Gas storage three-in-one operation and maintenance risk factor incidence relation identification method and equipment

Abstract

The invention provides a gas storage three-in-one operation and maintenance risk factor incidence relation identification method and equipment, and relates to the technical field of information extraction. The method comprises the following steps: identifying a plurality of operation and maintenance risk events in the operation and maintenance process of a target gas storage from a target text set; inputting the plurality of operation and maintenance risk events into a construction unit to construct a plurality of target event pairs; generating a first vector of each target event pair through a vector generation unit; extracting a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector of the target event pair; and inputting the target fusion vector of each target event pair into a prediction unit to obtain an association relationship between the two target operation and maintenance risk events in each target event pair. According to the method, the problems that in the operation and maintenance process of the gas storage, identification of the incidence relation of all the risk factors lacks an effective tool and is low in accuracy are solved.

Description

Technical Field

[0001] The present application relates to the technical field of information extraction, and in particular to a method and device for identifying correlation relationships of three-in-one operation and maintenance risk factors of a gas storage facility. Background Art

[0002] A gas storage facility is a facility for storing natural gas. As a complex system of "geological body-wellbore-ground", its operation stage not only plays a vital role in energy security and supply stability, but also faces many potential risks and challenges. During the operation stage of a gas storage facility, a large amount of safety management texts and event reporting texts will be generated. These texts record the risk factors of the gas storage facility in multiple dimensions, such as the natural state of the geological body, the engineering characteristics of the wellbore, and the operation of the ground facilities. There are multiple correlations between the risk factors, such as concurrency, composition, and causality. Extracting the correlation between the risk factors is of great significance for understanding the causes and laws of gas storage accidents.

[0003] At present, the analysis of the correlation between various risk factors of gas storage can be carried out by using association rule mining algorithms, that is, by extracting frequent item sets to discover potential relationships in the data. However, the association rule mining algorithm cannot provide causal relationships. It can only describe the correlation between risk factors, resulting in a large amount of implicit information that cannot be effectively mined and it is difficult to show the full picture of the accident. In recent years, with the widespread use of pre-trained language large models, these language large models can be used to predict the hidden correlations among various risk factors. However, this implicit causal extraction requires more complex feature construction. While focusing on overall features, it may ignore the importance of local features, resulting in the model being unable to fully consider subtle changes and key information in various risk factors, thereby affecting the accurate extraction of risk factor correlations.

[0004] In summary, there is a lack of effective tools for identifying the correlation between various risk factors in the operation and maintenance process of gas storage facilities, and the accuracy is low. Summary of the invention

[0005] The present application provides a method and device for identifying the correlation relationship of risk factors of three-in-one operation and maintenance of a gas storage facility, which is used to solve the problem that there is a lack of effective tools and low accuracy in identifying the correlation relationship of various risk factors in the operation and maintenance process of a gas storage facility.

[0006] In a first aspect, the present application provides a method for identifying the correlation relationship of three-in-one operation and maintenance risk factors of a gas storage, comprising:

[0007] Acquire a target text set related to operation and maintenance risks of a target gas storage, and identify multiple target operation and maintenance risk events of the target gas storage from the target text set;

[0008] Inputting multiple target operation and maintenance risk events into the construction unit to construct multiple target event pairs, the target event pairs including: any two target operation and maintenance risk events;

[0009] Generate a first vector for each target event pair by a vector generation unit;

[0010] Extracting a first local feature vector and a first overall feature vector from the first vector by a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector of the target event pair;

[0011] The target fusion vector of each target event pair is input into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

[0012] Optionally, the feature extraction unit includes: a first bidirectional gated recurrent unit and a multi-head self-attention mechanism, and extracting a first local feature vector and a first overall feature vector from the first vector through the feature extraction unit includes:

[0013] extracting a first local feature vector from the first vector by a first bidirectional gated recurrent unit;

[0014] A first overall feature vector is extracted from the first vector through a multi-head self-attention mechanism.

[0015] Optionally, the vector generation unit includes: a pre-trained language model and a word vector model, and generating a first vector for each target event pair through the vector generation unit includes:

[0016] Generate a second vector for each target event pair through the pre-trained language model;

[0017] Generate a third vector for each target event pair through the word vector model;

[0018] For each target event pair, the second vector and the third vector of the target event pair are fused into the first vector of the target event pair.

[0019] Optionally, a pre-trained named entity recognition model is pre-stored, and the named entity recognition model includes: a conversion unit, a second bidirectional gated recurrent unit, and a recognition unit, and identifies multiple target operation and maintenance risk events of the target gas storage from the target text set, including:

[0020] Each target text in the target text set is converted into a target word vector through a conversion unit;

[0021] Through the second bidirectional gated recurrent unit, feature extraction is performed on each target word vector to obtain a target feature vector;

[0022] The target feature vector of each target text is input into the recognition unit to identify the target operation and maintenance risk events existing in each target text.

[0023] Optional pre-training of named entity recognition models, including:

[0024] Acquire a sample text set related to operation and maintenance risks of a sample gas storage, the sample text set including a plurality of sample texts and an event label of each sample text, the event label being used to indicate that the sample text is a risk factor, an event type, or a non-entity object;

[0025] Convert each sample text into a sample word vector;

[0026] Perform feature extraction on each word vector to obtain a sample feature vector;

[0027] The named entity recognition model is trained through the sample feature vectors and event labels corresponding to multiple sample texts.

[0028] In a second aspect, the present application provides a training method for a relation extraction model, the relation extraction model comprising a construction unit, a vector generation unit and a feature extraction unit, the method comprising:

[0029] Acquire a sample text set related to operation and maintenance risks of a sample gas storage, and identify multiple sample operation and maintenance risk events of the sample gas storage from the sample text set;

[0030] Inputting a plurality of sample operation and maintenance risk events into a construction unit to construct at least one event triplet, wherein the event triplet includes: a sample event pair consisting of any two sample operation and maintenance risk events, and a relationship label between the sample event pairs;

[0031] Generate a fourth vector for each sample event pair and a label vector for each relationship label by a vector generation unit;

[0032] Extracting a second local feature vector and a second overall feature vector from the fourth vector by a feature extraction unit, and fusing the second local feature vector and the second overall feature vector to obtain a sample fusion vector;

[0033] The relationship extraction model is trained by fusion vectors and label vectors of samples corresponding to multiple event triplets.

[0034] In a third aspect, the present application provides a gas storage three-in-one operation and maintenance risk factor correlation identification device, comprising:

[0035] An acquisition module is used to acquire a target text set related to the operation and maintenance risk of a target gas storage, and identify multiple target operation and maintenance risk events of the target gas storage from the target text set;

[0036] A construction module is used to input multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs, where the target event pairs include: any two target operation and maintenance risk events;

[0037] A vector module, used for generating a first vector for each target event pair through a vector generating unit;

[0038] A feature module, configured to extract a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fuse the first local feature vector and the first overall feature vector to obtain a target fused vector of the target event pair;

[0039] The prediction module is used to input the target fusion vector of each target event pair into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

[0040] In a fourth aspect, the present application provides an electronic device, comprising: a processor, and a memory communicatively connected to the processor;

[0041] Memory stores computer-executable instructions;

[0042] When the processor executes the computer-executable instructions stored in the memory, it is used to implement a method for identifying the correlation relationship of risk factors of three-in-one operation and maintenance of a gas storage facility according to the first aspect of the invention.

[0043] In a fifth aspect, the present application provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, they are used to implement a method for identifying the correlation relationship of risk factors of the three-in-one operation and maintenance of a gas storage facility according to the invention content of the first aspect.

[0044] In a sixth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, is used to implement a method for identifying correlation relationships between risk factors of three-in-one operation and maintenance of a gas storage facility according to the invention content of the first aspect.

[0045] The present application provides a method and device for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir, including: obtaining a target text set related to the operation and maintenance risk of the target gas storage reservoir, and identifying multiple target operation and maintenance risk events of the target gas storage reservoir from the target text set; inputting multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs, the target event pairs including: any two target operation and maintenance risk events; generating a first vector for each target event pair through a vector generation unit; extracting a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector for the target event pair; inputting the target fusion vector of each target event pair into a prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair. The present application obtains the target fusion vector of the target event pair by fusing the first local feature vector and the first overall feature vector, so as to better understand the contextual information of the text, thereby analyzing the correlation relationship of the target event pair, and solves the problem of lack of effective tools and low accuracy in identifying the correlation relationship of various risk factors in the operation and maintenance process of the gas storage reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 A flow chart of a method for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage provided in an embodiment of the present application Figure 1 ;

[0048] Figure 2 A flow chart of a method for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage provided in an embodiment of the present application Figure 2 ;

[0049] Figure 3 A diagram of the named entity recognition model architecture provided for an embodiment of the present application;

[0050] Figure 4 A flowchart of a training method for a relationship extraction model provided in an embodiment of the present application;

[0051] Figure 5 A flow chart of a training method for a multi-feature fusion relationship extraction model provided in an embodiment of the present application;

[0052] Figure 6A schematic diagram of the structure of a gas storage three-in-one operation and maintenance risk factor correlation relationship identification device provided in an embodiment of the present application;

[0053] Figure 7 A schematic diagram of the structure of the electronic device hardware provided in an embodiment of the present application.

[0054] Reference numerals:

[0055] 310-acquisition module; 320-construction module; 330-vector module; 340-feature module; 350-prediction module;

[0056] 410 - processor; 420 - memory; 430 - communication component; 440 - bus. DETAILED DESCRIPTION

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0058] In the embodiments of the present application, words such as "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit the difference. It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or more advantageous than other embodiments or design. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way. In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more.

[0059] It should be noted that the "at..." in the embodiments of the present application can be the instant when a certain situation occurs, or it can be a period of time after a certain situation occurs, and the embodiments of the present application do not specifically limit this. In addition, the method for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir provided in the embodiments of the present application is only used as an example, and the method for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir can also include more or less content.

[0060] As a complex system of "geological body-wellbore-ground", the operation stage of gas storage not only plays a vital role in energy security and supply stability, but also faces many potential risks and challenges. During the operation stage of gas storage, a large amount of safety management texts and event reporting texts will be generated. These texts record the risk factors of gas storage in multiple dimensions, such as the natural state of the geological body, the engineering characteristics of the wellbore, and the operation of ground facilities. There are multiple correlations between the risk factors, such as concurrency, composition, and causality. Extracting the correlation between the risk factors is of great significance for understanding the causes and laws of gas storage accidents.

[0061] At present, the analysis of the correlation between various risk factors of gas storage can be carried out by using association rule mining algorithms, that is, by extracting frequent item sets to discover potential relationships in the data. However, the association rule mining algorithm cannot provide causal relationships. It can only describe the correlation between risk factors, resulting in a large amount of implicit information that cannot be effectively mined and it is difficult to show the full picture of the accident. In recent years, with the widespread use of pre-trained language large models, these language large models can be used to predict the hidden correlations among various risk factors. However, this implicit causal extraction requires more complex feature construction. While focusing on overall features, it may ignore the importance of local features, resulting in the model being unable to fully consider subtle changes and key information in various risk factors, thereby affecting the accurate extraction of risk factor correlations.

[0062] In summary, the correlation analysis of various risk factors in gas storage operation and maintenance has the problem of low accuracy.

[0063] Based on this, the embodiment of the present application provides a method and device for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage facility, which can be used in the field of information extraction technology, aiming to solve the above technical problems of the prior art.

[0064] Figure 1 A flow chart of a method for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage facility provided in an embodiment of the present application Figure 1 .like Figure 1 As shown, the method includes:

[0065] S101. Acquire a target text set related to operation and maintenance risks of a target gas storage, and identify multiple target operation and maintenance risk events of the target gas storage from the target text set.

[0066] Specifically, the target gas storage is the gas storage with the correlation between the operation and maintenance risk factors to be identified; the target text set is the text set of the correlation between the risk factors in the text to be identified, and the target operation and maintenance risk event is an event related to the operation and maintenance risk of the trinity of "geological body-wellbore-ground" of the gas storage, such as pig blockage. The target gas storage production site event report set, hazard and operability analysis (HAZOP) report, hidden danger management ledger, unsafe behavior or status report and other text data are collected, and the text data is comprehensively sorted out by corpus expansion and manual sorting to form the target text set. Multiple target operation and maintenance risk events of the target gas storage are identified from the target text set by the named body recognition method.

[0067] S102: Input multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs.

[0068] Specifically, based on the pre-trained relation extraction model, the correlation of the three-in-one operation and maintenance risk factors of the gas storage is identified. The pre-trained relation extraction model includes: a construction unit, a vector generation unit, a feature extraction unit and a prediction unit; the construction unit is used to combine the target operation and maintenance risk events in pairs in a certain order to construct a target event pair; the target event pair includes: any two target operation and maintenance risk events.

[0069] S103: Generate a first vector for each target event pair through a vector generating unit.

[0070] Specifically, the vector generation unit includes a model that converts text into vectors, such as a predictive training language model (BERT), a word vector model (word2vec), a document vector model (doc2vec), etc., which is used to convert the target event pair into a vector for subsequent processing; the first vector is a vector representing the text semantics of the target event pair.

[0071] S104 , extracting a first local feature vector and a first overall feature vector from the first vector by a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector of the target event pair.

[0072] Specifically, the feature extraction unit is used to extract vector features, and can be composed of different combinations of convolutional neural networks, time series network models, bidirectional gated recurrent units (BiGRU), and multi-head self-attention mechanisms. The first local feature vector is a vector indicating the local features of the text corresponding to the first vector; the first overall feature vector is a vector indicating the overall features of the text corresponding to the first vector; the target fusion vector includes the first local feature vector and the first overall feature vector.

[0073] S105: Input the target fusion vector of each target event pair into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

[0074] Specifically, the prediction unit is used to predict the correlation between target operation and maintenance risk events; the target fusion vector of each target event pair is input into the prediction unit, and the prediction unit predicts the correlation between two target operation and maintenance risk events according to the target fusion vector.

[0075] The embodiment of the present application provides a method for identifying the association relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir, including: obtaining a target text set related to the operation and maintenance risk of the target gas storage reservoir, and identifying multiple target operation and maintenance risk events of the target gas storage reservoir from the target text set; inputting multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs, the target event pairs including: any two target operation and maintenance risk events; generating a first vector for each target event pair through a vector generation unit; extracting a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector for the target event pair; inputting the target fusion vector of each target event pair into a prediction unit to obtain the association relationship between the two target operation and maintenance risk events in each target event pair. The present application obtains the target fusion vector of the target event pair by fusing the first local feature vector and the first overall feature vector, so as to better understand the contextual information of the text, thereby analyzing the association relationship of the target event pair, and solves the problem of lack of effective tools and low accuracy in identifying the association relationship of each risk factor in the operation and maintenance process of the gas storage reservoir.

[0076] Figure 2 A flow chart of a method for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage facility provided in an embodiment of the present application Figure 2 In this embodiment Figure 1 Based on the example, the method for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage facility is described in detail. Figure 2 As shown, the method includes:

[0077] S201. Obtain a target text set related to operation and maintenance risks of a target gas storage facility.

[0078] S201 is similar to S101 and will not be described in detail in this embodiment.

[0079] S202. Convert each target text in the target text set into a target word vector through a conversion unit; extract features from each target word vector to obtain a target feature vector through a second bidirectional gated recurrent unit; input the target feature vector of each target text into a recognition unit to identify target operation and maintenance risk events existing in each target text.

[0080] Specifically, a pre-trained named entity recognition model is pre-stored, and the named entity recognition model includes: a conversion unit, a second bidirectional gated recurrent unit, and a recognition unit. The conversion unit is used to convert text into a word vector, the second bidirectional gated recurrent unit is used to extract features, and the recognition unit is used to identify operation and maintenance risk events in the text according to the feature vector corresponding to the text. The target word vector is a vector representation of the target text, and the target feature vector is a vector of the target text feature.

[0081] Among them, the pre-training process of the named entity recognition model includes: obtaining a sample text set related to the operation and maintenance risks of the sample gas storage, the sample text set includes multiple sample texts and event labels for each sample text; converting each sample text into a sample word vector; performing feature extraction on each word vector to obtain a sample feature vector; and training the named entity recognition model through the sample feature vectors and event labels corresponding to each of the multiple sample texts.

[0082] Specifically, the sample text set is the text set used for model training; the event label is used to indicate whether the sample text is a risk factor, event type, or non-entity object.

[0083] In one example, after preprocessing the accident data by removing stop words, searching dictionaries, and using normative descriptions, accurate word segmentation is performed, and each piece of data is labeled using the BIO standard in the annotation tool (Brat) to obtain event labels. Specifically, the accident data can be labeled from three aspects: geological body risk, wellbore risk, and surface engineering risk to find out the causes of the accident. The BIO label set is shown in Table 1. The labeled accident data is used as a sample text set and input into the subsequent model for training.

[0084] Table 1

[0085] Corresponding tag name meaning Start Tag Middle Mark End Tag HUM Geological risk B-HUM I-HUM O-HUM MAT Wellbore risk B-MAT I-MAT O-MAT MAN Ground engineering risks B-MAN I-MAN O-MAN ACI Type of accident / incident B-ACI I-ACI O-ACI OTH Non-Entity O O 0

[0086] Figure 3 The architecture diagram of the named entity recognition model provided in the embodiment of the present application is as follows: Figure 3 As shown in Figure 1, the BERT+BiGRU+CRF model architecture is used for model training. N-1 , X N Etc. are the identifiers of the input layer, representing different elements of the sequence data received by the model, and the sample text set is input through the input layer. The BERT layer is used as the base and combined with BiGRU to better capture the context information of the sentence; then the output of the BiGRU layer is input to the fully connected layer, and then input to the conditional random field (CRF). The state transfer matrix in the CRF is used to capture the dependencies between entity labels. The output layer outputs the entity sequence to improve the effect of entity recognition. E1, E2, E N-1 、E NEqual to T1, T2, T N-1 , T N etc. are the identifiers of the input nodes and output nodes of the BERT layer, Trm is the identifier of the intermediate nodes of the BERT layer, GRU is the identifier of the gate recurrent unit (GRU) of the BiGRU layer, h1, h2, h N-1 、h N etc. are the identifiers of the nodes in the fully connected layer.

[0087] First, the preprocessed sample text set is input into the BERT model to obtain the contextual information of each word and generate the corresponding sample word vector. The advantage of the BERT model is that it can fully capture the semantic information and contextual relationship of the language, which helps to improve the accuracy of entity recognition.

[0088] Next, the output of the BERT model is used as the input of the BiGRU model. The BiGRU model will further learn the feature representation of the text sequence, especially the sequence information, and extract the features of each word vector to obtain the sample feature vector. The advantage of the BiGRU model is that it can effectively capture the long-term dependencies in the text sequence, which helps to improve the accuracy of entity recognition.

[0089] Among them, GRU is designed with two gate control units, namely the reset gate and the update gate. The reset gate discards the information z t and update the information updated by the update gate t As shown in formula (1) and formula (2) respectively. Where: σ is the sigmoid activation function, which is used as a gating signal to control the value within [0, 1]; w r and u r are the input weight matrix and the loop weight matrix of the reset gate respectively; x t is the input information of the current node; w z and u z are the input weight matrix and the loop weight matrix of the update gate respectively; h t-1 is the hidden layer state at the previous moment.

[0090] r t =σ(w r x t +u r h t-1 ) (1)

[0091] z t =σ(w z x t +u z h t-1 ) (2)

[0092] Create a new candidate hidden state in the tanh function ☉ is the Hadamard product, which multiplies the reset gate activation matrix by the corresponding element of its weight matrix, and then calculates the candidate hidden state As shown in formula (3), where w and u are the input weight matrix and the loop weight matrix of the unit state respectively.

[0093]

[0094] Let h t is the hidden layer state at the current moment. When updating information, h t As shown in formula (4).

[0095]

[0096] The bidirectional GRU network structure obtains the word vector from the BERT layer and inputs it into the forward GRU and backward GRU for feature extraction, which can make full use of the feature vector to connect the context information. CRF decoding is used to determine the label corresponding to the current sample text.

[0097] Finally, the output of the BiGRU model is connected to the CRF (Conditional Random Field) layer, and the named entity recognition model is trained through the sample feature vectors and event labels corresponding to multiple sample texts. The CRF layer can jointly model the entire sequence and consider the relationship between entity labels, thereby further improving the accuracy and continuity of entity recognition.

[0098] During the model training process, the F1 score will be used as the evaluation basis, and the models with good results will be saved. Among them, F1 is the balanced F score, which is the harmonic mean of precision and recall. After the training is completed, the named entity recognition model is obtained to realize the function of inputting sentences and outputting entities. The target text set is processed by the pre-trained named entity recognition model to obtain multiple target operation and maintenance risk events E1 = {e1, e2…e n}.

[0099] In one example, a sample text set is input into BERT to generate sample word vectors, and the training set, verification machine, and test set are divided according to a preset ratio of 8:1:1. In the text input stage, the maximum sentence truncation length is set to 256, and the number of sentences in each training batch is 16. In the word vector representation stage, the pre-trained model used is BERT-base-chinese, and the vector dimension is BERT's default 768 dimensions. In the semantic encoding stage, BERT's default 12-layer Transformer encoder is used. In the model training stage, the loss rate is set to 0.1 and the learning rate is set to 5e -5 The training rounds are 20. The accident text (part) after preprocessing is shown in Table 2, and the specific parameter settings are shown in Table 3.

[0100] Table 2

[0101]

[0102] Table 3

[0103] Parameter name Parameter meaning Parameter Value max_seq_length Maximum sentence truncation length 256 batch_size Number of sentences in a single batch 16 model_name BERT Version BERT-base-chinese learning_rate Learning Rate 5e-5 drop_out Loss Rate 0.1 epochs Training rounds 20 log_step Log output frequency 100

[0104] The training set is input into the bidirectional gated recurrent unit BiGRU, and then the output of the BiGRU layer is input into the conditional random field CRF for training. During the model training process, the F1 score is used as the evaluation basis, and the model with good effect is saved. After the training is completed, the relationship extraction model is used through the model loading and the code used to implement the function of inputting sentences and outputting entities, and multiple target operation and maintenance risk events E1 are obtained. The model recognition effect is shown in Table 4 below.

[0105] Table 4

[0106] Dataset Precision Recall F1 Validation Dataset 76.36％ 86.19％ 0.8097 Test Dataset 75.30％ 83.96％ 0.7939

[0107] S203: Input multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs, where the target event pairs include any two target operation and maintenance risk events.

[0108] Specifically, the events included in multiple target operation and maintenance risk events E1 are grouped in pairs in turn to obtain the target event pair set E2 = {…e ij …}, where e ij = <e i , e j > represents the target event pair, e i , e j represents target operation and maintenance risk event i, target operation and maintenance risk event j, i, j are integers, and 1≤i≠j≤n, n is the total number of target operation and maintenance risk events.

[0109] S204, generating a second vector for each target event pair through a pre-trained language model; generating a third vector for each target event pair through a word vector model; for each target event pair, merging the second vector and the third vector of the target event pair into a first vector of the target event pair.

[0110] Specifically, the vector generation unit includes: pre-trained language model and word vector model; using BERT pre-trained language model to transform e ij Mapped to the second vector, denoted by E 1mij , whose set is denoted by E 1m ; Use word2vec word vector model to transform e ij Mapped to the third vector, denoted by E 2mij , whose set is denoted by E 2m .

[0111] E 1m and E 2m Fusion forms the first vector

[0112] S205, extracting a first local feature vector from the first vector through a first bidirectional gated recurrent unit; extracting a first overall feature vector from the first vector through a multi-head self-attention mechanism, and fusing the first local feature vector and the first overall feature vector to obtain a target fused vector for the target event pair;

[0113] Specifically, the feature extraction unit includes: a first bidirectional gated recurrent unit and a multi-head self-attention mechanism. In the multi-view feature fusion stage, the first vector E of the gas storage field event pair extracted by BERT and word2vec is 3m The data are input into the BiGRU and multi-head self-attention mechanism models respectively, and the bidirectional perspective of BiGRU and the global perspective of the multi-head self-attention mechanism are used to extract the local features and overall features of the text.

[0114] The first local eigenvector output under the bidirectional perspective is e c , the first overall eigenvector output from a global perspective is e r , e c and e r Fusion forms the final feature vector Use it as the target fusion vector.

[0115] S206: Input the target fusion vector of each target event pair into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

[0116] S206 is similar to S105 and will not be described in detail in this embodiment.

[0117] A method for identifying the association relationship of three-in-one operation and maintenance risk factors of a gas storage reservoir provided in an embodiment of the present application includes: obtaining a target text set related to the operation and maintenance risk of a target gas storage reservoir, and identifying multiple target operation and maintenance risk events of the target gas storage reservoir from the target text set; inputting the multiple target operation and maintenance risk events into a construction unit to construct multiple target event pairs, the target event pairs including: any two target operation and maintenance risk events; generating a first vector for each target event pair through a vector generation unit; extracting a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector for the target event pair; inputting the target fusion vector of each target event pair into a prediction unit to obtain the association relationship between the two target operation and maintenance risk events in each target event pair. The present application obtains the target fusion vector of the target event pair by fusing the first local feature vector and the first overall feature vector, so as to better understand the contextual information of the text, thereby analyzing the correlation relationship of the target event pair, and solving the problem of low accuracy in the correlation relationship analysis of various risk factors in the operation and maintenance of the gas storage reservoir; by fusing two word embedding methods such as BERT and word2vec, their respective advantages are fully utilized to provide richer semantic information and capture the implicit causal relationship in the text. At the same time, by combining the advantages of BERT and word2vec in different levels of information and generalization ability, the model's understanding ability and robustness for causal relationship extraction tasks are enhanced. By introducing BiGRU as part of the model, it can simultaneously obtain information before and after the current moment, more comprehensively capture the contextual information in the input sequence, and help to better understand and model the correlation between risk factors in text sequences; by adding a multi-head attention mechanism, it can pay attention to the relationship between different positions in parallel, making up for the lack of traditional models in handling long-distance dependencies; by combining the bidirectional perspective (BiGRU) and the global feature perspective (multi-head attention mechanism), a multi-perspective learning framework is formed, which comprehensively utilizes information at multiple levels to more comprehensively understand and model data, thereby improving the robustness and generalization ability of the model.

[0118] Figure 4 A flow chart of a training method for a relation extraction model provided in an embodiment of the present application. Figure 2 Based on the embodiment, the training method of the relationship extraction model is described. Figure 4 As shown, the method includes:

[0119] In a second aspect, the present application provides a training method for a relation extraction model, the relation extraction model comprising a construction unit, a vector generation unit and a feature extraction unit, the method comprising:

[0120] S301. Obtain a sample text set related to operation and maintenance risks of a sample gas storage, and identify multiple sample operation and maintenance risk events of the sample gas storage from the sample text set.

[0121] S301 is similar to S101 and S202 and will not be described in detail in this embodiment.

[0122] S302: Input multiple sample operation and maintenance risk events into a construction unit to construct at least one event triplet, where the event triplet includes: a sample event pair consisting of any two sample operation and maintenance risk events, and a relationship label between the sample event pairs.

[0123] Specifically, by constructing units, the events included in multiple sample operation and maintenance risk events E1 are grouped in pairs in turn to obtain a sample event pair set E2 = {…e ij …}, where e ij = <e i , e j > represents a sample event pair, e i , e j represents sample operation and maintenance risk event i, sample operation and maintenance risk event j, i, j are integers, and 1≤i≠j≤n, n is the total number of sample operation and maintenance risk events. ij Perform relationship annotation to form event triples <e i , e j ,y ij >, denoted as E ij Among them, y ij ∈{0, 1, 2} is the relationship label of the sample event pair. Taking causal relationship as an example, 0 means e i With e j No association; 1 means e i Yes j Cause event; 2 means e i Yes j After the marking is completed, the event triple E3 = {…E ij …}.

[0124] S303 . Generate a fourth vector for each sample event pair and a label vector for each relationship label through a vector generating unit.

[0125] Specifically, the BERT pre-trained language model is used to convert e ij Mapped to vector E 1mij , whose set is denoted by E 1m ; Use word2vec word vector model to transform e ij Mapped to vector E 2mij , whose set is denoted by E 2m .

[0126] E 1m and E 2m The fusion generates a fourth vector for each sample event pair

[0127] y ij Perform one-hot encoding to generate label vector Y emij , whose collection constitutes the matrix Y Em .

[0128] S304 , extracting a second local feature vector and a second overall feature vector from the fourth vector through a feature extraction unit, and fusing the second local feature vector and the second overall feature vector to obtain a sample fusion vector.

[0129] S301 is similar to S204 and will not be described in detail in this embodiment.

[0130] S305 , training a relationship extraction model through sample fusion vectors and label vectors corresponding to multiple event triplets.

[0131] Specifically, the order of the event triples E3 in the gas storage field is randomly disrupted, and the sample fusion vectors and label vectors corresponding to multiple event triples are divided into training sets, validation sets, and test sets according to a preset ratio, such as 8:1:1. The training set and validation set are input into the model for training to obtain a model that meets the set optimization goals, so as to extract the correlation between gas storage risk factors and build a risk feature library of correlation between gas storage fields.

[0132] In one example, the number of training sets, validation sets, and test sets into which multiple event triplets are divided is shown in Table 5.

[0133] Table 5

[0134] Dataset name Interpretation quantity train.txt Training Dataset 1700 dev.txt Validation Dataset 212 test.txt Test Dataset 213

[0135] During the model training process, it will be divided into multiple cycles (Epochs) for training. In each Epoch, the data is divided into multiple batches and input into the model. After the training set completes the training of an entire Epoch, the validation set is input into the model for evaluation. F1 is used as the evaluation indicator, and the model W with the highest F1 value is saved as the relationship extraction model. In the testing phase, the test set is input again to obtain the sample fusion vector and label vector, and the model W is tested to determine the correlation between the event pairs in the gas storage field in the data, and the correlation risk feature library is output, which contains the correlation information between the event pairs in the gas storage field, providing support for the accurate identification of operation and maintenance risk factors.

[0136] In one example, the relationship recognition effect of the relationship extraction model is shown in Table 6 below.

[0137] Table 6

[0138] Dataset meaning Precision Recall F1 dev.txt Validation Dataset 76.66％ 76.77％ 0.7670 test.txt Test Dataset 80.18％ 80.22％ 0.8019

[0139] In one example, Figure 5 A flow chart of a training method for a multi-feature fusion relationship extraction model provided in an embodiment of the present application, such as Figure 5 As shown in the figure, we first enter the event entity recognition stage, perform text preprocessing and annotation on the gas storage accident corpus, and then output the risk factor set based on the BERT-BiGRU-CRF model. Then we enter the multi-feature fusion stage, extract the vector matrix through the BERT model and generate word vector features based on Word2Vec, and then obtain the fused event feature vector through the bidirectional BIGRU and Multi-Head Self-Attention layers, perform model training and test the optimal model, and output the gas storage association risk feature library.

[0140] This embodiment provides a method for training a relationship extraction model. The trained relationship extraction model can be used to execute a method for identifying the correlation relationship of risk factors of three-in-one operation and maintenance of a gas storage facility in the above embodiment. The technical effects are similar and will not be described in detail in this embodiment.

[0141] The embodiment of the present application can divide the functional modules of the electronic device or the main control device according to the above method example. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0142] Figure 6 A schematic diagram of a gas storage three-in-one operation and maintenance risk factor correlation identification device provided in an embodiment of the present application. Figure 6 As shown, the gas storage three-in-one operation and maintenance risk factor correlation relationship identification device includes: an acquisition module 310, a construction module 320, a vector module 330, a feature module 340 and a prediction module 350.

[0143] An acquisition module 310 is used to acquire a target text set related to the operation and maintenance risk of a target gas storage, and identify multiple target operation and maintenance risk events of the target gas storage from the target text set;

[0144] A construction module 320 is used to input a plurality of target operation and maintenance risk events into a construction unit to construct a plurality of target event pairs, wherein the target event pairs include: any two target operation and maintenance risk events;

[0145] A vector module 330, configured to generate a first vector for each target event pair through a vector generation unit;

[0146] A feature module 340 is configured to extract a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fuse the first local feature vector and the first overall feature vector to obtain a target fused vector of the target event pair;

[0147] The prediction module 350 is used to input the target fusion vector of each target event pair into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

[0148] Optionally, the feature module 340 is specifically used to:

[0149] A first local feature vector is extracted from the first vector through a first bidirectional gated recurrent unit; and a first overall feature vector is extracted from the first vector through a multi-head self-attention mechanism.

[0150] Optionally, the vector module 330 is specifically used for:

[0151] The second vector of each target event pair is generated through the pre-trained language model. The third vector of each target event pair is generated through the word vector model. For each target event pair, the second vector and the third vector of the target event pair are fused into the first vector of the target event pair.

[0152] Optionally, the acquisition module 310 is specifically configured to:

[0153] Each target text in the target text set is converted into a target word vector through the conversion unit; the feature of each target word vector is extracted through the second bidirectional gated recurrent unit to obtain a target feature vector; the target feature vector of each target text is input into the recognition unit to identify the target operation and maintenance risk events existing in each target text.

[0154] Optionally, the acquisition module 310 is specifically configured to:

[0155] A sample text set related to the operation and maintenance risks of a sample gas storage facility is obtained, wherein the sample text set includes multiple sample texts and an event label for each sample text, wherein the event label is used to indicate that the sample text is a risk factor, an event type, or a non-entity object; each sample text is converted into a sample word vector; features are extracted from each word vector to obtain a sample feature vector; and a named entity recognition model is trained using the sample feature vectors and event labels corresponding to each of the multiple sample texts.

[0156] The present embodiment provides a device for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage reservoir, which can execute a method for identifying the correlation relationship between the three-in-one operation and maintenance risk factors of a gas storage reservoir in the above-mentioned embodiment. The implementation principle and technical effect are similar, and will not be described in detail in the present embodiment.

[0157] In the specific implementation of the aforementioned device for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir, each module can be implemented as a processor, and the processor can execute computer execution instructions stored in the memory, so that the processor executes the aforementioned method for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir.

[0158] Figure 7 This is a schematic diagram of the structure of the electronic device hardware provided in the embodiment of the present application. Figure 7 As shown, the electronic device includes: at least one processor 410 and a memory 420. The electronic device also includes a communication component 430. The processor 410, the memory 420 and the communication component 430 are connected via a bus 440.

[0159] In the specific implementation process, at least one processor 410 executes the computer execution instructions stored in the memory 420, so that at least one processor 410 executes a method for identifying the correlation relationship of the three-in-one operation and maintenance risk factors of a gas storage reservoir as executed by the electronic device side as above.

[0160] The specific implementation process of the processor 410 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0161] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The steps of the method disclosed in the invention may be directly embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0162] The memory may include a high-speed random-access memory (RAM) and may also include a non-volatile memory (NVM), such as at least one disk memory.

[0163] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0164] The above functions implemented by the electronic device and the main control device introduce the scheme provided by the embodiment of the present application. It is understandable that in order to implement the above functions, the electronic device or the main control device includes a hardware structure and / or software module corresponding to each function. In combination with the units and algorithm steps of each example described in the embodiment disclosed in the embodiment of the present application, the embodiment of the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the technical solution of the embodiment of the present application.

[0165] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, it is used to implement the above-mentioned method for identifying the correlation relationship of risk factors of the three-in-one operation and maintenance of a gas storage facility.

[0166] The above-mentioned readable storage medium can be any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0167] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium 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 be located in a dedicated integrated circuit. Of course, the processor and the readable storage medium can also exist as discrete components in an electronic device or a main control device.

[0168] The present application also provides a computer program product, which includes: a computer program, which is stored in a readable storage medium. At least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0169] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0170] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments, and the above embodiments are only used to illustrate the technical solution of the present application rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying the correlation relationship of three-in-one operation and maintenance risk factors of a gas storage facility, characterized in that: include: Acquire a target text set related to operation and maintenance risks of a target gas storage, and identify a plurality of target operation and maintenance risk events of the target gas storage from the target text set; Inputting a plurality of the target operation and maintenance risk events into a construction unit to construct a plurality of target event pairs, wherein the target event pairs include: any two of the target operation and maintenance risk events; Generate a first vector for each target event pair by a vector generating unit; Extracting a first local feature vector and a first overall feature vector from the first vector by a feature extraction unit, and fusing the first local feature vector and the first overall feature vector to obtain a target fusion vector of the target event pair; The target fusion vector of each target event pair is input into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

2. The method according to claim 1, characterized in that The feature extraction unit includes: a first bidirectional gated recurrent unit and a multi-head self-attention mechanism, and the extracting of a first local feature vector and a first overall feature vector from the first vector by the feature extraction unit includes: extracting a first local feature vector from the first vector by using the first bidirectional gated recurrent unit; A first overall feature vector is extracted from the first vector through the multi-head self-attention mechanism.

3. The method according to claim 1, characterized in that The vector generation unit includes: a pre-trained language model and a word vector model, and the first vector of each target event pair is generated by the vector generation unit, including: Generate a second vector for each target event pair using the pre-trained language model; Generate a third vector for each target event pair through the word vector model; For each of the target event pairs, the second vector and the third vector of the target event pair are fused into a first vector of the target event pair.

4. The method according to any one of claims 1 to 3, characterized in that A pre-trained named entity recognition model is pre-stored, and the named entity recognition model includes: a conversion unit, a second bidirectional gated loop unit and a recognition unit. The identifying of multiple target operation and maintenance risk events of the target gas storage from the target text set includes: Convert each target text in the target text set into a target word vector by the conversion unit; Performing feature extraction on each of the target word vectors through the second bidirectional gated recurrent unit to obtain a target feature vector; The target feature vector of each target text is input into the recognition unit to identify the target operation and maintenance risk events existing in each target text.

5. The method according to claim 4, characterized in that The pre-training process of the named entity recognition model includes: Acquire a sample text set related to operation and maintenance risks of a sample gas storage, wherein the sample text set includes a plurality of sample texts and an event label of each sample text, wherein the event label is used to indicate that the sample text is a risk factor, an event type, or a non-entity object; Convert each of the sample texts into a sample word vector; Performing feature extraction on each of the word vectors to obtain a sample feature vector; The named entity recognition model is trained by using the sample feature vectors and the event labels corresponding to each of the plurality of sample texts.

6. A training method for a relation extraction model, characterized in that: The relationship extraction model includes a construction unit, a vector generation unit and a feature extraction unit, and the method includes: Acquire a sample text set related to operation and maintenance risks of a sample gas storage, and identify a plurality of sample operation and maintenance risk events of the sample gas storage from the sample text set; Inputting a plurality of the sample operation and maintenance risk events into the construction unit to construct at least one event triplet, wherein the event triplet includes: a sample event pair consisting of any two of the sample operation and maintenance risk events, and a relationship label between the sample event pairs; Generate, by means of the vector generating unit, a fourth vector of each of the sample event pairs and a label vector of each of the relationship labels; Extracting a second local feature vector and a second overall feature vector from the fourth vector by the feature extraction unit, and fusing the second local feature vector and the second overall feature vector to obtain a sample fusion vector; The relationship extraction model is trained by using the sample fusion vectors and the label vectors corresponding to a plurality of the event triplets.

7. A gas storage three-in-one operation and maintenance risk factor correlation identification device, characterized in that: include: An acquisition module, used to acquire a target text set related to the operation and maintenance risk of a target gas storage, and identify a plurality of target operation and maintenance risk events of the target gas storage from the target text set; A construction module, used for inputting a plurality of the target operation and maintenance risk events into a construction unit to construct a plurality of target event pairs, wherein the target event pairs include: any two of the target operation and maintenance risk events; A vector module, used for generating a first vector of each target event pair through a vector generating unit; A feature module, configured to extract a first local feature vector and a first overall feature vector from the first vector through a feature extraction unit, and fuse the first local feature vector and the first overall feature vector to obtain a target fused vector of the target event pair; The prediction module is used to input the target fusion vector of each target event pair into the prediction unit to obtain the correlation relationship between the two target operation and maintenance risk events in each target event pair.

8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; When the processor executes the computer-executable instructions stored in the memory, the instructions are used to implement the method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images