Method and device for intelligent multi-modal evaluation of corrosion risk of process pipelines of refining and chemical plants, electronic equipment, storage medium and program product

Abstract

This application provides a method, apparatus, electronic device, storage medium, and program product for multimodal intelligent assessment of corrosion risk in process pipelines of refining and chemical plants. The method includes: acquiring multimodal operation and maintenance data of process pipelines in refining and chemical plants; performing structured processing on the multimodal operation and maintenance data to generate a structured dataset; inputting the structured dataset into a pre-trained natural language processing model, performing semantic parsing on the structured dataset according to a multi-head self-attention mechanism, and outputting entity relationships and feature vectors of the operation and maintenance data; creating a risk knowledge graph based on the entity relationships and feature vectors of the operation and maintenance data; and performing dimensional analysis on the risk knowledge graph using a clustering algorithm to generate corrosion risk identification results for process pipelines in refining and chemical plants. This avoids the problems of existing technologies relying on human experience and identification lag in judging whether there is a risk in process pipelines of refining and chemical plants, and improves the reliability of process pipelines in refining and chemical plants.

Description

Technical Field

[0001] This application relates to the field of equipment operation and maintenance technology, and in particular to a multimodal intelligent assessment method, device, electronic equipment, storage medium and program product for corrosion risk assessment of process pipelines in refining and chemical plants. Background Technology

[0002] Refining and chemical processing units are among the core facilities in petroleum refining. Process pipelines in these units are susceptible to corrosion thinning and stress cracking under harsh conditions of high temperature, high pressure, and strong corrosion. Traditional pipeline maintenance relies on periodic manual inspections, which suffers from problems such as delayed response, numerous blind spots, and weak early warning capabilities.

[0003] In the existing technology, the main methods for inspecting process pipelines in refining and chemical plants are physical inspection methods such as ultrasonic thickness measurement or radiographic testing, which are used to assess the pipeline wall thickness and weld quality to determine whether there are any risks in the process pipelines of the refining and chemical plants.

[0004] However, existing methods rely on human experience, which leads to a lag in risk identification. Summary of the Invention

[0005] This application provides a multimodal intelligent assessment method, device, electronic equipment, storage medium, and program product for corrosion risk assessment of process pipelines in refining and chemical plants, in order to solve the problem of delayed risk identification in the prior art.

[0006] In a first aspect, embodiments of this application provide a multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants, including:

[0007] Acquire multimodal operation and maintenance data of process pipelines in refining and chemical plants;

[0008] The multimodal operation and maintenance data is processed in a structured manner to generate a structured dataset;

[0009] The structured dataset is input into a pre-trained natural language processing model, and semantic parsing of the structured dataset is performed according to a multi-head self-attention mechanism to output entity relationships and feature vectors of the operation and maintenance data.

[0010] A risk knowledge graph is created based on the entity relationships and feature vectors of the operation and maintenance data;

[0011] The risk knowledge graph is analyzed dimensionally using a clustering algorithm to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0012] In one possible implementation, the step of structuring the multimodal operation and maintenance data to generate a structured dataset includes: organizing the multimodal operation and maintenance data according to a pre-created knowledge representation framework to obtain an organized dataset; preprocessing the organized dataset using corpus augmentation technology and temporal feature alignment technology to generate a preprocessed dataset; and segmenting the preprocessed dataset using a domain-specific word segmentation tool to generate a structured dataset.

[0013] In one possible implementation, before inputting the structured dataset into the pre-trained natural language processing model, the method further includes: adjusting the natural language processing model using a refining unit operation and maintenance corpus to obtain a pre-trained natural language processing model; labeling the structured dataset according to the data in the refining unit operation and maintenance corpus to obtain a labeled dataset; and adjusting the attention weights of the pre-trained natural language processing model according to the semantic relevance of the text in the labeled dataset to obtain an adjusted natural language processing model, wherein the adjusted natural language processing model is used to identify entity relationships in the operation and maintenance data.

[0014] In one possible implementation, the step of adjusting the natural language processing model using a refining and chemical plant operation and maintenance corpus to obtain a pre-trained natural language processing model includes: creating a model training set for the natural language processing model based on the refining and chemical plant operation and maintenance corpus; performing predictions on the model training set using a masked language model task to generate prediction results; judging the sentence coherence of the model training set using a next sentence prediction task to generate judgment results; and adjusting the natural language processing model based on the prediction results and the judgment results to obtain the pre-trained natural language processing model.

[0015] In one possible implementation, the step of creating a risk knowledge graph based on the entity relationships and feature vectors of the operation and maintenance data includes: creating risk knowledge graph nodes based on the entity relationships and feature vectors of the operation and maintenance data and the process diagram of the refining and chemical unit's process pipelines; generating a risk topology structure based on the risk knowledge graph nodes and node relationship mappings; and creating a risk knowledge graph based on the risk topology structure.

[0016] In one possible implementation, the step of performing dimensional analysis on the risk knowledge graph using a clustering algorithm to generate corrosion risk identification results for process pipelines in refining and chemical plants includes: dividing the risk nodes in the knowledge graph to generate feature vectors of multiple dimensions; standardizing the feature vectors of multiple dimensions to generate standardized feature vectors; clustering the standardized feature vectors using a clustering algorithm to generate multiple data clusters; and performing feature analysis on the multiple data clusters to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0017] Secondly, embodiments of this application provide a multimodal intelligent assessment device for corrosion risk of process pipelines in refining and chemical plants, comprising:

[0018] The acquisition module is used to acquire multimodal operation and maintenance data of process pipelines in refining and chemical plants;

[0019] The text processing module is used to perform structured processing on the multimodal operation and maintenance data to generate a structured dataset;

[0020] The semantic parsing module is used to input the structured dataset into a pre-trained natural language processing model, perform semantic parsing on the structured dataset according to the multi-head self-attention mechanism, and output the entity relationships and feature vectors of the operation and maintenance data.

[0021] A module is created to generate a risk knowledge graph based on the entity relationships and feature vectors of the operation and maintenance data.

[0022] The dimensional analysis module is used to perform dimensional analysis on the risk knowledge graph through clustering algorithms to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0023] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0024] The memory stores computer-executed instructions;

[0025] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0026] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0027] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0028] The multimodal intelligent assessment method, device, electronic equipment, storage medium, and program product for corrosion risk of refining and chemical plant process pipelines provided in this application acquire multimodal operation and maintenance data of refining and chemical plant process pipelines, generate a structured dataset based on the multimodal operation and maintenance data, perform semantic parsing of the structured dataset using a natural language processing model and a multi-head self-attention mechanism, create a risk knowledge graph based on the entity relationships and feature vectors of the output operation and maintenance data, perform dimensional analysis of the risk knowledge graph using a clustering algorithm, and generate corrosion risk identification results. This avoids the problems of existing technologies that rely on human experience and have identification lag in judging whether there is a risk in refining and chemical plant process pipelines. Attached Figure Description

[0029] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0030] Figure 1 This is a schematic diagram of the system structure of a computer device provided in an embodiment of this application;

[0031] Figure 2 A flowchart illustrating the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants provided in this application. Figure 1 ;

[0032] Figure 3 This is a schematic diagram showing the performance comparison of different models provided in the embodiments of this application;

[0033] Figure 4 A flowchart illustrating the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants provided in this application. Figure 2 ;

[0034] Figure 5 A process diagram of the refining and chemical plant process piping provided in the embodiments of this application;

[0035] Figure 6 A schematic diagram of cluster center analysis provided in an embodiment of this application;

[0036] Figure 7 This is a visualization diagram of the clustering results provided in an embodiment of this application;

[0037] Figure 8 A schematic diagram of the multimodal intelligent assessment device for corrosion risk of process pipelines in refining and chemical plants provided in this application;

[0038] Figure 9 A schematic diagram of the structure of the electronic device provided in this application.

[0039] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0040] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0041] First, let's explain the names involved in this application:

[0042] Multimodal: refers to information that is of a variety of types or forms, with a dataset containing information in different "modalities" or "channels".

[0043] Multidimensional: The data or analysis contains multiple different attributes, variables or features.

[0044] Refining and chemical plants are among the core facilities in petroleum refining. Process pipelines in these plants are susceptible to corrosion thinning and stress cracking under harsh conditions of high temperature, high pressure, and strong corrosion. Traditional pipeline maintenance relies on periodic manual inspections, which suffers from slow response times, numerous blind spots, and weak early warning capabilities. Current technologies primarily utilize physical inspection methods such as ultrasonic thickness measurement or radiographic testing to assess pipeline wall thickness and weld quality, determining the presence of risks. However, these methods rely on manual experience, leading to delays in risk identification.

[0045] To address the aforementioned technical problems, this application proposes the following technical concept: The inventors considered structuring multimodal operation and maintenance data, constructing a structured dataset, introducing a natural language processing model, and combining it with a multi-head self-attention mechanism to perform semantic parsing on the structured dataset, obtaining entity relationships and feature vectors of the operation and maintenance data. Furthermore, considering the creation of a risk knowledge graph, a clustering algorithm is used to perform dimensional analysis on the risk knowledge graph, generating corrosion risk identification results for refining and chemical plant process pipelines. This avoids the problems of existing technologies relying on human experience and lagging identification when judging whether there are risks in refining and chemical plant process pipelines.

[0046] Figure 1 This is a schematic diagram of the system architecture of the computer device provided in an embodiment of this application. Figure 1As shown, the computer device includes: a receiving device 101, a processing device 102, and a display device 103.

[0047] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the method for identifying risks in process pipelines of refining and chemical plants. In other feasible embodiments of this application, the above architecture may include more or fewer components than illustrated, or combine some components, or split some components, or arrange different components, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of both.

[0048] In the specific implementation process, the receiving device 101 can be an input / output interface or a communication interface, which can acquire multimodal operation and maintenance data of the process pipeline of the refining and chemical unit.

[0049] The processing device 102 can generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0050] The display device 103 can be used to display the corrosion risk identification results of the process pipelines of the above-mentioned refining and chemical equipment.

[0051] The display device can also be a touch screen, used to receive user commands while displaying the above content, so as to realize the operation interaction with the user.

[0052] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions from memory and executing those instructions, or it can be implemented by a chip circuit.

[0053] Furthermore, the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0054] Figure 2 A flowchart illustrating the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants provided in this application. Figure 1 ,like Figure 2 As shown, the method includes:

[0055] S201: Obtain multimodal operation and maintenance data of process pipelines in refining and chemical units.

[0056] In this embodiment, the multimodal operation and maintenance data includes text data such as historical accident records, on-site operation logs, equipment maintenance reports, process operation instructions, and risk analysis documents of the refining and chemical unit.

[0057] Specifically, textual descriptions are used to organize the semantic information of pipeline corrosion.

[0058] The original text structure includes, but is not limited to, “risk description”, “corrosion phenomenon”, “cause analysis”, and “response measures”.

[0059] Specifically, after structuring, a knowledge expression framework is formed with "corrosion event - inducing mechanism - prevention and control measures" as the main line.

[0060] Specifically, by linking process time-series data, corrosion monitoring trends, and other multimodal auxiliary information, a corrosion risk research framework is constructed with text as the semantic backbone and multi-source data as support.

[0061] S202: Perform structured processing on multimodal operation and maintenance data to generate structured datasets.

[0062] Specifically, the multimodal operation and maintenance data is organized according to the pre-created knowledge representation framework, the organized dataset is preprocessed using corpus augmentation technology, and the preprocessed operation and maintenance dataset is segmented using a domain-specific word segmentation tool to generate a structured dataset.

[0063] Specifically, multi-source data is fused and preprocessed. Text data (such as accident reports and inspection logs), time-series data (such as temperature, pressure, and flow signals collected by process sensors), and corrosion monitoring data (such as ultrasonic thickness measurement and electrochemical corrosion rate) generated during the operation and maintenance of process pipelines in refining and chemical units are collected. The multi-source data is uniformly cleaned, standardized, and spatiotemporally aligned to construct a dataset with standardized structure and multimodal association.

[0064] For example, based on the text record in the inspection log dated September 12, 2023, at 10:25 AM, which states "White crystals and a damp feel were found at the flange of the R-101 outlet pipe, suspected to be a precursor to acid leakage," and by correlating the temperature and pressure time-series data of that pipe section in the DCS system for the corresponding period (September 11, 2023, 6:00 PM to September 12, 2023, 12:00 PM), it was found that from 6:00 PM the previous day, the pipe temperature continuously rose from the normal value of around 120℃ to a peak of 152℃ during the time recorded in the text record. Simultaneously, the pressure fluctuated from 2.8 MPa to 3.4 MPa, with the standard deviation of the fluctuation increasing by 40%. The time-series data exhibits a synergistic characteristic of a continuous abnormal increase in temperature and significant pressure fluctuation. The text-aligned content of the time-series characteristics is shown in Table 1 below.

[0065] Table 1. Text Alignment of Abnormal Temporal Features

[0066]

[0067] For example, based on the text description in the corrosion assessment report dated October 20, 2023, that "significant corrosion thinning was found at the bend of the P-202 acid transport pipeline, with the thickness gauge reading decreasing by approximately 0.5 mm compared to the previous month," and by correlating the continuous wall thickness data of the corresponding monitoring point (UT-P202-ELBOW01) in the ultrasonic online thickness measurement system from August 1 to October 20, 2023, the analysis shows that the wall thickness of this pipe section decreased from an initial 12.5 mm to 11.8 mm at the time of recording, a cumulative reduction of 0.7 mm, which is basically consistent with the qualitative description of "approximately 0.5 mm" in the text. Furthermore, the calculated average corrosion rate reached 3.14 mm / year, falling into the category of high corrosion rate. The textual correlation information of the corrosion data is shown in Table 2 below.

[0068] Table 2 Corrosion Data Text Correlation

[0069]

[0070] S203: Input the structured dataset into the pre-trained natural language processing model, perform semantic parsing on the structured dataset based on the multi-head self-attention mechanism, and output the entity relationships and feature vectors of the operation and maintenance data.

[0071] Specifically, text data serves as the semantic backbone, integrating multimodal operation and maintenance data such as time series and corrosion data. Entities and composite relationships are extracted through a domain-adaptive BERT model and a multi-head attention mechanism, and multimodal features such as time series sensor data and online corrosion monitoring trends are collaboratively integrated within the semantic framework.

[0072] Specifically, a deep learning model based on a multi-head self-attention mechanism is used to perform semantic analysis on the word vectorized text, extracting entity relationships and feature vectors from the text.

[0073] In this embodiment, the pre-trained natural language processing model is the BERT model.

[0074] For example, Figure 3 This is a schematic diagram comparing the performance of different models provided in the embodiments of this application.

[0075] like Figure 3As shown, the F1 scores differ by 14.4 percentage points, indicating significant differences in the processing of specialized texts across different architectures. The BERT-Transformer model, combining BERT's pre-trained knowledge with the attention mechanism of the Transformer architecture, fine-tunes the text of refining equipment to establish domain semantic understanding, capturing complex dependencies between equipment, media, and risk. Its precision of 92.1% indicates high accuracy in extracting relationships and a low false positive rate; its recall of 90.3% indicates that the model can cover most real-world relationships; and its F1 score is optimal, achieving a balance between precision and recall. The ERNIE3.0 model's performance analysis shows that all metrics are below 90%. Although it incorporates external knowledge, its coverage of specialized domain knowledge is limited, exhibiting an incomplete knowledge base of chemical engineering terminology and insufficient depth of understanding of equipment-specific relationships. The traditional BiLSTM-CRF model has the lowest values ​​across all metrics, struggling to accurately identify complex semantic relationships, missing numerous implicit relationships, and exhibiting problems such as inability to model long-distance dependencies, weak understanding of terminology combinations, and limited relational reasoning capabilities.

[0076] S204: Create a risk knowledge graph based on the entity relationships and feature vectors of operation and maintenance data.

[0077] Specifically, risk knowledge graph nodes are created based on the entity relationships and feature vectors of operation and maintenance data and the process diagrams of refining and chemical unit process pipelines. Risk topology is generated based on node relationship mapping, and risk knowledge graph is created based on risk topology.

[0078] S205: Use clustering algorithms to perform dimensional analysis on the risk knowledge graph and generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0079] Specifically, risk nodes in the knowledge graph are divided using a clustering algorithm to generate multi-dimensional feature vectors. These multi-dimensional feature vectors are then standardized and analyzed to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0080] As can be seen from the above embodiments, by acquiring multimodal operation and maintenance data of process pipelines in refining and chemical plants, generating a structured dataset based on the multimodal operation and maintenance data, using a natural language processing model and a multi-head self-attention mechanism to perform semantic parsing on the structured dataset, creating a risk knowledge graph based on the entity relationships and feature vectors of the output operation and maintenance data, and performing dimensional parsing on the risk knowledge graph through a clustering algorithm to generate corrosion risk identification results, the problem of existing technologies relying on human experience and having identification lag in judging whether there are risks in process pipelines of refining and chemical plants is avoided.

[0081] In one embodiment of this application, step S202 includes:

[0082] S2021: Based on the pre-created knowledge representation framework, the multimodal operation and maintenance data is processed to obtain the processed dataset.

[0083] In this embodiment, the multimodal operation and maintenance data is the raw data, and the structure of the raw data is divided into: risk and hazard, accident type, cause analysis and control measures.

[0084] For example, a pre-created knowledge representation framework can be an "event-cause-measure" knowledge representation framework.

[0085] S2022: The processed dataset is preprocessed using corpus augmentation and temporal feature alignment techniques to generate a preprocessed dataset.

[0086] In this embodiment, corpus augmentation techniques include, but are not limited to, synonym replacement, back translation, template generation, and adversarial generation.

[0087] S2023: Use a domain-specific word segmentation tool to segment the preprocessed dataset to generate a structured dataset.

[0088] Specifically, a domain-enhanced word segmentation tool was used to segment the text information of the refining and chemical plant, identifying proper nouns such as safety accidents, protective equipment, emergency actions, and leaks.

[0089] For example, the text without word segmentation is as follows: When a safety accident or emergency occurs, clearly specify the necessary protective equipment that the discoverer should be equipped with and the means by which to check and confirm.

[0090] For example, the segmented text is as follows: / When / a safety accident / and / an emergency / occurs / , clearly / the discoverer / should / be / equipped / with / the / necessary / protective equipment / and / by / what / means / to / inspect / confirm / .

[0091] As can be seen from the above embodiments, the multimodal operation and maintenance data is organized by a pre-created knowledge representation framework, and the organized dataset is preprocessed using corpus expansion technology and temporal feature alignment technology to generate a preprocessed dataset, thereby improving the model's semantic understanding of text. The preprocessed dataset is then segmented using a domain-specific word segmentation tool to generate a structured dataset, thereby improving the model's accuracy and efficiency in text recognition.

[0092] Figure 4 A flowchart illustrating the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants provided in this application. Figure 2 ,like Figure 4 As shown, before step S203, the following steps are also included:

[0093] S301: The natural language processing model is adjusted using the refining and chemical plant operation and maintenance corpus to obtain a pre-trained natural language processing model.

[0094] Specifically, a training set for a natural language processing model is created based on the corpus of refining and chemical plant operation and maintenance. The training set is then used to make predictions through a masked language model task to generate prediction results. The coherence of sentences in the training set is then judged through a next sentence prediction task to generate judgment results. Based on the prediction results and judgment results, the natural language processing model is adjusted to obtain a pre-trained natural language processing model.

[0095] S302: Annotate the structured dataset based on the data in the refining and chemical plant operation and maintenance corpus to obtain an annotated dataset.

[0096] Specifically, the BIO text annotation method is used to annotate the structured dataset after word segmentation.

[0097] BIO text annotation categorizes each word into one of three types: B (Beginning), I (Inside), and O (Outside). Specifically, B marks the first word of a named entity, I marks the non-first word part of the named entity, and O marks the non-entity part.

[0098] Specifically, the word index is converted into a word vector tensor with contextual semantics through the embedding layer of the BERT model, thus maintaining the integrity of the semantic structure.

[0099] S303: Adjust the attention weights of the pre-trained natural language processing model based on the semantic relevance of the text in the labeled dataset to obtain the adjusted natural language processing model, which is used to identify entity relationships in the operation and maintenance data.

[0100] Specifically, attention weights are adjusted through domain gating networks to improve the accuracy of terminology recognition.

[0101] Specifically, in the self-attention mechanism, each input vector The data is mapped to three vectors: query vector Q, key vector K, and value vector V, which are transformed by a linear transformation matrix. , and ,get:

[0102]

[0103] By calculating the dot product of the query vector Q and the key vector K, the relevance (i.e., attention weight) of each word in the sequence to other words is obtained.

[0104] The correlation is normalized using the softmax function:

[0105]

[0106] in, It is the dimension of the key vector K, used to prevent the dot product from becoming too large.

[0107] As can be seen from the above embodiments, by adjusting the natural language processing model using the refining and chemical plant operation and maintenance corpus, the model's ability to understand domain-specific terms and semantic relationships can be improved. By annotating the structured dataset, the model can identify entity relationships in the operation and maintenance data. The attention weight of the natural language processing model can be adjusted according to the semantic relevance of the text in the annotated dataset, thereby improving the accuracy of the model in identifying entities.

[0108] In one embodiment of this application, step S301 includes:

[0109] S3011: Create a training set for a natural language processing model based on the corpus of refining and chemical plant operation and maintenance.

[0110] Specifically, the model is pre-trained using the MLM (Masked Language Model) task and the next sentence prediction task, and then fine-tuned for downstream tasks using the trained model.

[0111] Specifically, various types of information are encoded at the input of the BERT model, including token embedding (TE), segment embedding (SE), and position embedding (PE).

[0112] Specifically, TE maps each tag to a corresponding vector representation through a vocabulary, generating different segment embeddings for each sentence, indicating whether the tag belongs to the first sentence or the second sentence.

[0113] Specifically, PE (Personal Token) records the position of a token within a sentence. For a given token, its final input to the BERT model is the sum of the three:

[0114]

[0115] In the formula, Indicates the final input; Indicates the marked input; This indicates segmented input; This indicates the position input.

[0116] S3012: Predict the model training set through the masked language model task and generate prediction results.

[0117] Specifically, the MLM task is used for pre-training, randomly masking 15% of the labels in the input sequence, and requiring the model to predict the masked labels based on the context.

[0118] Specifically, for the obscured markers The goal of the model is to train it by maximizing the probability of the occluded markers:

[0119]

[0120] In the formula, The loss function is represented during the training of the masked language model; M represents the set of masked tags. Indicates removal Other tags besides those.

[0121] S3013: Determine the sentence coherence in the model training set through the next sentence prediction task and generate a judgment result.

[0122] Specifically, sentence pairs are constructed from the model training set. In 50% of cases, the second sentence is the actual follow-up sentence of the first sentence in the original document (positive example); in the other 50% of cases, the second sentence is randomly extracted from other locations in the corpus (negative example). The model needs to determine whether the two sentences are consecutive in the original text.

[0123] S3014: Adjust the natural language processing model based on the prediction and judgment results to obtain a pre-trained natural language processing model.

[0124] Specifically, after pre-training is completed, fine-tuning is performed for the task while keeping the model architecture unchanged. Further training is then conducted on the adjusted task, using supervised learning with labeled data for the given task to adjust the model's parameters.

[0125] As can be seen from the above embodiments, by jointly training the model through the masked language model task and the next sentence prediction task, and adjusting the model according to the prediction results and judgment results, the model's ability to understand the context of domain terms can be improved.

[0126] In one embodiment of this application, step S204 includes:

[0127] S2041: Create risk knowledge graph nodes based on the entity relationships and feature vectors of the operation and maintenance data and the process diagrams of the refining and chemical unit's process pipelines.

[0128] Figure 5 A process diagram of the refining and chemical plant process piping provided in the embodiments of this application.

[0129] In this embodiment, a directed graph structure consisting of nodes and directed edges is used to represent knowledge in text operation and maintenance, where nodes represent entities or concepts, and directed edges represent semantic relationships between entities or concepts.

[0130] In this embodiment, the basic unit of knowledge is organized and stored in the form of a triple consisting of a head entity, a relation, and a tail entity.

[0131] Specifically, entities and relationships are identified and extracted from multimodal data through information extraction; knowledge fusion is performed to align and merge multi-source knowledge; the fused knowledge is transformed into a machine-processable and computable representation, i.e., knowledge representation; and the structured knowledge is persistently saved to a database through knowledge storage.

[0132] S2042: Generate a risk topology structure based on the nodes and node relationship mapping of the risk knowledge graph.

[0133] Specifically, through node relationship mapping, risk factors such as acid settling tank corrosion, reactor coking, and seal failure are linked with multi-dimensional data such as process parameters, equipment status, and maintenance records to form a complete risk evolution topology.

[0134] S2043: Create a risk knowledge graph based on the risk topology.

[0135] Specifically, graph database technology is used to construct a knowledge graph of the operation and maintenance risks of process pipelines in refining and chemical plants.

[0136] In this embodiment, entities, attributes and relationships are extracted based on multiple open information sources. Through the triplet data model, data collection, information extraction and knowledge fusion steps are performed, and the extracted elements are stored in the knowledge graph in a structured form.

[0137] In this embodiment, data relationships are presented in a graphical form using visualization technology.

[0138] As can be seen from the above embodiments, knowledge graph nodes are created by using entity relationships and feature vectors of text data and pipeline process diagrams. A risk topology structure is generated based on the node mapping relationship. A risk knowledge graph is created based on the risk topology structure. By creating a risk knowledge graph, the visualization and reasoning of risk knowledge can be realized.

[0139] In one embodiment of this application, step S205 includes:

[0140] S2051: Divide the risk nodes in the knowledge graph and generate feature vectors with multiple dimensions.

[0141] Specifically, risk nodes in the knowledge graph are divided, risk nodes are extracted from the knowledge graph, and a feature vector containing four dimensions, namely "semantic similarity", "frequency of occurrence", "severity" and "scope of influence", is constructed for each node.

[0142] S2052: Standardizes feature vectors of multiple dimensions to generate standardized feature vectors.

[0143] Specifically, the feature vectors are Z-score standardized to eliminate the influence of dimensions.

[0144] S2053: Clustering standardized feature vectors using a clustering algorithm to generate multiple data clusters.

[0145] Specifically, the number of clusters, k, in risk analysis, is related to the number of risk levels.

[0146] For example, a security risk assessment system may contain n risk indicators (n k), randomly select 2 indicators from n risk indicators as the initial cluster centers. and Calculate the cluster distance of each risk indicator relative to the initial cluster center:

[0147]

[0148] In the formula, and The cluster distance from each risk indicator to the initial cluster center; Let i be the i-th data point in the dataset; Indicates cluster center Location; Indicates cluster center The location.

[0149] Specifically, compare the arrival times of each data point at the initial cluster center. and The cluster distance is used to assign data points to the clusters corresponding to the nearest cluster centers, thus forming clusters with the initial cluster centers as the cores. Based on the results of the first cluster analysis, the cluster centers of each cluster are recalculated. and The new cluster center is the mean of all data points within the cluster after the first clustering.

[0150] Specifically, based on the determined new cluster center and Calculate each risk indicator relative to the cluster center and The cluster distance. Specifically, all feature data are processed to standardize the dimensions and orders of magnitude of each indicator, eliminating the influence of dimensions.

[0151] Specifically, the elbow method is used to first determine the k value in combination with the silhouette coefficient method. The elbow method observes the inflection point of the sum of squared errors (SSE) within the cluster corresponding to different k values. The silhouette coefficient evaluates the tightness and separation of the clusters of each sample, and the k value corresponding to its peak value is taken to enhance the reliability of the results.

[0152] Specifically, the K-means algorithm is used to cluster the standardized multidimensional data. After clustering, the risk feature nodes of each cluster are defined by analyzing the centroids of each cluster across all feature dimensions.

[0153] Specifically, the K-means clustering algorithm is used to segment the risk nodes related to corrosion in the knowledge graph. Risk nodes directly related to corrosion are extracted from the knowledge graph, and a corrosion feature vector containing four dimensions, namely "corrosion rate trend", "wall thickness reduction degree", "medium corrosivity" and "environmental corrosion intensity", is constructed for each node.

[0154] Specifically, the feature vectors are Z-score standardized to eliminate the influence of dimensions.

[0155] Specifically, to determine the optimal number of clusters, a range of candidate k values ​​(k=2 to 10) was set, and clustering calculations were performed sequentially. The clustering effect was evaluated by analyzing the sum of squared intra-cluster deviations corresponding to different k values.

[0156] For example, N=50 risk nodes were extracted from the risk knowledge graph of the refining and chemical plant's process pipelines, and a 4-dimensional feature vector was constructed for each node: semantic similarity, frequency of occurrence, severity, and scope of impact. All features were Z-score standardized. The data matrix X has dimensions (50, 4), and the candidate k values ​​are set to a range of k=2, 3, 4, 5, 6, 7, 8, 9, 10. Data X is clustered to obtain k clusters. SSE is the sum of squares of the Euclidean distances from all sample points to the centroid of their respective clusters. It measures the compactness within a cluster. The smaller the SSE, the more similar the samples within the cluster.

[0157]

[0158] In the formula, k represents the number of clusters; Indicates the i-th cluster; This represents the centroid of the i-th cluster; Cluster One of the sample points; This represents the square of the Euclidean distance from the sample point to the centroid.

[0159] For example, a range of candidate k values ​​is set (k=2 to 10), and clustering calculations are performed sequentially. The clustering effect is evaluated by analyzing the sum of squared deviations within clusters corresponding to different k values. The specific SSE calculation results are shown in Table 3 below:

[0160] Table 3. SSE results calculated for different K values

[0161]

[0162] Figure 6 This is a schematic diagram of cluster center analysis provided in an embodiment of this application.

[0163] Figure 7 This is a schematic diagram of the clustering results provided in an embodiment of this application.

[0164] like Figure 6 and Figure 7 As shown, SSE drops sharply when k increases from 2 to 4. From k=4 onwards, the curve flattens out, and the rate of SSE decrease decreases significantly. k=4 can be considered the "inflection point" or "elbow" of the curve. The "SSE Decrease Value" column in the table shows that: from k 2 to 3, SSE decreases by 55.5; from k 3 to 4, SSE decreases by 30.0; from k 4 to 5, the SSE decreases sharply to 17.0, and then continues to decrease slowly. k=4 is a clear "elbow point." This indicates that when the data is divided into 4 clusters, the benefit of increasing the number of clusters to improve "intra-cluster tightness" (SSE decrease) begins to decrease significantly. Choosing k=4 ensures good clustering results while avoiding the generation of too many meaningless sub-clusters.

[0165] S2053: Perform feature analysis on multiple data clusters to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0166] For example, using a defined k value (k=4), the standardized corrosion risk feature vector is clustered, and corrosion risk nodes within each cluster are considered to have similar multidimensional corrosion characteristics. Based on the centroid values ​​of the samples within each cluster on the two core features of corrosion rate trend and wall thickness reduction, the corrosion risk is divided into four levels: Cluster 1 (C1) represents extremely high corrosion risk; Cluster 2 (C2) represents high corrosion risk; Cluster 3 (C3) represents medium corrosion risk; and Cluster 4 (C4) represents low corrosion risk. The risk level classification criteria are shown in Table 4 below:

[0167] Table 4 Risk Level Classification Standards

[0168]

[0169] For example, the risk feature identification results are shown in Table 5 below:

[0170] Table 5 Risk Identification Results

[0171]

[0172] As can be seen from the above embodiments, risk nodes in the knowledge graph are divided by clustering algorithms to obtain feature vectors of multiple dimensions. The feature vectors are then standardized and analyzed to generate pipeline corrosion risk identification results, providing data-driven basis for pipeline corrosion prevention and control.

[0173] Figure 8 This is a schematic diagram of the multimodal intelligent assessment device for corrosion risk of process pipelines in refining and chemical plants provided in this application, as shown below. Figure 8 As shown, the multimodal intelligent assessment device 80 for corrosion risk of process pipelines in refining and chemical plants provided in this embodiment includes: an acquisition module 801, a text processing module 802, a semantic parsing module 803, a creation module 804, and a dimension parsing module 805.

[0174] The acquisition module 801 is used to acquire multimodal operation and maintenance data of process pipelines in refining and chemical plants.

[0175] The text processing module 802 is used to perform structured processing on multimodal operation and maintenance data to generate structured datasets.

[0176] The semantic parsing module 803 is used to input the structured dataset into the pre-trained natural language processing model, perform semantic parsing on the structured dataset according to the multi-head self-attention mechanism, and output the entity relationships and feature vectors of the operation and maintenance data.

[0177] Create module 804 to create a risk knowledge graph based on entity relationships and feature vectors from operation and maintenance data.

[0178] The dimensional analysis module 805 is used to perform dimensional analysis on the risk knowledge graph through clustering algorithms to generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0179] In one possible implementation, the text processing module 802 includes:

[0180] The data processing unit is used to process multimodal operation and maintenance data according to a pre-created knowledge representation framework to obtain a processed dataset.

[0181] The preprocessing unit is used to preprocess the sorted dataset using corpus augmentation and temporal feature alignment techniques to generate a preprocessed dataset.

[0182] The word segmentation unit is used to segment the preprocessed dataset using a domain-specific word segmentation tool to generate a structured dataset.

[0183] In one possible implementation, the multimodal intelligent assessment device 80 for corrosion risk of process pipelines in refining and chemical plants further includes:

[0184] The first adjustment module is used to adjust the natural language processing model using the refining and chemical plant operation and maintenance corpus to obtain a pre-trained natural language processing model.

[0185] The standard module is used to annotate structured datasets based on data from the refining and chemical plant operation and maintenance corpus, resulting in an annotated dataset.

[0186] The second adjustment module is used to adjust the attention weights of the pre-trained natural language processing model based on the semantic relevance of the text in the labeled dataset, so as to obtain the adjusted natural language processing model, which is used to identify entity relationships in the operation and maintenance data.

[0187] In one possible implementation, the first adjustment module includes:

[0188] The first creation unit is used to create a model training set for a natural language processing model based on the corpus of refining and chemical plant operation and maintenance.

[0189] The prediction unit is used to make predictions on the model training set through the masked language model task and generate prediction results.

[0190] The judgment unit is used to judge the coherence of sentences in the model training set based on the next sentence prediction task and generate a judgment result.

[0191] The adjustment unit is used to adjust the natural language processing model based on the prediction and judgment results to obtain a pre-trained natural language processing model.

[0192] In one possible implementation, the creation module 804 includes:

[0193] The second creation unit is used to create risk knowledge graph nodes based on the entity relationships and feature vectors of the operation and maintenance data and the process diagrams of the refining and chemical unit's process pipelines.

[0194] The generation unit is used to generate a risk topology based on the nodes and node relationship mapping of the risk knowledge graph.

[0195] The third creation unit is used to create a risk knowledge graph based on the risk topology.

[0196] In one possible implementation, the dimension parsing module 805 includes:

[0197] The partitioning unit is used to divide risk nodes in the knowledge graph and generate feature vectors with multiple dimensions.

[0198] The standardization unit is used to standardize feature vectors of multiple dimensions to generate standardized feature vectors.

[0199] Clustering units are used to cluster standardized feature vectors using clustering algorithms to generate multiple data clusters.

[0200] The feature parsing unit is used to perform feature parsing on multiple data clusters and generate corrosion risk identification results for process pipelines in refining and chemical plants.

[0201] The multimodal intelligent assessment device for corrosion risk of process pipelines in refining and chemical plants provided in this embodiment can execute the method provided in the above-mentioned method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0202] Figure 9 A schematic diagram of the structure of the electronic device provided in this application. Figure 9 As shown, the electronic device 90 provided in this embodiment includes at least one processor 901 and a memory 902. Optionally, the electronic device 90 further includes a communication component 903. The processor 901, memory 902, and communication component 903 are connected via a bus.

[0203] In the specific implementation process, at least one processor 901 executes computer execution instructions stored in memory 902, causing at least one processor 901 to execute the above-mentioned process pipeline risk identification method for refining and chemical equipment.

[0204] The specific implementation process of processor 901 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0205] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0206] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0207] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0208] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for identifying risks in process pipelines of a refining and chemical plant.

[0209] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-mentioned method for identifying risks in process pipelines of refining and chemical plants.

[0210] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0211] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0212] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0213] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0214] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0215] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0216] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0217] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants, characterized in that, include: Acquire multimodal operation and maintenance data of process pipelines in refining and chemical plants; The multimodal operation and maintenance data is processed in a structured manner to generate a structured dataset; The structured dataset is input into a pre-trained natural language processing model, and semantic parsing of the structured dataset is performed according to a multi-head self-attention mechanism to output entity relationships and feature vectors of the operation and maintenance data. A risk knowledge graph is created based on the entity relationships and feature vectors of the operation and maintenance data; The risk knowledge graph is analyzed dimensionally using a clustering algorithm to generate corrosion risk identification results for process pipelines in refining and chemical plants.

2. The method according to claim 1, characterized in that, The step of structuring the multimodal operation and maintenance data to generate a structured dataset includes: The multimodal operation and maintenance data is processed according to the pre-created knowledge representation framework to obtain the processed dataset; The sorted dataset is preprocessed using corpus augmentation and temporal feature alignment techniques to generate a preprocessed dataset. The preprocessed dataset is segmented using a domain-specific word segmentation tool to generate a structured dataset.

3. The method according to claim 1, characterized in that, Before inputting the structured dataset into the pre-trained natural language processing model, the method further includes: The natural language processing model was adjusted by using the corpus of refining and chemical plant operation and maintenance data to obtain a pre-trained natural language processing model. The structured dataset is labeled based on the data in the refining and chemical plant operation and maintenance corpus to obtain a labeled dataset; The attention weights of the pre-trained natural language processing model are adjusted based on the semantic relevance of the text in the labeled dataset to obtain the adjusted natural language processing model, wherein the adjusted natural language processing model is used to identify entity relationships in the operation and maintenance data.

4. The method according to claim 3, characterized in that, The process of adjusting the natural language processing model using a corpus of refining and chemical plant operation and maintenance data to obtain a pre-trained natural language processing model includes: A training set for a natural language processing model was created based on a corpus of refining and chemical plant operation and maintenance data. The model training set is used to make predictions through a masked language modeling task to generate prediction results; The system uses the next sentence prediction task to determine the coherence of sentences in the model training set and generates a judgment result. The natural language processing model is adjusted based on the prediction results and the judgment results to obtain a pre-trained natural language processing model.

5. The method according to claim 1, characterized in that, The step of creating a risk knowledge graph based on the entity relationships and feature vectors of the operation and maintenance data includes: Risk knowledge graph nodes are created based on the entity relationships and feature vectors of operation and maintenance data, as well as the process diagrams of refining and chemical plant process pipelines. A risk topology is generated based on the nodes and node relationship mapping of the risk knowledge graph. A risk knowledge graph is created based on the risk topology.

6. The method according to any one of claims 1 to 5, characterized in that, The step of performing dimensional analysis on the risk knowledge graph using a clustering algorithm to generate corrosion risk identification results for process pipelines in refining and chemical plants includes: Risk nodes in the knowledge graph are segmented to generate feature vectors with multiple dimensions; The feature vectors of the multiple dimensions are standardized to generate standardized feature vectors; The standardized feature vectors are clustered using a clustering algorithm to generate multiple data clusters; Feature analysis is performed on the multiple data clusters to generate corrosion risk identification results for process pipelines in refining and chemical plants.

7. A multimodal intelligent assessment device for corrosion risk of process pipelines in a refining and chemical plant, characterized in that, include: The acquisition module is used to acquire multimodal operation and maintenance data of process pipelines in refining and chemical plants; The text processing module is used to perform structured processing on the multimodal operation and maintenance data to generate a structured dataset; The semantic parsing module is used to input the structured dataset into a pre-trained natural language processing model, perform semantic parsing on the structured dataset according to the multi-head self-attention mechanism, and output the entity relationships and feature vectors of the operation and maintenance data. A module is created to generate a risk knowledge graph based on the entity relationships and feature vectors of the operation and maintenance data. The dimensional analysis module is used to perform dimensional analysis on the risk knowledge graph through clustering algorithms to generate corrosion risk identification results for process pipelines in refining and chemical plants.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory, causing the processor to perform the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The invention includes a computer program that, when executed by a processor, implements the multimodal intelligent assessment method for corrosion risk of process pipelines in refining and chemical plants as described in any one of claims 1 to 6.

Images