Root cause analysis methods, systems and storage media for defects in oil and gas exploration equipment

Abstract

This invention provides a method, system, and storage medium for root cause analysis of defects in oil and gas exploration equipment. The method includes: performing reasoning analysis on root cause samples and defect description samples to obtain reasoning thought chain samples; performing thought chain filtering and logical filtering on the reasoning thought chain samples to obtain filtered thought chains; training an equipment defect root cause analysis model based on the filtered thought chains, root cause samples, and defect description samples; inputting the defect description to be analyzed into the converged equipment defect root cause analysis model for cause analysis to obtain the target defect root cause and the defect reasoning thought chain of the target defect root cause. In this embodiment, based on the converged equipment defect root cause analysis model, the root cause analysis of equipment defects on the defect description to be analyzed can be effectively performed, improving the efficiency of root cause analysis of defects in oil and gas exploration equipment.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method, system and storage medium for analyzing the root causes of defects in oil and gas exploration equipment. Background Technology

[0002] In the field of oil and gas exploration, the trouble-free operation of exploration equipment such as core components of drilling platforms and downhole logging equipment is the core prerequisite for ensuring the continuous progress of exploration operations and controlling operating costs. Accurately locating the root cause of equipment defects is the core support for achieving preventive maintenance of equipment.

[0003] The root cause analysis of existing oil and gas exploration equipment defects generally involves manually disassembling the equipment to trace the fault source, resulting in low efficiency in the analysis of the root cause of equipment defects. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and storage medium for root cause analysis of defects in oil and gas exploration equipment, so as to solve the problem of low efficiency in root cause analysis of equipment defects in the prior art.

[0005] This invention is implemented as follows: a method for analyzing the root causes of defects in oil and gas exploration equipment, the method comprising:

[0006] Obtain root cause samples and defect description samples, and perform reasoning analysis on the root cause samples and defect description samples to obtain reasoning chain samples;

[0007] The reasoning thought chain sample is subjected to text analysis to obtain text analysis results, and the reasoning thought chain sample is screened based on the text analysis results to obtain candidate thought chains.

[0008] The candidate thought chains are logically tested to obtain logical test results, and the candidate thought chains are logically filtered based on the logical test results to obtain filtered thought chains.

[0009] The screening thought chain, the root cause samples, and the defect description samples are combined to obtain model training samples. The equipment defect root cause analysis model is then trained based on the model training samples until the equipment defect root cause analysis converges.

[0010] Obtain the description of the defects to be analyzed in the target oil and gas exploration equipment, and input the description of the defects to be analyzed into the converged root cause analysis model of the equipment defects for cause analysis, so as to obtain the root cause of the target defects and the defect reasoning chain of the root cause of the target defects.

[0011] Preferably, the root cause sample and the defect description sample are subjected to reasoning analysis to obtain a reasoning chain sample, including:

[0012] Entity analysis is performed on the defect description sample to obtain entity analysis results, and key entities are determined based on the entity analysis results;

[0013] A task tree is generated based on the key entities, and the entity content of the key entities in the defect description sample is obtained.

[0014] The task tree is used to call the intelligent agent to perform reasoning analysis on the entity content, obtain the entity thought chain, and combine the entity thought chain with the root cause sample to obtain the reasoning thought chain sample.

[0015] Preferably, text analysis is performed on the reasoning chain sample to obtain text analysis results, including:

[0016] Obtain a list of core keywords, and perform word matching between the reasoning chain sample and the list of core keywords to obtain the core keywords;

[0017] The reasoning chain sample is segmented into words to obtain sample words, and the sample words are then identified by type to obtain entity types.

[0018] Obtain the length of the reasoning thought chain sample, and combine the length of the thought chain, the entity type, and the number of keywords of the core keywords to obtain the text analysis result.

[0019] Preferably, the reasoning thought chain samples are screened based on the text analysis results to obtain candidate thought chains, including:

[0020] If the number of keywords in any of the reasoning thought chain samples is less than the number threshold, then the reasoning thought chain sample is deleted.

[0021] If the entity type in any of the reasoning thought chain samples does not match the preset type, then the reasoning thought chain sample is deleted.

[0022] If the length of the reasoning chain of any of the reasoning chain samples is not within the preset length range, then the reasoning chain sample is deleted.

[0023] The remaining inference thought chain samples are set as the candidate thought chains.

[0024] Preferably, the candidate thought chain is subjected to logical detection to obtain the logical detection result, including:

[0025] Entity extraction is performed on the candidate thought chains to obtain thought chain entities, and entity relationships between different thought chain entities are obtained.

[0026] Different thought chain entities and corresponding entity relationships are combined to obtain thought chain triples, and a thought chain knowledge graph is constructed based on the thought chain triples.

[0027] The thought chain knowledge graph is disambiguated, and the disambiguated thought chain knowledge graph is then fused to obtain a fused knowledge graph.

[0028] The entity relations in the thought chain triplet are replaced with inverse relations to obtain inverse relation triplets. The inverse relation triplets and the thought chain triplets are then used with the fused knowledge graph to generate content, resulting in inverse relation generated text and thought chain generated text.

[0029] Logical detection is performed on the text generated by the reverse relationship and the text generated by the thought chain, respectively, to obtain the logical detection results.

[0030] Preferably, the candidate thought chains are logically filtered based on the logical detection results to obtain filtered thought chains, including:

[0031] If any of the reverse relation generated texts in the logical detection results is logically correct, then the thought chain triple corresponding to the reverse relation generated text is determined as an incorrect triple.

[0032] Delete the candidate thought chain corresponding to the erroneous triple, and set the remaining candidate thought chains as the filtering thought chain.

[0033] Preferably, training the root cause analysis model for equipment defects based on the model training samples includes:

[0034] The defect description sample is input into the equipment defect root cause analysis model for cause analysis, and a predictive reasoning chain for predicting the root cause of the defect is obtained.

[0035] The first model loss is calculated based on the root cause of the predicted defect and the root cause sample, and the second model loss is calculated based on the predictive reasoning chain and the screening chain.

[0036] The total model loss is determined based on the first model loss and the second model loss, and the parameters of the equipment defect root cause analysis model are updated based on the total model loss.

[0037] Another objective of this invention is to provide a root cause analysis system for defects in oil and gas exploration equipment, the system comprising:

[0038] The reasoning analysis module is used to obtain root cause samples and defect description samples, and to perform reasoning analysis on the root cause samples and defect description samples to obtain reasoning thought chain samples.

[0039] The thought chain filtering module is used to perform text analysis on the reasoning thought chain sample, obtain text analysis results, and filter the reasoning thought chain sample according to the text analysis results to obtain candidate thought chains.

[0040] The logic filtering module is used to perform logic detection on the candidate thought chains, obtain logic detection results, and perform logic filtering on the candidate thought chains based on the logic detection results to obtain filtered thought chains.

[0041] The model training module is used to combine the screening thought chain, the root cause samples, and the defect description samples to obtain model training samples, and to train the equipment defect root cause analysis model based on the model training samples until the equipment defect root cause analysis converges.

[0042] The cause analysis module is used to obtain the description of the defects to be analyzed in the target oil and gas exploration equipment, and input the description of the defects to be analyzed into the converged root cause analysis model of the equipment defects for cause analysis, so as to obtain the root cause of the target defect and the defect reasoning chain of the root cause of the target defect.

[0043] Preferably, the reasoning analysis module is further used for:

[0044] Entity analysis is performed on the defect description sample to obtain entity analysis results, and key entities are determined based on the entity analysis results;

[0045] A task tree is generated based on the key entities, and the entity content of the key entities in the defect description sample is obtained.

[0046] The task tree is used to call the intelligent agent to perform reasoning analysis on the entity content, obtain the entity thought chain, and combine the entity thought chain with the root cause sample to obtain the reasoning thought chain sample.

[0047] In this embodiment of the invention, by performing reasoning analysis on root cause samples and defect description samples, reasoning thought chain samples can be effectively generated. By performing text analysis on the reasoning thought chain samples, reasoning thought chain screening can be effectively performed, improving the accuracy of candidate thought chains. By performing logical detection on the candidate thought chains, logical screening can be effectively performed, further improving the accuracy of the screened thought chains. By combining the screened thought chains, root cause samples, and defect description samples, model training samples can be effectively generated. Based on the model training samples, the equipment defect root cause analysis model is trained, so that the trained equipment defect root cause analysis model can effectively perform equipment defect root cause analysis on the defect description to be analyzed, automatically generating the target defect root cause and the defect reasoning thought chain of the target defect root cause, eliminating the need for manual decomposition for defect root cause analysis, and improving the efficiency of oil and gas exploration equipment defect root cause analysis. Attached Figure Description

[0048] Figure 1 This is a flowchart of the root cause analysis method for oil and gas exploration equipment defects provided in the first embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the root cause analysis method for oil and gas exploration equipment defects provided in the first embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of the structure of the root cause analysis system for oil and gas exploration equipment defects provided in the second embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of the structure of the terminal device provided in the third embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] To illustrate the technical solution described in this invention, specific embodiments are described below.

[0054] Example 1

[0055] Please see Figure 1 This is a flowchart of the root cause analysis method for oil and gas exploration equipment defects provided in the first embodiment of the present invention. This root cause analysis method for oil and gas exploration equipment defects can be applied to any equipment or system. The root cause analysis method for oil and gas exploration equipment defects includes the following steps:

[0056] Step S10: Obtain root cause samples and defect description samples, and perform reasoning analysis on the root cause samples and defect description samples to obtain reasoning chain samples.

[0057] This process involves obtaining historical defect descriptions and their corresponding root causes, performing data preprocessing on these descriptions to obtain root cause samples and defect description samples, and then using a pre-trained large-scale language model to perform reasoning analysis on the root cause samples and defect description samples to obtain reasoning chain samples.

[0058] Optionally, for pre-trained large-scale language models, a prompt word template is set. The prompt includes "role definition," "task context," and "inference constraints." The role definition can be set to "You are an equipment safety expert with experience in offshore oil and gas exploration, and your task is to analyze the root cause of equipment defects." The task context can include "known defect description and known root cause category." The inference constraints can be set as follows:

[0059] Please proceed with step-by-step reasoning strictly according to the knowledge in the following four areas:

[0060] Seismic exploration: Consider the propagation characteristics of seismic waves (velocity, attenuation), stratigraphic structure (such as the distribution and morphology of salt domes, faults, and folds), formation pressure system, reservoir properties (porosity, permeability) prediction results, and the impact of near-surface complex conditions (such as shallow gas and water layers) on equipment vibration, load, positioning accuracy, and signal interference.

[0061] Downhole logging: Considering the wellbore environment (high temperature, high pressure, highly corrosive fluids, etc.) , The following factors are considered: brine, formation rock mechanical properties (strength, drillability), wellbore trajectory (dogleg, inclination), downhole fluid properties (density, viscosity, sand content), and compatibility and failure risk of logging instruments with harsh environments (high temperature, high pressure, strong magnetic field, radioactivity).

[0062] Core analysis: Considering the impact of rock mineral composition (such as the influence of clay mineral type and content on water sensitivity and acid sensitivity), microstructural characteristics (pore structure, fracture development), rock mechanical parameters (compressive / tensile strength, elastic modulus, Poisson's ratio), and formation sensitivity (stress sensitivity, rate sensitivity, water sensitivity, salt sensitivity, acid sensitivity) on drilling tool wear, completion string stability, fracturing effect, and long-term production.

[0063] Geological modeling: Consider the uncertainties of three-dimensional geological structure (fault sealing, reservoir heterogeneity, interlayer distribution), fluid distribution characteristics (oil-water / gas-water interface, saturation), the direction and magnitude of the geostress field, and the dynamic changes of reservoir (pressure exhaustion, compaction and settlement) on drilling engineering design, well location deployment, production string stress, and the stability of surface facilities (such as platforms and pipelines).

[0064] The output requires the instruction: "The final reasoning basis must be a clear thought process," in order to obtain a sample of the reasoning thought chain.

[0065] Furthermore, reasoning analysis is performed on the root cause sample and the defect description sample to obtain a reasoning chain sample, including:

[0066] Entity analysis is performed on the defect description sample to obtain entity analysis results, and key entities are determined based on the entity analysis results. Specifically, the entity types in the entity analysis results are matched with specified types, and the entity objects corresponding to the successfully matched entity types are determined as key entities. The specified types can be set according to requirements, for example, the specified types can be set as equipment type, location type, phenomenon type, etc.

[0067] A task tree is generated based on the key entities, and the entity content of the key entities in the defect description sample is obtained; wherein, the task tree is used to call the corresponding agent. For example, the task tree involving the geological structure calls the geological modeling agent, the task tree involving vibration / stress calls the rock mechanics agent, the task tree involving fluid corrosion calls the corrosion chemistry agent, and the task tree involving equipment failure calls the equipment engineering agent.

[0068] The task tree is used to call the intelligent agent to perform reasoning analysis on the entity content to obtain the entity thought chain, and the entity thought chain and the root cause sample are combined to obtain the reasoning thought chain sample.

[0069] For example, the defect description is: "Pressure fluctuation amplitude of drilling pump on South China Sea platform ±15%, frequency 2.2Hz", and the root cause category is: "Geological environment related failure".

[0070] Agent analysis:

[0071] "Equipment type", "Drilling pump";

[0072] Location, South China Sea salt gypsum layer block;

[0073] "Phenomenon", ["Pressure fluctuation", "Frequency 2.2Hz"];

[0074] Generate a task tree:

[0075] Geological modeling agent (analyzes the distribution of salt and gypsum layers), rock mechanics agent (calculates the source of vibration frequencies), and equipment engineering agent (verifies the vibration resistance of seals).

[0076] Multi-agent cooperative response:

[0077] For the geological modeling agent, input "block coordinates (these block coordinates also need to be generated by the main controller)", call the Petrel model, and output "salt-gypsum layer thickness 62m (pressure sensitivity coefficient 0.38)";

[0078] For the Rock Mechanics Agent, input "salt-gypsum layer thickness 62m + pump pressure curve → ABAQUS calculation of creep wave", output "resonance frequency 2.3Hz (matching actual measurement 2.2Hz, error 4.5%)";

[0079] For the Equipment Engineering Agent, input "Vibration frequency 2.3Hz, query API 6A specification", and output "Sealing ring resonance failure risk level II";

[0080] Combine the output content:

[0081] Geological modeling: The thickness of the salt-gypsum layer in Block B of the South China Sea reaches 62m (exceeding the safety threshold of 50m) → high creep risk;

[0082] Rock mechanics: Creep stress wave frequency 2.3Hz → resonates with the drilling pump's natural frequency of 2.2Hz;

[0083] Equipment Engineering: Resonance caused API 6A seal failure;

[0084] The root cause of the equipment defect: salt paste layer creep - equipment resonance coupling failure.

[0085] Step S20: Perform text analysis on the reasoning thought chain sample to obtain the text analysis results, and filter the reasoning thought chain sample according to the text analysis results to obtain candidate thought chains.

[0086] Specifically, text analysis is performed on the reasoning thought chain samples to obtain the thought chain length, entity type, and number of core keywords. Based on the thought chain length, entity type, and number of core keywords, the reasoning thought chain samples are screened to improve the accuracy of candidate thought chains. By using the thought chain length, entity type, and number of core keywords, obviously unqualified reasoning thought chain samples can be quickly eliminated.

[0087] Optionally, text analysis is performed on the reasoning chain sample to obtain text analysis results, including:

[0088] Obtain a core keyword list and perform word matching between the reasoning chain sample and the core keyword list to obtain core keywords. The keywords in the core keyword list can be set according to requirements. In this step, geological environment keyword matching can be performed, defining a set of core keywords reflecting four aspects of marine exploration and oil and gas production: seismic exploration, downhole logging, core analysis, and geological modeling. For example: ["seismic wave", "logging curve", "core porosity", "geological model", "formation pressure", "salt-gypsum layer", "reservoir properties", "well trajectory", "drilling fluid", "corrosion rate"], etc. For each reasoning chain sample, count the number of times it contains the above keywords to obtain the keyword count.

[0089] The reasoning chain sample is segmented into words to obtain sample words, and the sample words are then type-identified to obtain entity types; wherein, entity types are used to represent the semantic types of the corresponding words;

[0090] Obtain the length of the reasoning thought chain sample, and combine the length of the thought chain, the entity type, and the number of keywords of the core keywords to obtain the text analysis result.

[0091] Furthermore, based on the text analysis results, the reasoning thought chain samples are screened to obtain candidate thought chains, including:

[0092] If the number of keywords in any of the reasoning thought chain samples is less than the number threshold, then the reasoning thought chain sample is deleted; wherein, the number threshold can be set according to requirements;

[0093] If the entity type in any of the reasoning thought chain samples does not match the preset type, the reasoning thought chain sample is deleted. In this process, matching the entity type with the preset type achieves the effect of basic logical coherence check, which checks whether the reasoning thought chain sample contains obvious logical connectors (such as "therefore", "because", "leading to", "therefore", etc.) or whether its expression presents a clear derivation process of "defect description - geological environment influence - root cause", so as to eliminate processes that are extremely vague, structurally chaotic or obviously irrelevant.

[0094] If the length of any reasoning thought chain sample is not within the preset length range, the reasoning thought chain sample is deleted. The preset length range can be set according to the requirements. By comparing the length of the reasoning thought chain sample with the preset length range, the effect of length rationality screening is achieved, and reasoning thought chain samples that are too short (insufficient information) or too long (may contain irrelevant information or redundancy) are eliminated.

[0095] The remaining reasoning thought chain samples are set as candidate thought chains; wherein, an optimal quantity parameter can be set. If the number of chains that pass the screening is greater than the optimal quantity parameter, then thought chains with the optimal quantity parameter are set as candidate thought chains according to the rule score (e.g., the higher the number of keywords, the higher the score) or randomly selected.

[0096] Step S30: Perform logical detection on the candidate thought chain to obtain the logical detection result, and perform logical screening on the candidate thought chain according to the logical detection result to obtain the screened thought chain;

[0097] In this process, logical checks are performed on the candidate thought chains to detect whether there are any logical errors in each candidate thought chain. Based on the results of the logical checks, candidate thought chains with logical errors are deleted, thereby achieving the effect of logical screening of candidate thought chains.

[0098] Optionally, logical detection is performed on the candidate thought chain to obtain the logical detection result, including:

[0099] Entity extraction is performed on the candidate thought chains to obtain thought chain entities, and entity relationships between different thought chain entities are obtained; wherein, entity extraction is performed on the candidate thought chains (e.g., equipment components "drill bit" "casing"; geological concepts "salt gypsum layer" "pore pressure"); environmental factors "high temperature" "high pressure", "high pressure", etc.). "Corrosion" extracts the relationships between entities in the thought chain, resulting in entity relationships such as: "cause", "because", "accelerated", "restricted by", "not applicable to".

[0100] Different thought chain entities and their corresponding entity relationships are combined to obtain thought chain triples, and a thought chain knowledge graph is constructed based on the thought chain triples; wherein, two thought chain entities and their corresponding entity relationships are combined to form a triple, for example (high temperature sealing, resulting in rubber softening and failure), thought chain triples are extracted for each candidate thought chain, and a thought chain knowledge graph is constructed based on the extracted thought chain triples;

[0101] The thought chain knowledge graph is disambiguated, and then the disambiguated thought chain knowledge graphs are merged to obtain a merged knowledge graph. Specifically, entity disambiguation of the thought chain knowledge graphs aligns different entities with the same meaning across different thought chain knowledge graphs. For example, all proper nouns in the oil exploration field are collected, and a proper noun dictionary is created. Each entity in the thought chain knowledge graph is compared with the proper noun dictionary to determine if there are words with the same meaning. If so, the entity is replaced with the standard word from the proper noun dictionary, thus achieving entity disambiguation of the thought chain knowledge graph. By merging the disambiguated thought chain knowledge graphs, an entity relationship network can be effectively formed. For example, after merging, a closed-loop entity relationship network is formed: "real-time parameters → fault symptoms → intermediate causes → root causes → maintenance records".

[0102] The entity relations in the thought chain triplet are replaced with inverse relations to obtain inverse relation triplets. The inverse relation triplets and the thought chain triplets are then used with the fused knowledge graph to generate content, resulting in inverse relation generated text and thought chain generated text.

[0103] Logical detection is performed on the text generated by the reverse relationship and the text generated by the thought chain, respectively, to obtain the logical detection results.

[0104] Further, the candidate thought chains are logically filtered based on the logical detection results to obtain the filtered thought chains, including: if the detection result of any of the reverse relation generated texts in the logical detection results is logically correct, then the thought chain triple corresponding to the reverse relation generated text is determined as an incorrect triple; the candidate thought chain corresponding to the incorrect triple is deleted, and the remaining candidate thought chains are set as the filtered thought chains.

[0105] Step S40: Combine the screening thought chain, the root cause sample, and the defect description sample to obtain the model training sample, and train the equipment defect root cause analysis model according to the model training sample until the equipment defect root cause analysis converges.

[0106] The model training samples cover typical scenarios such as seismic exploration equipment, downhole logging instruments, and pipeline corrosion. In this step, a domain knowledge enhancement architecture can be introduced on the basis of the general large language model LLM, using the CodeLLaMA-7B model as the base model. Its structured logical reasoning ability is superior to that of general text generation models. A multi-objective joint loss function is designed to enhance domain adaptation, using standard cross-entropy loss. On the basis of the basic loss function, a domain keyword coverage penalty term is added to ensure that the output contains necessary geological terms (such as "salt-gypsum layer creep"), "..." In terms of hyperparameter optimization strategy, the optimal configuration was determined through directional parameter space search, with length_reward set to 1.1. The essence of length_reward is a guiding signal used by the Large Language Model (LLM) to dynamically adjust the length of the output content when generating text. length_reward < 1.0 penalizes excessively long generated results, forcing the model to output more concisely (potentially losing key details); length_reward = 1.0 shows no length bias (the model has free rein); length_reward > 1.0 (e.g., 1.1) rewards longer generated results, encouraging the model to supplement details (such as descriptions of geological environment associations). Different values ​​were tested in the range [0.8, 1.8] with a step size of 0.1 (11 experiments in total), using text key information coverage. The results showed that length_reward = 1.1 was the best value.

[0107] Optionally, the root cause analysis model for equipment defects is trained based on the model training samples, including:

[0108] The defect description sample is input into the equipment defect root cause analysis model for cause analysis, and a predictive reasoning chain for predicting the root cause of the defect is obtained.

[0109] The first model loss is calculated based on the root cause of the predicted defect and the root cause sample, and the second model loss is calculated based on the predictive reasoning chain and the screening chain.

[0110] The total model loss is determined based on the first model loss and the second model loss, and the parameters of the equipment defect root cause analysis model are updated based on the total model loss.

[0111] Step S50: Obtain the description of the defects to be analyzed of the target oil and gas exploration equipment, and input the description of the defects to be analyzed into the converged root cause analysis model of the equipment defects for cause analysis, so as to obtain the root cause of the target defect and the defect reasoning chain of the root cause of the target defect;

[0112] The process involves logically testing the root cause of the target defect and the reasoning chain of the defect. If the logical test of the root cause of the target defect and the reasoning chain of the defect is qualified, the root cause of the target defect and the reasoning chain of the defect are added to the model training samples to expand the model training samples, thereby realizing the continuous iteration and optimization of the model.

[0113] Please see Figure 2When the description of the defect to be analyzed is "the water injection pump was shut down for maintenance and it was found that the corresponding temperature could not be measured correctly. If the temperature is abnormal, it may cause the water injection pump to shut down abnormally, which will impact the production process", the description of the defect to be analyzed is input into the converged equipment defect root cause analysis model for cause analysis, so as to output the root cause of the target defect "the underground pressure of the drilling platform is too high" and the defect reasoning chain "1. The rock and soil geology is relatively hard; 2. The oil and gas exploration depth is deep, resulting in excessive pressure; 3. The drilling platform is at the bottom, and the pressure measuring device is prone to failure; 4. Therefore, the excessive underground pressure is the root cause".

[0114] In this embodiment, by performing reasoning analysis on root cause samples and defect description samples, reasoning thought chain samples can be effectively generated. By performing text analysis on the reasoning thought chain samples, the reasoning thought chain samples can be effectively screened, improving the accuracy of candidate thought chains. By performing logical detection on the candidate thought chains, the candidate thought chains can be effectively logically screened, further improving the accuracy of the screened thought chains. By combining the screened thought chains, root cause samples, and defect description samples, model training samples can be effectively generated. Based on the model training samples, the equipment defect root cause analysis model can be trained, so that the trained equipment defect root cause analysis model can effectively perform equipment defect root cause analysis on the defect description to be analyzed, automatically generating the target defect root cause and the defect reasoning thought chain of the target defect root cause, eliminating the need for manual decomposition for defect root cause analysis, and improving the efficiency of oil and gas exploration equipment defect root cause analysis.

[0115] Example 2

[0116] Please see Figure 3 This is a schematic diagram of the structure of the oil and gas exploration equipment defect root cause analysis system 100 provided in the second embodiment of the present invention, including:

[0117] The reasoning analysis module 10 is used to obtain root cause samples and defect description samples, and to perform reasoning analysis on the root cause samples and defect description samples to obtain reasoning thought chain samples.

[0118] Optionally, the reasoning and analysis module 10 is further configured to: perform entity analysis on the defect description sample, obtain entity analysis results, and determine key entities based on the entity analysis results;

[0119] A task tree is generated based on the key entities, and the entity content of the key entities in the defect description sample is obtained.

[0120] The task tree is used to call the intelligent agent to perform reasoning analysis on the entity content, obtain the entity thought chain, and combine the entity thought chain with the root cause sample to obtain the reasoning thought chain sample.

[0121] The thought chain filtering module 11 is used to perform text analysis on the reasoning thought chain sample, obtain the text analysis results, and filter the reasoning thought chain sample according to the text analysis results to obtain candidate thought chains.

[0122] Optionally, the thinking chain filtering module 11 is also used to: obtain a core keyword list and perform word matching between the reasoning thinking chain sample and the core keyword list to obtain core keywords;

[0123] The reasoning chain sample is segmented into words to obtain sample words, and the sample words are then identified by type to obtain entity types.

[0124] Obtain the length of the reasoning thought chain sample, and combine the length of the thought chain, the entity type, and the number of keywords of the core keywords to obtain the text analysis result.

[0125] Furthermore, the thinking chain filtering module 11 is also used to: delete the thinking chain sample if the number of keywords in any of the thinking chain samples is less than the number threshold.

[0126] If the entity type in any of the reasoning thought chain samples does not match the preset type, then the reasoning thought chain sample is deleted.

[0127] If the length of the reasoning chain of any of the reasoning chain samples is not within the preset length range, then the reasoning chain sample is deleted.

[0128] The remaining inference thought chain samples are set as the candidate thought chains.

[0129] The logic filtering module 12 is used to perform logic detection on the candidate thought chain, obtain the logic detection result, and perform logic filtering on the candidate thought chain according to the logic detection result to obtain the filtered thought chain.

[0130] Optionally, the logic filtering module 12 is further configured to: extract entities from the candidate thought chains to obtain thought chain entities, and obtain entity relationships between different thought chain entities;

[0131] Different thought chain entities and corresponding entity relationships are combined to obtain thought chain triples, and a thought chain knowledge graph is constructed based on the thought chain triples.

[0132] The thought chain knowledge graph is disambiguated, and the disambiguated thought chain knowledge graph is then fused to obtain a fused knowledge graph.

[0133] The entity relations in the thought chain triplet are replaced with inverse relations to obtain inverse relation triplets. The inverse relation triplets and the thought chain triplets are then used with the fused knowledge graph to generate content, resulting in inverse relation generated text and thought chain generated text.

[0134] Logical detection is performed on the text generated by the reverse relationship and the text generated by the thought chain, respectively, to obtain the logical detection results.

[0135] Furthermore, the logic filtering module 12 is also used to: if the detection result of any of the reverse relation generated texts in the logic detection results is logically correct, then the thought chain triple corresponding to the reverse relation generated text is determined as an incorrect triple;

[0136] Delete the candidate thought chain corresponding to the erroneous triple, and set the remaining candidate thought chains as the filtering thought chain.

[0137] The model training module 13 is used to combine the screening thought chain, the root cause sample, and the defect description sample to obtain the model training sample, and to train the equipment defect root cause analysis model based on the model training sample until the equipment defect root cause analysis converges.

[0138] Optionally, the model training module 13 is also used to: input the defect description sample into the equipment defect root cause analysis model for cause analysis, and obtain a predictive reasoning chain for predicting the root cause of the defect;

[0139] The first model loss is calculated based on the root cause of the predicted defect and the root cause sample, and the second model loss is calculated based on the predictive reasoning chain and the screening chain.

[0140] The total model loss is determined based on the first model loss and the second model loss, and the parameters of the equipment defect root cause analysis model are updated based on the total model loss.

[0141] The cause analysis module 14 is used to obtain the description of the defects to be analyzed of the target oil and gas exploration equipment, and input the description of the defects to be analyzed into the converged root cause analysis model of the equipment defects for cause analysis, so as to obtain the root cause of the target defect and the defect reasoning chain of the root cause of the target defect.

[0142] In this embodiment, by performing reasoning analysis on root cause samples and defect description samples, reasoning thought chain samples can be effectively generated. By performing text analysis on the reasoning thought chain samples, the reasoning thought chain samples can be effectively screened, improving the accuracy of candidate thought chains. By performing logical detection on the candidate thought chains, the candidate thought chains can be effectively logically screened, further improving the accuracy of the screened thought chains. By combining the screened thought chains, root cause samples, and defect description samples, model training samples can be effectively generated. Based on the model training samples, the equipment defect root cause analysis model can be trained, so that the trained equipment defect root cause analysis model can effectively perform equipment defect root cause analysis on the defect description to be analyzed, automatically generating the target defect root cause and the defect reasoning thought chain of the target defect root cause, eliminating the need for manual decomposition for defect root cause analysis, and improving the efficiency of oil and gas exploration equipment defect root cause analysis.

[0143] Example 3

[0144] Figure 4 This is a structural block diagram of a terminal device 2 provided in the third embodiment of this application. For example... Figure 4 As shown, the terminal device 2 in this embodiment includes a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program for a root cause analysis method for defects in oil and gas exploration equipment. When the processor 20 executes the computer program 22, it implements the steps in each embodiment of the root cause analysis method for defects in oil and gas exploration equipment described above.

[0145] For example, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, the processor 20 and the memory 21.

[0146] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0147] The memory 21 can be an internal storage unit of the terminal device 2, such as a hard drive or memory of the terminal device 2. The memory 21 can also be an external storage device of the terminal device 2, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal device 2. Furthermore, the memory 21 can include both internal and external storage units of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 can also be used to temporarily store data that has been output or will be output.

[0148] Furthermore, the functional modules in the various embodiments of this application 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0149] If an integrated module 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. This computer-readable storage medium can be non-volatile or volatile. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the contents of a computer-readable storage medium may be appropriately added to or subtracted from the contents as required by the legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, a computer-readable storage medium may not include electrical carrier signals and telecommunication signals.

[0150] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method of oil and gas exploration equipment defect root cause analysis, characterized in that, The method comprises: obtaining a root cause sample and a defect description sample, and performing reasoning analysis on the root cause sample and the defect description sample to obtain a reasoning thought chain sample; performing text analysis on the reasoning thought chain sample to obtain a text analysis result, and performing thought chain screening on the reasoning thought chain sample according to the text analysis result to obtain a candidate thought chain; performing logic detection on the candidate thought chain to obtain a logic detection result, and performing logic screening on the candidate thought chain according to the logic detection result to obtain a screened thought chain; combining the screened thought chain, the root cause sample and the defect description sample to obtain a model training sample, and training a device defect root cause analysis model according to the model training sample until the device defect root cause analysis converges; obtaining a to-be-analyzed defect description of a target oil and gas exploration device, and inputting the to-be-analyzed defect description into the converged device defect root cause analysis model to perform cause analysis to obtain a to-be-analyzed defect root cause and a defect reasoning thought chain of the to-be-analyzed defect root cause; performing reasoning analysis on the root cause sample and the defect description sample to obtain a reasoning thought chain sample, comprising: performing entity analysis on the defect description sample to obtain an entity analysis result, and determining a key entity according to the entity analysis result; generating a task tree according to the key entity, and obtaining entity content of the key entity in the defect description sample; performing reasoning analysis on the entity content by calling an intelligent agent according to the task tree to obtain an entity thought chain, and combining the entity thought chain and the root cause sample to obtain the reasoning thought chain sample.

2. The oil and gas exploration equipment defect root cause analysis method of claim 1, wherein, performing text analysis on the reasoning thought chain sample to obtain a text analysis result, comprising: obtaining a core keyword list, and performing vocabulary matching on the reasoning thought chain sample and the core keyword list to obtain a core keyword; performing word segmentation on the reasoning thought chain sample to obtain sample word segmentation, and performing type identification on the sample word segmentation to obtain an entity type; obtaining a thought chain length of the reasoning thought chain sample, and combining the thought chain length, the entity type and a keyword quantity of the core keyword to obtain the text analysis result.

3. The oil and gas exploration equipment defect root cause analysis method of claim 2, wherein, performing thought chain screening on the reasoning thought chain sample according to the text analysis result to obtain a candidate thought chain, comprising: if the keyword quantity of any reasoning thought chain sample is less than a quantity threshold, deleting the reasoning thought chain sample; if the entity type in any reasoning thought chain sample does not match a preset type, deleting the reasoning thought chain sample; if the thought chain length of any reasoning thought chain sample is not within a preset length range, deleting the reasoning thought chain sample; setting the remaining reasoning thought chain samples as the candidate thought chain.

4. The oil and gas exploration equipment defect root cause analysis method of claim 1, wherein, performing logic detection on the candidate thought chain to obtain a logic detection result, comprising: performing entity extraction on the candidate thought chain to obtain thought chain entities, and obtaining entity relationships between different thought chain entities; combining different thought chain entities and corresponding entity relationships to obtain thought chain triples, and constructing a thought chain knowledge graph according to the thought chain triples; performing entity disambiguation on the thought chain knowledge graph, and fusing the thought chain knowledge graph after entity disambiguation to obtain a fused knowledge graph; performing inverse relationship replacement on the entity relationships in the thought chain triples to obtain inverse relationship triples, and performing content generation on the inverse relationship triples and the thought chain triples with the fused knowledge graph to obtain inverse relationship generated text and thought chain generated text; respectively performing logic detection on the inverse relationship generated text and the thought chain generated text to obtain a logic detection result.

5. The oil and gas exploration equipment defect root cause analysis method of claim 4, wherein, performing logic screening on the candidate thought chain according to the logic detection result to obtain a screened thought chain, including: if the detection result of any inverse relationship generated text in the logic detection result is logically correct, determining the thought chain triple corresponding to the inverse relationship generated text as an error triple; deleting the candidate thought chain corresponding to the error triple, and setting the remaining candidate thought chain as the screened thought chain.

6. The oil and gas exploration equipment defect root cause analysis method of claim 1, wherein, training a device defect root cause analysis model according to the model training sample, including: inputting the defect description sample into the device defect root cause analysis model to perform cause analysis, and obtaining a predicted defect root cause prediction reasoning thought chain; calculating a first model loss according to the predicted defect root cause and the root cause sample, and calculating a second model loss according to the predicted reasoning thought chain and the screened thought chain; determining a total model loss according to the first model loss and the second model loss, and updating parameters of the device defect root cause analysis model according to the total model loss.

7. An oil and gas exploration equipment defect root cause analysis system, characterized by, The system includes: a reasoning analysis module configured to obtain root cause samples and defect description samples, and perform reasoning analysis on the root cause samples and the defect description samples to obtain reasoning thought chain samples; a thought chain screening module configured to perform text analysis on the reasoning thought chain samples to obtain a text analysis result, and perform thought chain screening on the reasoning thought chain samples according to the text analysis result to obtain candidate thought chains; a logic screening module configured to perform logic detection on the candidate thought chains to obtain a logic detection result, and perform logic screening on the candidate thought chains according to the logic detection result to obtain screened thought chains; a model training module configured to combine the screened thought chains, the root cause samples and the defect description samples to obtain model training samples, and train a device defect root cause analysis model according to the model training samples until the device defect root cause analysis converges; a cause analysis module configured to obtain a to-be-analyzed defect description of a target oil and gas exploration device, and input the to-be-analyzed defect description into the converged device defect root cause analysis model to perform cause analysis, and obtain a to-be-analyzed defect root cause and a defect reasoning thought chain of the to-be-analyzed defect root cause; the reasoning analysis module is further configured to: performing entity analysis on the defect description sample to obtain an entity analysis result, and determining a key entity according to the entity analysis result; generating a task tree according to the key entity, and obtaining entity content of the key entity in the defect description sample; performing reasoning analysis on the entity content by invoking an intelligent agent according to the task tree to obtain an entity thought chain, and combining the entity thought chain with the root cause sample to obtain the reasoning thought chain sample.

8. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program, when executed by a processor, implements the steps of the defect root cause analysis method of the oil and gas exploration equipment according to any one of claims 1 to 6.

Images