units, directly labeling node impacts across the entire process would not only be computationally intensive but also difficult to guarantee accuracy. Therefore, this application's embodiments introduce a 2-Hops neighborhood propagation mechanism, using the shut-down unit as the central node and performing local propagation and labeling only within its two-step neighborhood, thereby achieving high-precision partial labeling of critical affected nodes. Based on this, the constructed impact prediction model is used to learn and infer the impact propagation patterns, thereby achieving impact assessment of units and storage tank nodes across the entire process.

[0093] First, this application will be based on material balance constraints, with shutdown nodes... Determine the feed edge set For each feed edge Identify the corresponding upstream node The system then calculates the impact label value of the node based on the edge flow and the upstream node attributes (inventory or load), thus completing the labeling of the upstream 1-Hop node of the shutdown node. The calculation formula is as follows:

[0094]

[0095] Next, based on the work stoppage point Determine the set of discharge edges For each discharge edge Identify the corresponding downstream nodes The system calculates the impact label value of the node based on the edge flow and downstream node attributes (inventory or load), thus completing the labeling of the downstream 1-Hop node of the shutdown node. The calculation formula is as follows:

[0096]

[0097] Similarly, using nodes Determine the feed edge set Complete the node according to the following formula. upstream 1-Hop node The annotation:

[0098]

[0099] Similarly, using nodes Determine the set of discharge edges According to the type of material in the flow stream, Divided into two categories: one category is , indicating edge With edge The materials are of the same type; one type is... , indicating edge With edge The types of materials are different.

[0100] Therefore, for the discharge edge set Complete the node according to the following formula. Downstream 1-Hop node The annotation:

[0101]

[0102] For the discharge edge collection Complete the node according to the following formula. Downstream 1-Hop node The annotation:

[0103]

[0104] In summary, this application's embodiments, based on material balance constraints and combined with a 2-Hops neighborhood propagation mechanism, achieve the derivation and annotation of impact labels for all unit nodes and tank nodes within a 2-Hops range of the shutdown node. This method ensures the physical rationality of impact propagation, effectively reduces computational complexity, and obtains high-precision partial annotation results within a local range. This constructs an impact annotation strategy based on material balance and neighborhood propagation, generating a dataset with partial supervision signals, providing technical support for subsequent semi-supervised learning of impact prediction models and full-process impact assessment.

[0105] Step 104: Based on the refining process flow chart structure, feature matrix, and multiple labeled nodes, perform semi-supervised learning training on the preset graph neural network model to obtain the influence prediction model.

[0106] In this embodiment, after obtaining the refining process flow diagram structure, feature matrix, and multiple labeled nodes, a graph neural network is constructed. This graph neural network is then trained using semi-supervised learning combined with the labeled nodes to obtain a trained impact prediction model. The graph neural network model can learn patterns for extracting and propagating impact information from node features and graph topology through internal graph convolution and other operations. Through the semi-supervised learning paradigm, the model can fully utilize limited labeled data and a large amount of unlabeled graph structure information for training, effectively learning the complex and non-linear propagation patterns of unplanned shutdown impacts in complex production networks. This significantly improves the model's generalization ability and practicality, enabling it to accurately assess the impact on nodes across the entire process even with only local labeled information.

[0107] In one possible implementation, before training the graph neural network model, this embodiment of the application will also construct an initial model of the graph neural network-based impact prediction model based on graph convolutional layers and linear transformation layers, which is used to characterize the positive and negative impact propagation relationship of unplanned shutdown nodes in the graph structure.

[0108] For details, please refer to Figure 2 The diagram shown is a structural schematic of an impact prediction model provided in an embodiment of this application. The impact prediction model for unplanned shutdowns of refining and chemical plants based on graph neural networks constructed in this embodiment of the application may include the following structure:

[0109] 1) Feature extraction module: Maps the input node features through linear embedding and nonlinear activation functions to form a basic feature representation with a uniform scale.

[0110] 2) Directional projection module: Based on basic features, forward mapping and reverse mapping mechanisms are introduced to generate direction-sensitive forward and backward features respectively, so as to characterize the directional differences in material flow relationships.

[0111] 3) Bidirectional graph convolution module: Constructs forward and reverse neighborhoods using the adjacency matrix and its transpose, and performs multi-layer graph convolution propagation within these neighborhoods to capture local dependency patterns and global topological structure information between nodes.

[0112] 4) Feature fusion module: Adaptively combines the results of forward convolution and inverse convolution through a gating weighting mechanism to obtain a comprehensive feature representation that incorporates upstream and downstream dependency information.

[0113] 5) Residual Enhancement Mapping Module: Apply multi-layer nonlinear mapping and residual updates to the fused features to enhance feature expressiveness and avoid gradient vanishing.

[0114] 6) Output module: Based on the normalization process, a linear transformation is introduced to output the predicted value of the impact of shutdown, so as to ensure the rationality and interpretability of the prediction results under the material balance constraint.

[0115] Thus, the impact prediction model in this embodiment can utilize a multilayer perceptron to deeply extract input features to enhance feature representation capabilities. It also introduces the adjacency matrix and its transpose matrix through a parallel graph convolutional network to achieve joint learning of forward propagation and backward feedback patterns. Furthermore, a feature fusion gating mechanism can be used to achieve adaptive integration of bidirectional features. Finally, a feedforward network is introduced and combined with residual connections and normalization operations to improve the stability and generalization performance of model training, thereby achieving high-precision assessment of the impact of unplanned shutdowns.

[0116] In one possible implementation, the embodiments of this application can input the refining process flow chart structure and feature matrix into the constructed graph neural network model to obtain the predicted value of each node, and determine the loss value of the graph neural network model based on the difference between the influence label value and the predicted value of each labeled node. Then, according to the principle of minimizing the loss value, the model parameters of the graph neural network model are iteratively updated until the preset iteration termination condition is met, and the trained influence prediction model is output.

[0117] Specifically, in this embodiment, the aforementioned training data is used, with SmoothL1Loss as the loss function, and the training step size and learning rate are set. The Adam parameter optimization algorithm is introduced to perform semi-supervised training on the impact prediction model, allowing the model to gradually learn the propagation patterns of unplanned shutdown impacts during the convergence process. After training, the converged model is applied to the analysis of target refinery unit shutdown events, outputting the impact assessment results for the entire system, thereby providing reliable decision support for emergency dispatching in petrochemical enterprises.

[0118] Specifically, in this application embodiment, the impact prediction model is semi-supervised trained as follows:

[0119] Let the set of labeled nodes be... The set of unlabeled nodes is The corresponding impact tag is The model predicts the output as follows: Then the loss function Defined as:

[0120]

[0121]

[0122] in, It is a hyperparameter (usually set to 1).

[0123] During training, the model addresses unlabeled nodes. Semi-supervised learning is achieved by using graph convolution propagation mechanisms to participate in feature updates.

[0124] During training, set the learning rate. With training stride The Adam optimization algorithm is used to iteratively update the model parameters. :

[0125]

[0126] Thus, through multiple rounds of forward and backward propagation, the model gradually learns the propagation patterns of the positive and negative impacts of unplanned shutdowns in the graph structure. After training convergence, the trained model is used to predict the full-process impact of shutdown events on the target unit, obtaining the impact assessment results for each unit and storage tank node.

[0127] In one possible implementation, this application embodiment will describe in detail the model training process based on production process data from a certain refining and chemical enterprise:

[0128] The refining system comprises approximately 14 main secondary processing unit nodes and 18 storage tank nodes, connected by over 140 material flow edges, covering typical processes such as atmospheric and vacuum distillation, catalytic cracking, hydrotreating, and aromatics units. This node and edge information is used to construct a graph structure model of the refining process and generate corresponding adjacency and transpose matrices to support propagation calculations in the bidirectional graph convolution module. This application extracts approximately 50,000 operational data points from a production history database, covering unit operating load, product output, storage tank levels, material flow direction, and flow rate. First, Z-Score normalization is used to normalize data of different dimensions. Then, an impact labeling strategy based on material balance and neighborhood propagation is employed to derive and label the impact labels of locally affected units and storage tank nodes at shutdown nodes, resulting in a dataset with partial supervisory signals for semi-supervised training of the model. The impact prediction model for unplanned shutdowns of refining and chemical plants based on graph neural networks, as described in this application embodiment, comprises a feature extraction module, a directional projection module, a bidirectional graph convolution module, a gated feature fusion module, a residual augmentation mapping and output module. During training, SmoothL1Loss is selected as the loss function, Adam is used as the optimization algorithm, the learning rate is set to 0.001, the training stride is set to 30, and the batch input sample size is 32. In semi-supervised training, only partially labeled node labels are used for supervised computation, while the structural information of unlabeled nodes is combined to guide the model in learning the propagation patterns of shutdown impacts within the network. (Reference) Figure 3The diagram shown illustrates a comparison of loss curves provided in an embodiment of this application. As the number of iterations increases, both the training loss and validation loss of the model exhibit a steady decreasing trend, converging around the 23rd iteration. The training loss decreases from the initial 87.9542 to 0.0005, while the validation loss stabilizes at around 0.0002, indicating that the model maintains good predictive performance on unseen data, avoiding overfitting. (Reference) Figure 4 The diagram shown is a comparison table of model prediction results and labeled values ​​provided in an embodiment of this application. After the model converges, this application applies it to predict unplanned shutdown scenarios of Unit 1 catalytic cracking unit, and the prediction results are as shown above. Figure 4 As shown, the model's prediction results are largely consistent with the known labels, achieving a high level of accuracy. For example, the predicted values ​​for key nodes such as the continuous reforming unit, No. 1 atmospheric and vacuum distillation unit, delayed coking unit, residue hydrotreating unit, and S-zorb unit are highly consistent with the labels, indicating that the model can effectively capture the propagation patterns of shutdown impacts between upstream and downstream units and storage tanks. For some unlabeled nodes (such as No. 1 hydrotreating unit, aromatics unit, and jet fuel feedstock tank group), the model can still provide reasonable prediction values, demonstrating the advantages of semi-supervised learning in handling missing labels. In summary, the overall results show that the model provided in this application can accurately assess the impact of unplanned shutdowns of refining and chemical units and provide a reliable basis for emergency dispatching and production decisions.

[0129] After obtaining the trained impact prediction model, the embodiments of this application can use the impact prediction model to predict the impact of unplanned shutdowns at each node in the refining system, so as to obtain the impact prediction value output by the impact prediction model.

[0130] refer to Figure 5 The diagram shown is a flowchart of an unplanned work stoppage impact assessment method provided in an embodiment of this application. The specific implementation process of this method is as follows:

[0131] Step 501: Based on the real-time production data of the target refining system, determine the feature matrix of the refining process flow diagram structure.

[0132] In this embodiment of the application, the refining process flow diagram structure includes multiple processing unit nodes and storage tank nodes. The nodes are connected by material flow edges, and the feature matrix includes the attribute features of each node and the flow characteristics of each material flow edge.

[0133] Step 502: Input the real-time features into the trained impact prediction model to obtain the predicted value of the unplanned shutdown impact of the node.

[0134] In this embodiment, the impact prediction model is obtained based on the graph neural network and semi-supervised training described in the foregoing embodiments.

[0135] It is worth mentioning that in the embodiments of this application, the prediction process of the unplanned shutdown impact of each node during the training process is the same as the prediction process in the actual application process. Therefore, this process can be referred to the detailed introduction of the aforementioned training process, and will not be elaborated further here.

[0136] In one possible implementation, combined with Figure 6 The diagram illustrates a model training and application process according to an embodiment of this application. This embodiment first models the refining process flow chart, abstracting the refining production process of the refining production system into a graph structure. Next, it processes production data and constructs features, processing real-time and historical production data to generate node and edge features. Through data labeling and dataset construction, influence labels are assigned to some nodes based on material balance and neighborhood propagation strategies, forming a training dataset. Based on this, an influence prediction model based on a graph neural network is constructed and semi-supervised training is performed to allow the model to learn the influence propagation rules. After training, the model can be applied to actual refining production process monitoring, continuously monitoring to determine whether unplanned shutdown events occur in each unit. If no such events occur, the monitoring status is returned; if they do occur, the converged influence prediction model is used to predict the impact, quickly assessing the overall process impact, and finally outputting the impact assessment results to support scheduling decisions.

[0137] Please see Figure 7 As shown, based on the same technical concept, this application also provides a computer device 70. In one embodiment, the computer device can be a device specifically used for assessing the impact of unplanned shutdowns, or it can be a device used for overall control of refinery production. The computer device, as shown... Figure 7 As shown, it includes a memory 701, a communication module 703, and one or more processors 702.

[0138] The memory 701 is used to store computer programs executed by the processor 702. The memory 701 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0139] Memory 701 may be volatile memory, such as random-access memory (RAM); memory 701 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 701 may be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 701 may be a combination of the above-described memories.

[0140] Processor 702 may include one or more central processing units (CPUs) or digital processing units, etc. Processor 702 is used to implement the aforementioned method for assessing the impact of unplanned shutdown events when it calls the computer program stored in memory 701.

[0141] The communication module 703 is used to communicate with the industrial control system.

[0142] This application embodiment does not limit the specific connection medium between the memory 701, communication module 703, and processor 702 described above. This application embodiment... Figure 7 The memory 701 and the processor 702 are connected via a bus 707, which is in... Figure 7 The diagram uses thick lines to describe the components. The connections between other components are for illustrative purposes only and should not be considered limiting. The 707 bus can be divided into address bus, data bus, control bus, etc. For ease of description, ​ It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0143] The memory 701 stores a computer storage medium, which stores computer-executable instructions. The computer-executable instructions are used to implement the impact assessment method for unplanned shutdown events in the embodiments of this application. The processor 702 is used to execute the impact assessment method for unplanned shutdown events in the above embodiments.

[0144] Based on the same inventive concept, embodiments of this application also provide a storage medium storing a computer program that, when run on a computer, causes the computer to perform the steps in the impact assessment method for unplanned shutdown events according to various exemplary embodiments of this application described above.

[0145] In some possible implementations, various aspects of the impact assessment method for unplanned shutdown events provided in this application can also be implemented in the form of a computer program product, which includes a computer program that, when run on a computer device, causes the computer device to perform the steps in the impact assessment method for unplanned shutdown events according to various exemplary embodiments of this application described above. For example, the computer device can perform the steps of the various embodiments.

[0146] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0147] The program product of the embodiments of this application may employ a portable compact disc read-only memory (CD-ROM) and include a computer program, and may run on a computer device. However, the program product of this application is not limited thereto. In this application, the readable storage medium may be any tangible medium that contains or stores a program, and the computer program included therein may be used by or in conjunction with a command execution system, apparatus, or device.

[0148] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a readable computer program. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.

[0149] Computer programs contained on readable media may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0150] Computer programs for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages.

[0151] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0152] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0153] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0154] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0155] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for assessing the impact of unplanned work stoppage events, characterized in that, The method includes: Based on the real-time production data of the target refining system, the feature matrix of the refining process flow diagram structure is determined; the refining process flow diagram structure includes multiple processing unit nodes and storage tank nodes, and each node is connected by material flow edges. The feature matrix includes the attribute features of each node and the flow characteristics of each material flow edge. The feature matrix is ​​input into the trained impact prediction model to obtain the predicted value of the unplanned shutdown impact of each node. The impact prediction model is obtained by semi-supervised training of a graph neural network based on the refining process diagram structure, the feature matrix, and multiple labeled nodes with impact label values. The impact label values ​​are determined based on material balance constraints and neighborhood propagation strategies.

2. The method as described in claim 1, characterized in that, The impact prediction model is obtained based on the following steps: Based on the refining and chemical production process information of the target refining and chemical system, the refining and chemical process diagram structure is constructed. Based on the historical data generated by the target refining system, a historical feature matrix of the refining flow chart structure is determined. Based on preset material balance constraints and neighborhood propagation strategies, the refining process flow diagram structure is marked with influence labels to obtain multiple labeled nodes with influence label values. Based on the refining process flow chart structure, the feature matrix, and multiple labeled nodes, a pre-defined graph neural network model is trained using semi-supervised learning to obtain the influence prediction model.

3. The method as described in claim 2, characterized in that, The construction of the refining process diagram structure based on the refining production process information of the target refining system includes: Based on the device information in the refining and chemical production process information, each processing device node and storage tank node is determined. Based on the material transport relationships between nodes, the material flow direction edge is determined to construct the refining process flow diagram structure.

4. The method as described in claim 2, characterized in that, Based on preset material balance constraints and a neighborhood propagation strategy, the refining process flow diagram structure is labeled with influence tags to obtain multiple labeled nodes with influence tag values, including: The simulated shutdown node is determined from the refining process flow diagram structure, and based on the material balance constraints, the influence label value corresponding to the one-hop neighbor node directly connected to the simulated shutdown node is calculated; the influence label value is determined based on the proportional relationship between the flow magnitude of the corresponding connection edge and the node attribute value of the target node; Based on the influence label value of the first-hop neighbor node, the influence label value of the second-hop neighbor node is recursively calculated.

5. The method as described in claim 2, characterized in that, The step of semi-supervised learning training of a preset graph neural network model based on the refining process flow diagram structure, the feature matrix, and the multiple labeled nodes to obtain the impact prediction model includes: The refining process flow diagram structure and the feature matrix are input into the graph neural network model to obtain the predicted value of the impact of unplanned shutdowns at each node. The loss value of the graph neural network model is determined based on the difference between the impact label value of each labeled node and the predicted impact value of unplanned work stoppage. Based on the loss value, the model parameters of the graph neural network model are iteratively updated until the preset iteration termination condition is met, and the trained impact prediction model is output.

6. The method as described in claim 5, characterized in that, The feature matrix further includes an adjacency matrix and a transpose matrix. The adjacency matrix represents the forward connections between nodes, and the transpose matrix represents the reverse connections between nodes. The step of inputting the refining process flow diagram structure and the feature matrix into the graph neural network model to obtain the predicted impact values ​​of unplanned shutdowns at each node includes: Based on the forward propagation path corresponding to the adjacency matrix, neighborhood feature convolution processing is performed to obtain the forward convolution result; Based on the reverse propagation path corresponding to the transpose matrix, neighborhood feature convolution processing is performed to obtain the reverse convolution result; The forward convolution result and the inverse convolution result are fused to obtain the predicted value of the unplanned shutdown impact of each node.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

8. A computer storage medium storing computer program instructions thereon, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 6.

9. A computer program product comprising computer program instructions, characterized in that, When executed by a processor, the computer program instructions implement the steps of the method according to any one of claims 1 to 6.

Images