A drill bit recommendation method, device and equipment

Abstract

The present application relates to a kind of drill bit recommendation method, device and equipment, it is related to well drilling engineering technical field, wherein, including: based on formation information knowledge source and drill bit technical knowledge source, generate multiple theoretical candidate drill bits and corresponding theoretical confidence through large language model;Based on historical record knowledge source and driller demand knowledge source, generate multiple experience candidate drill bits and corresponding experience confidence through similar case search;Based on engineering constraint, the multiple theoretical candidate drill bits and multiple experience candidate drill bits are filtered;According to the comparison result of the theoretical confidence of each theoretical candidate drill bit after screening and the experience confidence of each experience candidate drill bit after screening, the theoretical candidate drill bit after screening and experience candidate drill bit are merged, and the drill bit recommendation result of target well is obtained;The drill bit recommendation result includes multiple recommended drill bits and corresponding recommended reason and risk prompt.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of drilling engineering technology, specifically to a drill bit recommendation method, apparatus, and equipment. Background Technology

[0002] In existing technologies, drill bit recommendations rely on multi-source heterogeneous data. However, the different data are stored in isolation and lack deep correlation, resulting in unsatisfactory engineering feasibility, reliability, and adaptability of the drill bit recommendation results. Summary of the Invention

[0003] The purpose of the embodiments in this specification is to provide a drill bit recommendation method, apparatus, and equipment to overcome the problems of poor engineering feasibility, reliability, and adaptability of drill bit recommendation results in existing methods.

[0004] To address the aforementioned technical problems, this specification provides a drill bit recommendation method, comprising: Based on the knowledge sources of formation information and drill bit technology, multiple theoretical candidate drill bits and their corresponding theoretical confidence scores are generated through a large language model. Based on historical knowledge sources and driller demand knowledge sources, multiple candidate drill bits with experience and corresponding experience confidence are generated through similar case retrieval. Based on engineering constraints, the multiple theoretical candidate drill bits and multiple empirical candidate drill bits are screened. Based on the comparison between the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening, the screened theoretical candidate drill bits and empirical candidate drill bits are merged to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding reasons for recommendation and risk warnings.

[0005] Furthermore, embodiments of this specification provide a drill bit recommendation device, comprising: The first generation module is used to generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the knowledge sources of formation information and drill bit technology, through a large language model. The second generation module is used to generate multiple candidate drill bits and their corresponding experience confidence levels based on historical knowledge sources and driller demand knowledge sources through similar case retrieval. The filtering module is used to filter the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints. The merging module is used to merge the screened theoretical candidate drill bits and the empirical candidate drill bits based on the comparison results of the theoretical confidence scores of each screened theoretical candidate drill bit and the empirical confidence scores of each screened empirical candidate drill bit, to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding recommendation reasons and risk warnings.

[0006] Furthermore, embodiments of this specification provide a computer device, including: Memory, used to store computer programs; A processor for executing the computer program to implement the above-described drill bit recommendation method.

[0007] Furthermore, embodiments of this specification provide a computer storage medium storing computer program instructions that, when executed, implement the aforementioned drill bit recommendation method.

[0008] Furthermore, embodiments of this specification provide a computer program product, including a computer program that, when executed by a processor, implements the aforementioned drill bit recommendation method.

[0009] As can be seen from the technical solutions provided in the embodiments of this specification above, these embodiments can generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on formation information knowledge sources and drill bit technology knowledge sources through a large language model; generate multiple empirical candidate drill bits and their corresponding empirical confidence scores based on historical data knowledge sources and driller demand knowledge sources through similar case retrieval; screen the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints; and merge the screened theoretical candidate drill bits and empirical candidate drill bits according to the comparison results of the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and their corresponding recommendation reasons and risk warnings. By generating candidate drill bits through dual paths and combining them with engineering constraint screening, the comprehensiveness and engineering feasibility of the recommendation results are directly improved; the dynamic fusion of candidate drill bits from the two paths based on the confidence score comparison results avoids the failure of recommendations from a single path when data is sparse or theoretically insufficient, effectively enhancing the reliability and adaptability of the recommendation results. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the accompanying drawings used in the description of the embodiments or prior art will be briefly introduced below.

[0011] Figure 1 This is a flowchart of a drill bit recommendation method provided in the embodiments of this specification; Figure 2 The overall architecture diagram of the drill bit recommendation system based on large model and multi-source knowledge fusion provided in the embodiments of this specification; Figure 3 A flowchart of a drill bit recommendation method based on large model and multi-source knowledge fusion provided for embodiments of this specification; Figure 4This is a schematic diagram illustrating the fusion of multi-source heterogeneous drilling knowledge provided in the embodiments of this specification; Figure 5 A diagram illustrating the dual-path candidate drill bit generation mechanism based on theoretical recommendation and empirical verification, provided for embodiments of this specification; Figure 6 A diagram illustrating the adaptive learning and knowledge update optimization mechanism provided in the embodiments of this specification; Figure 7 This document provides a flowchart for the knowledge retrieval enhancement generation process in the embodiments of this specification. Figure 8 This is a schematic diagram of the structural composition of a drill bit recommendation device provided in the embodiments of this specification; Figure 9 This is a schematic diagram of the structural composition of a computer device provided in the embodiments of this specification. Detailed Implementation

[0012] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0013] It should be noted that the terms "first," "second," etc., used in this specification, claims, and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] In some embodiments, drill bits are the core tools for directly breaking rocks in oil and gas drilling operations. Their type, structure, and parameter selection directly affect drilling machinery's rate of penetration, footage capability, operating costs, and downhole safety. Drill bit adaptability varies significantly under different formation conditions, and incorrect drill bit selection can lead to decreased drilling speed, premature drill bit failure, or even downhole accidents.

[0015] In some embodiments, drill bit recommendation can be the process of selecting the optimal drill bit from a variety of drill bit models under specific drilling conditions, taking into account factors such as formation lithology, drilling parameters, historical operation records, and operational objectives. This process involves a comprehensive judgment of multi-dimensional information, including geological, engineering, and experiential data, and is a key decision-making step in drilling engineering design.

[0016] This specification provides a drill bit recommendation method, as illustrated in the embodiments below. Figure 1 As shown, the specific implementation may include the following steps: S101: Based on the knowledge sources of formation information and drill bit technology, multiple theoretical candidate drill bits and their corresponding theoretical confidence scores are generated through a large language model.

[0017] In some embodiments, the formation information knowledge source is used to characterize the lithological, physical, geological structural, and drilling characteristics of the target well formation; the drill bit technology knowledge source is used to characterize the specifications of the drill bit and its historical performance in the target well formation.

[0018] In some embodiments, step S101 may specifically include: based on the formation information knowledge source and the drill bit technology knowledge source, generating multiple theoretical candidate drill bits and their corresponding theoretical confidence levels through the large language model according to the four-element instruction format composed of formation description, operation scenario, historical adjacent well operation records and output expectations.

[0019] The stratigraphic information knowledge source is used to characterize the lithological features, physical properties, geological structure, and drilling characteristics of the formation. It consists of a standardized stratigraphic feature vector extracted from the geological logging report. This vector includes lithological feature components, physical property components, geological structure components, and drilling characteristic components. Each component is encoded by a domain-adaptive pre-trained language model to form a multi-dimensional numerical description of the formation conditions of the target well.

[0020] The drill bit technology knowledge source is used to characterize the specifications and parameters of drill bits and their historical performance in formations. It consists of a static parameter library and a dynamic profile of drill bit formation adaptability. The static parameter library records the drill bit type, cutting tooth size, tooth arrangement, IADC classification code, manufacturer-recommended drill pressure and speed range, and applicable formation description. The dynamic profile is based on historical usage data and is calculated separately for each formation type, including the average mechanical drilling rate, standard deviation of mechanical drilling rate, 10th percentile, 50th percentile, 90th percentile of footage, and reliability score, forming a quantitative description of the drill bit's performance under different formation conditions.

[0021] The large language model can be a generative pre-trained language model that has been fine-tuned using drill bit-recommended corpus. The training samples used in this fine-tuning process are constructed using formation descriptions, operation scenarios, historical adjacent well operation records, and output expected quadruple instruction formats.

[0022] When generating multiple theoretical candidate drill bits and their corresponding theoretical confidence scores, the stratigraphic feature vectors from the stratigraphic information knowledge source and the static parameters and dynamic profile data of the drill bit from the drill bit technology knowledge source can be organized into input text according to the four-element instruction format and input into the large language model. The large language model outputs one or more theoretical candidate drill bit models and the theoretical confidence score value corresponding to each candidate drill bit. The theoretical confidence score value represents the degree of matching between the drill bit and the current stratigraphic conditions based on the theoretical reasoning level.

[0023] By inputting formation information and drill bit technical parameters into a specially fine-tuned large language model in the form of structured instructions, automatic reasoning based on drilling theory knowledge is realized. It can provide theoretically based drill bit recommendations and matching quantification evaluations without relying on historical cases, providing a theoretical candidate set and confidence basis for subsequent dual-path fusion.

[0024] In some embodiments, the above-mentioned generation of multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source, according to a four-element instruction format composed of formation description, operation scenario, historical adjacent well operation records, and output expectations, through the large language model, may further include: determining the formation description field in the four-element instruction format according to the formation information knowledge source; the formation description field includes lithology type, formation depth, drillability level, abrasiveness index, and formation pressure information; determining the operation scenario field in the four-element instruction format according to the driller's requirements knowledge source; the operation scenario field includes well type, well depth range, drilling fluid type, operation priority, and budget constraint information; and further determining the formation description field in the four-element instruction format according to the driller's requirements knowledge source; the operation scenario field includes well type, well depth range, drilling fluid type, operation priority, and budget constraint information. Historical knowledge sources are used to determine the historical adjacent well operation record field in the four-element instruction format. The historical adjacent well operation record field includes a summary of successful or failed cases under similar working conditions, the drill bit model used, the actual mechanical drilling rate, the actual footage, and wear status information. The output expectation field in the four-element instruction format is determined. The output expectation field is used to specify the recommended drill bit model, operation parameter window, expected footage range, failure mode warning, and confidence level description to be output. The formation description field, the operation scenario field, the historical adjacent well operation record field, and the output expectation field are combined into a four-element instruction, which is input into the large language model to obtain multiple theoretical candidate drill bits and the theoretical confidence level corresponding to each theoretical candidate drill bit.

[0025] The formation description field is used to provide complete geological information of the formation of the target well, including lithology, formation depth range, drillability level, abrasiveness index and formation pressure type. This information is directly extracted from the formation information knowledge source and organized into natural language description sentences according to a preset template.

[0026] The operation scenario field is used to describe the engineering conditions and target constraints of the current drilling operation, including well type, current well depth range, drilling fluid system type, driller-specified operation priority, and budget constraints per drill bit. This information is extracted from the driller's requirements knowledge source and converted into standardized expressions.

[0027] The historical adjacent well operation record field is used to provide adjacent well operation experience similar to the current working conditions, including a summary of successful or failed cases under similar formation conditions, the drill bit type used in the case, the actual mechanical drilling rate achieved, the actual footage, and the drill bit wear rating at the time of tripping. This information is obtained from historical knowledge sources through formation similarity retrieval.

[0028] The Output Expectations field specifies the information elements that the output of the large language model should include, including recommended drill bit models, suggested drilling pressure and rotation speed parameter windows, the 10th, 50th, and 90th percentiles of the expected footage, warnings of possible failure modes, and confidence level descriptions.

[0029] The formation description field, operation scenario field, historical adjacent well operation record field, and output expectation field are concatenated into a four-element instruction text in a fixed order and input into the large language model. The large language model generates one or more theoretical candidate drill bit models according to the format specified by the output expectation field, and attaches the theoretical confidence value and confidence source description for each candidate drill bit in the output.

[0030] By adopting a four-element instruction format that corresponds to the actual decision-making logic of drilling engineers, the large language model can obtain complete decision information elements at the input level. The output results directly include key information such as parameter windows, footage ranges, and failure warnings required on-site, which significantly improves the engineering usability and information completeness of the theoretical recommendation results.

[0031] In some embodiments, step S101 may specifically include: generating multiple theoretical candidate drill bits based on the formation information knowledge source and the drill bit technology knowledge source through the large language model, and then inputting each theoretical candidate drill bit into a preset confidence evaluation model to obtain the theoretical confidence level corresponding to each theoretical candidate drill bit.

[0032] The confidence assessment model can be a regression model or a classification model that is independent of the large language model. The input of the model includes the concatenation result of the target well formation feature vector in the formation information knowledge source and the dynamic profile feature vector of the candidate drill bit in the formation in the drill bit technology knowledge source. The output is a confidence value between 0 and 1.

[0033] The confidence assessment model can be pre-trained using successful and unsuccessful cases from historical drilling data. In the training samples, positive examples are successful formation-drill bit combinations in actual use, and negative examples are unsuccessful formation-drill bit combinations in actual use. Through supervised learning, the model can output the matching probability based on the degree of matching between formation conditions and drill bit performance characteristics.

[0034] After generating candidate drill bit models using the large language model, dynamic profile features of each candidate drill bit under the target well formation type are extracted from the drill bit technology knowledge source. These features are then concatenated with formation feature vectors from the formation information knowledge source and input into the confidence assessment model. The confidence assessment model outputs the theoretical confidence score corresponding to each candidate drill bit. This theoretical confidence score is independent of the generation process of the large language model and serves as a secondary verification result of the theoretical matching degree of the candidate drill bits.

[0035] By introducing an independent confidence assessment model to perform secondary verification of theoretical candidate drill bits, the potential confidence estimation bias in the large language model generation results is eliminated, making the theoretical confidence more objective and stable, and providing a more reliable quantitative basis for subsequent dual-path fusion.

[0036] In some embodiments, step S101 may further include: constructing a simplified instruction format containing formation description, operation scenario and output expectation based on the formation information knowledge source and the drill bit technology knowledge source; generating multiple theoretical candidate drill bits through the large language model; retrieving the number of successful cases matching each theoretical candidate drill bit from the historical knowledge source; and determining the theoretical confidence level corresponding to each theoretical candidate drill bit based on the retrieval results.

[0037] The simplified instruction format removes the historical adjacent well operation record field from the quaternary instruction format, retaining only the formation description field, operation scenario field, and output expectation field. This makes the reasoning of the large language model rely entirely on the theoretical matching relationship between formation conditions and drill bit technical parameters, and is not affected by historical case data.

[0038] After the large language model generates one or more theoretical candidate drill bit models based on simplified instructions, it retrieves the number of successful cases matching each theoretical candidate drill bit from the historical knowledge source. During the retrieval, the candidate drill bit model is used as the primary key, and filtering is performed in combination with the target well formation type and operation scenario. The number of successful case entries in the historical knowledge source where both formation similarity and operation similarity exceed preset thresholds is counted.

[0039] The theoretical confidence level of each candidate drill bit can be determined based on the number of successful cases retrieved. The number of successful cases and the theoretical confidence level are converted using a piecewise mapping function or a monotonically increasing function. The more successful cases there are, the higher the theoretical confidence level. When the number of successful cases is zero, the theoretical confidence level is set to the system's preset benchmark value.

[0040] By simplifying the input instructions of the large language model and introducing historical case retrieval results as the basis for determining confidence, the generation process of theoretical candidate drill bits avoids the interference of historical data sparsity or historical data quality fluctuations on the inference results. At the same time, the number of historical successful cases is used to calibrate the theoretical confidence, so as to achieve a reasonable estimate of the theoretical recommendation confidence in the case of insufficient historical data.

[0041] S102: Based on historical knowledge sources and driller demand knowledge sources, multiple candidate drill bits with experience and their corresponding experience confidence scores are generated through similar case retrieval.

[0042] In some embodiments, the historical knowledge source is used to characterize historical use cases of drill bits in adjacent wells of the target well; the driller demand knowledge source is used to characterize the personalized needs of drilling operations in the target well.

[0043] In some embodiments, step S102 may specifically include: based on the historical knowledge source and the driller's demand knowledge source, retrieving multiple successful cases of neighboring wells of the target well according to formation similarity and operation similarity; and generating multiple empirical candidate drill bits and the empirical confidence level corresponding to each empirical candidate drill bit based on the multiple successful cases of neighboring wells.

[0044] The historical knowledge source is used to store usage cases of drill bits in adjacent wells. Each case includes formation information, drill bit model, operating parameters, and drilling performance evaluation indicators. The drilling performance evaluation indicators are determined by a combination of mechanical drilling speed, footage, drill bit wear rating, and whether a downhole accident occurred.

[0045] The driller's demand knowledge source is used to characterize the personalized demands of drilling operations at the target well. It consists of a standardized demand vector extracted from text instructions or structured forms input by the driller. This vector includes weighted requirements for performance indicators such as mechanical drilling speed, footage, cost, reliability, availability, and risk aversion.

[0046] When generating candidate drill bits based on experience and their corresponding experience confidence scores, multiple successful cases of neighboring wells of the target well can be retrieved based on historical knowledge sources and driller demand knowledge sources, according to formation similarity and operation similarity.

[0047] Formation similarity is used to measure the degree of similarity between a target well and its neighboring wells in terms of geological conditions. It is obtained by combining the matching degree of lithology type, the similarity of drillability level, the difference of abrasiveness index, and the degree of overlap of depth range.

[0048] Operational similarity is used to measure the degree of similarity between the target well and adjacent wells in terms of engineering operation conditions. It is obtained by integrating the consistency of well type, the matching degree of drilling fluid type, and the similarity of target footage.

[0049] The overall similarity is determined jointly by formation similarity and operational similarity. Cases from adjacent wells with an overall similarity higher than a preset threshold can be selected from historical knowledge sources as search results. The drill bit models used in each case are extracted as empirical candidate drill bits. The empirical confidence level of each empirical candidate drill bit is calculated by weighting the overall similarity of each case with drilling performance evaluation indicators. Cases with higher overall similarity and better drilling performance contribute more to the empirical confidence level.

[0050] By constructing a comprehensive similarity retrieval mechanism that includes both formation and operational dimensions, the source of empirical candidate drill bits not only considers the similarity of geological conditions, but also takes into account the matching of engineering operational conditions such as well type, drilling fluid, and footage target. The retrieval results are closer to actual working conditions, and the calculation of empirical confidence reflects both the similarity of cases and the quality of success, providing a high-quality empirical candidate set for subsequent dual-path fusion.

[0051] In some embodiments, the above-mentioned retrieval of multiple successful neighboring well cases based on formation similarity and operational similarity, based on the historical knowledge source and the driller's demand knowledge source, may further include: calculating the formation similarity between the target well and each neighboring well based on the lithological type matching degree, drillability level similarity, abrasiveness index difference, and depth range overlap between the target well and each neighboring well; calculating the operational similarity between the target well and each neighboring well based on the well type consistency, drilling fluid type matching degree, and target footage similarity between the target well and each neighboring well; calculating the comprehensive similarity between the target well and each neighboring well based on the formation similarity and the operational similarity; and selecting neighboring wells with a comprehensive similarity higher than a preset threshold to obtain multiple successful neighboring well cases.

[0052] Formation similarity can be calculated based on the quantitative results of four dimensions between the target well and each adjacent well: lithological type matching degree, drillability level similarity, abrasiveness index difference, and depth interval overlap. Among them, lithological type matching degree is measured by semantic distance in the lithological classification system, drillability level similarity is normalized by the reciprocal of the level difference, abrasiveness index difference is normalized by the inverse ratio of the index difference, and depth interval overlap is the proportion of the overlap depth to the total depth of the target well section.

[0053] The similarity of operations is calculated based on the quantitative results of the target well and each neighboring well in three dimensions: well type consistency, drilling fluid type matching degree, and target footage proximity. Well type consistency is measured by categorical variable matching, drilling fluid type matching degree is measured by drilling fluid system categorical matching degree, and target footage proximity is measured by normalized inverse proportion of the target footage difference.

[0054] The overall similarity score is obtained by weighted summation of formation similarity and operational similarity, where the weights for formation similarity and operational similarity are dynamically set based on the exploration stage of the target well and the richness of data types. Neighboring wells with overall similarity scores higher than a preset threshold are selected to obtain multiple successful neighboring well cases. The drill bit models used in each case are extracted as empirical candidate drill bits. An empirical confidence score is calculated based on the overall similarity score of each case, and a positive correlation is established between the overall similarity score and the empirical confidence score.

[0055] By clarifying the specific calculation dimensions included in stratigraphic similarity and operational similarity, the similarity calculation becomes engineering interpretable. The introduction of dynamic weights enables the retrieval mechanism to adaptively adjust according to the complexity of stratigraphic conditions and the quality of historical data. The calculation of empirical confidence is directly related to the comprehensive similarity, ensuring that the ranking criteria for empirical candidate drill bits are clear and transparent.

[0056] In some embodiments, the above-mentioned generation of multiple candidate drill bits and the corresponding empirical confidence scores of each candidate drill bit based on the multiple successful neighboring well cases may further include: extracting the drill bit models used in each successful neighboring well case as candidate drill bits; and calculating the empirical confidence scores of each candidate drill bit based on the comprehensive similarity and drilling performance of each successful neighboring well case.

[0057] When generating multiple candidate drill bits and their corresponding empirical confidence scores based on successful cases from multiple adjacent wells, the drill bit models used in each successful case from adjacent wells are extracted as candidate drill bits. The retrieved successful cases are grouped and merged according to drill bit model; when the same model appears in multiple cases, it is merged into one candidate drill bit. For each candidate drill bit, all its corresponding successful cases are collected, and the empirical confidence score of the candidate drill bit is calculated based on the comprehensive similarity of each case and the drilling performance.

[0058] Drilling performance is quantified by comprehensively measuring the actual mechanical drilling rate, actual footage, drill bit wear rating, and accident indicators recorded in the case studies to obtain an performance score. For each candidate drill bit, the empirical confidence level of the candidate drill bit is obtained by summing the products of the overall similarity of the cases and the performance scores of the cases, and dividing the sum by the total of the overall similarity and performance scores of all cases. When the same drill bit model appears in multiple cases, the empirical confidence level of the drill bit increases with the number of successful cases, and cases with high overall similarity and good drilling performance have a higher weight in contributing to the confidence level.

[0059] By merging the performance of the same drill bit model in different successful cases, the empirical confidence level can comprehensively reflect the effect of multiple successful verifications of the drill bit under similar working conditions, avoiding the excessive influence of the randomness of a single case on the confidence level assessment. The confidence level calculation results tend to stabilize as the number of verifications accumulates.

[0060] In some embodiments, step S102 may specifically include: based on the historical knowledge source and the driller's demand knowledge source, retrieving multiple successful cases of adjacent wells of the target well according to formation similarity to obtain a first candidate set; performing secondary screening on the cases in the first candidate set according to operation similarity to obtain a second candidate set; extracting the drill bit model used in each case in the second candidate set as empirical candidate drill bits, and calculating the empirical confidence level based on the comprehensive score of each case in the two searches.

[0061] Based on historical knowledge sources and driller demand knowledge sources, multiple successful cases of adjacent wells are retrieved according to formation similarity to obtain the first candidate set. Formation similarity is calculated based on four dimensions between the target well and each adjacent well: lithological type matching degree, drillability level similarity, abrasiveness index difference, and depth range overlap. During the search, a formation similarity threshold is set, and only adjacent well cases with formation similarity exceeding the threshold are retained.

[0062] The cases in the first candidate set are further filtered based on the job similarity to obtain the second candidate set. The job similarity is calculated based on three dimensions: consistency of well type, matching degree of drilling fluid type, and similarity of target footage between the target well and each adjacent well. During the second filtering, a job similarity threshold is set, and only cases with job similarity exceeding the threshold are retained.

[0063] The drill bit models used in each case in the second candidate set can be extracted as empirical candidate drill bits. The comprehensive score is calculated by combining the first score of each case in the formation similarity retrieval stage and the second score in the operation similarity secondary screening stage. The comprehensive score is then normalized and used as the empirical confidence level of each empirical candidate drill bit.

[0064] By adopting a two-step sequential screening mechanism that prioritizes strata analysis before operational considerations, the retrieval process first ensures the basic similarity of geological conditions, and then further screens the matching of engineering operation conditions. Compared with a single comprehensive retrieval method, this approach has higher screening efficiency and stronger engineering logic.

[0065] In some embodiments, step S102 may specifically include: based on the historical knowledge source and the driller's demand knowledge source, retrieving multiple successful and failed cases of adjacent wells of the target well according to formation similarity; extracting the drill bit models used in the successful cases as positive candidate drill bits, and extracting the drill bit models used in the failed cases as negative candidate drill bits; calculating the positive recommendation probability and negative avoidance probability of each candidate drill bit according to the occurrence frequency of the positive and negative candidate drill bits, and generating empirical candidate drill bits and corresponding empirical confidence levels according to the positive recommendation probability and negative avoidance probability.

[0066] Based on historical knowledge sources and driller demand knowledge sources, multiple successful and failed cases of adjacent wells are retrieved for the target well according to formation similarity and operational similarity. The calculation methods for formation similarity and operational similarity are not detailed here; the overall similarity is obtained by fusing the two. Both successful and failed cases are retained during the search. Successful cases are those whose drilling performance evaluation meets the preset excellent standard, while failed cases refer to those that have experienced failures such as drill bit tooth breakage, mud packing, twisting, premature wear, or whose drilling performance evaluation is below the preset standard.

[0067] The drill bit models used in successful cases are extracted as positive candidate drill bits, and those used in failed cases are extracted as negative candidate drill bits. For each candidate drill bit model, its frequency of occurrence in successful cases and in failed cases is calculated, with both success and failure frequencies weighted and accumulated based on the overall similarity of the cases. Based on the weighted success frequency of positive candidate drill bits and the weighted failure frequency of negative candidate drill bits, the positive recommendation probability and negative avoidance probability of each candidate drill bit are calculated. The positive recommendation probability is the ratio of the weighted success frequency to the sum of the weighted success frequencies of all candidate drill bits, and the negative avoidance probability is the ratio of the weighted failure frequency to the sum of the weighted failure frequencies of all candidate drill bits. The difference between the positive recommendation probability and the negative avoidance probability is used as the empirical confidence level; a positive difference indicates the drill bit is included in the empirical candidate set, while a negative difference indicates the drill bit is excluded.

[0068] By simultaneously introducing both successful and unsuccessful cases for two-way verification, the generation of experience-based candidate drill bits not only relies on the accumulation of successful experiences but also accepts the exclusion constraint of lessons learned from failures. The positive recommendation probability and negative avoidance probability can effectively identify drill bit models with high success rates and low failure rates under similar working conditions, significantly improving the reliability of experience-based recommendations in risk-sensitive scenarios.

[0069] In some embodiments, step S102 may specifically include: based on the historical knowledge source and the driller's demand knowledge source, retrieving multiple successful cases of adjacent wells of the target well according to formation similarity and operation similarity; assigning timeliness weights to the retrieved successful cases according to their time sequence, with cases that are closer in time having higher timeliness weights; and calculating the empirical confidence of each empirical candidate drill bit by weighting the similarity and timeliness weights of each case.

[0070] Based on historical knowledge sources and driller demand knowledge sources, multiple successful cases from neighboring wells are retrieved for the target well according to formation similarity and operational similarity, resulting in a set of retrieval results. Successful cases are assigned time-sensitivity weights based on their chronological order, with higher weights for cases occurring more recently. Time-sensitivity weights can be determined using, but are not limited to, an exponential decay function, with the decay rate set according to the drilling technology update speed and drill bit model iteration cycle of the block. For each candidate drill bit, all corresponding successful cases are collected. The empirical confidence level of the candidate drill bit is calculated by weighting the comprehensive similarity of each case with its time-sensitivity weight. One specific calculation method is to sum the products of the comprehensive similarity and time-sensitivity weights of each case, and then divide the sum by the total product of all cases' comprehensive similarity and time-sensitivity weights to obtain the normalized empirical confidence level. When multiple successful cases of the same drill bit model are distributed across different time intervals, recent cases contribute significantly more to the empirical confidence level than earlier cases, while earlier cases only serve as a reference when relevant recent cases are lacking.

[0071] By introducing a time-sensitivity weight into the calculation of experience confidence, the experience recommendation results can reflect the progress of drill bit technology and the application effect of new models, avoiding the system's over-reliance on early success cases and ignoring the technologically superior drill bits that have emerged in recent years. At the same time, it retains the reference value of early cases in data-sparse scenarios, and achieves a dynamic balance of the time sensitivity of experience knowledge.

[0072] S103: Based on engineering constraints, the multiple theoretical candidate drill bits and multiple empirical candidate drill bits are screened.

[0073] In some embodiments, the engineering constraints include size matching constraints, drill pressure safety constraints, and formation type adaptation constraints; the size matching constraints characterize the matching relationship between the drill bit outer diameter and the target wellbore size; the drill pressure safety constraints characterize the safety relationship between the recommended drill pressure and the drill string buckling critical load; and the formation type adaptation constraints characterize the adaptation relationship between the drill bit type and the formation drillability level.

[0074] Size matching constraints are used to limit the matching relationship between the drill bit outer diameter and the target wellbore size. It requires that the gap between the drill bit outer diameter and the wellbore size be within the range allowed by engineering specifications. If the gap is too small, it will lead to difficulty in running the drill bit or the risk of stuck drill bit. If the gap is too large, it will affect the drilling stability of the drill bit and the wellbore quality.

[0075] The drill pressure safety constraint is used to limit the safe relationship between the recommended drill pressure and the drill string buckling critical load. The drill string buckling critical load is dynamically calculated based on the current well depth and drill string combination. The maximum recommended drill pressure must not exceed the product of the drill string buckling critical load and the safety factor. Drill pressure exceeding this range may cause helical buckling or sinusoidal buckling of the drill string, leading to drill string failure.

[0076] Formation type compatibility constraints are used to limit the compatibility between drill bit type and formation drillability level. Hard constraint rules are established based on drilling engineering common sense. When the drillability level is in a specific range, drill bit types that are not compatible with that range are excluded.

[0077] In some embodiments, step S103 may specifically include: screening theoretical candidate drill bits and empirical candidate drill bits that simultaneously satisfy scale matching constraints, drill pressure safety constraints and formation type adaptation constraints.

[0078] When screening theoretical and empirical candidate drill bits, size matching constraints, drilling pressure safety constraints, and formation type adaptation constraints are applied simultaneously. Candidate drill bits must meet all three constraints at the same time. Candidate drill bits that do not meet any of the constraints are directly excluded and do not enter the subsequent processing flow.

[0079] By applying three hard engineering constraints simultaneously for rigorous screening, it is ensured that the candidate drill bits entering the subsequent ranking meet the engineering feasibility requirements in three dimensions: physical dimensions, drill string mechanical safety, and formation type adaptability. This fundamentally eliminates the possibility of physically infeasible solutions in the recommended results and ensures the engineering feasibility of the recommended results.

[0080] In some embodiments, step S103 may specifically include: screening theoretical candidate drill bits and empirical candidate drill bits that satisfy at least one of size matching constraints, drill pressure safety constraints and formation type adaptation constraints.

[0081] When screening theoretical and empirical candidate drill bits, candidates must meet at least one of the following constraints: size matching, drill pressure safety, and formation type suitability. A candidate drill bit is retained for further processing only if it meets any one of these three constraints; otherwise, it is excluded. This screening method allows candidate drill bits to be retained even when some engineering constraints cannot be met, and the subsequent recommendation and ranking process adjusts the scores based on constraint compliance.

[0082] By adopting a relaxed screening criterion that satisfies at least one constraint, more candidate drill bits are retained for the ranking process while ensuring the safety of the basic engineering. This is suitable for scenarios with sparse historical data or special geological conditions, avoiding the situation where the candidate set is too small due to strict screening and no recommended results are available. It provides a more sufficient source of candidates for subsequent confidence fusion and ranking.

[0083] In some embodiments, step S103 may specifically include: applying size matching constraints, drill pressure safety constraints, and formation type adaptation constraints in sequence according to preset priorities, retaining theoretical candidate drill bits and empirical candidate drill bits that meet the current constraints, and then continuing to apply the next constraint for screening.

[0084] When screening theoretical and empirical candidate drill bits, size matching constraints, drill pressure safety constraints, and formation type adaptation constraints are applied sequentially according to preset priorities. The preset priorities are: size matching constraint as the highest priority, drill pressure safety constraint as the second highest priority, and formation type adaptation constraint as the lowest priority. First, the size matching constraint is applied, retaining candidate drill bits that meet the constraint and eliminating those that do not. Then, the drill pressure safety constraint is applied to the retained candidate drill bits, retaining those that meet the constraint and eliminating those that do not. Finally, the formation type adaptation constraint is applied to the remaining candidate drill bits, retaining those that meet the constraint and eliminating those that do not. After this three-level progressive screening, the remaining candidate drill bits are the screening results and enter the subsequent processing flow.

[0085] By applying engineering constraints in a priority order for screening, the highest priority size matching constraint takes effect at the forefront of the screening process. This can quickly eliminate candidate drill bits with mismatched physical dimensions, reduce the computational overhead of subsequent constraint applications, and at the same time, the priority order reflects the difference in importance of different constraints in drilling engineering, ensuring that the screening logic is consistent with the safety priority principle in engineering practice.

[0086] In some embodiments, step S103 may specifically include: using size matching constraints as hard constraints, and using drill pressure safety constraints and formation type adaptation constraints as soft constraints, to screen theoretical candidate drill bits and empirical candidate drill bits that satisfy size matching constraints and at least one of drill pressure safety constraints or formation type adaptation constraints.

[0087] When screening theoretical and empirical candidate drill bits, size matching constraints are set as hard constraints, while drill pressure safety constraints and formation type adaptation constraints are set as soft constraints. The distinction between hard and soft constraints is based on whether a violation of the constraint directly leads to an engineering safety accident. Violation of the size matching constraint directly causes the drill bit to fail to descend or become stuck, constituting a hard constraint. Violation of the drill pressure safety constraint may cause drill string buckling, and violation of the formation type adaptation constraint may cause drill bit failure; both constitute soft constraints. During screening, the size matching constraint is applied first to exclude candidate drill bits that do not meet this hard constraint. Among the candidate drill bits that pass the hard constraint screening, further screening is conducted to ensure that at least one of the drill pressure safety constraint or formation type adaptation constraint is met. That is, a candidate drill bit, provided it meets the size matching constraint, only needs to meet either the drill pressure safety constraint or the formation type adaptation constraint to be retained for subsequent processing.

[0088] By setting the size matching constraint as a hard constraint and applying it strictly, and setting the drill pressure safety constraint and formation type adaptation constraint as soft constraints and using a relaxed combination of conditions, a reasonable elastic space is provided for drill pressure parameters and formation adaptability while ensuring the absolute safety of physical dimensions. This is suitable for scenarios where there is some uncertainty between the drill pressure safety boundary and formation adaptation conditions, and avoids the mistaken exclusion of feasible candidate drill bits due to excessively strict soft constraints.

[0089] In some embodiments, step S103 may specifically include: applying size matching constraints, drilling pressure safety constraints, and formation type adaptation constraints to screen theoretical candidate drill bits and empirical candidate drill bits respectively, obtaining screening results corresponding to each constraint, and selecting theoretical candidate drill bits and empirical candidate drill bits that appear in at least two screening results as screened candidate drill bits.

[0090] When screening theoretical and empirical candidate drill bits, size matching constraints, drill pressure safety constraints, and formation type adaptation constraints are applied independently to obtain screening results for each constraint. Specifically, size matching constraints are applied to all candidate drill bits, and those that meet the constraints are retained, resulting in the first screening result set. Drill pressure safety constraints are applied to all candidate drill bits, and those that meet the constraints are retained, resulting in the second screening result set. Formation type adaptation constraints are applied to all candidate drill bits, and those that meet the constraints are retained, resulting in the third screening result set. Theoretical and empirical candidate drill bits that appear in at least two screening result sets are selected as final candidate drill bits; that is, if a candidate drill bit appears at least twice in all three screening result sets, it is retained for further processing.

[0091] By applying three constraints and using a multi-result set voting mechanism for screening, candidate drill bits only need to meet any two of the three constraints to pass the screening. This is more inclusive than the strict screening that meets all three constraints at the same time, and more secure than the lenient screening that only meets a single constraint. The voting mechanism can identify candidate drill bits that perform well in most constraint dimensions, while excluding candidate drill bits that meet the requirements in only one dimension but have serious defects in other dimensions, thus achieving a balance between engineering safety and candidate sufficiency.

[0092] S104: Based on the comparison results of the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening, the theoretical candidate drill bits and the empirical candidate drill bits after screening are merged to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding recommendation reasons and risk warnings.

[0093] In some embodiments, step S104 may specifically include: determining a merging strategy based on the comparison results of the theoretical confidence scores of each theoretical candidate drill bit after screening and the empirical confidence scores of each empirical candidate drill bit after screening; merging the screened theoretical candidate drill bits and empirical candidate drill bits according to the merging strategy to obtain multiple recommended drill bits; and generating a drill bit recommendation result based on the multiple recommended drill bits.

[0094] The selected theoretical candidate drill bits can be a set of drill bit models generated from the theoretically recommended paths retained after engineering constraint screening. Each theoretical candidate drill bit corresponds to a theoretical confidence score, which represents the matching reliability derived from drilling theory knowledge. The selected empirical candidate drill bits can be a set of drill bit models generated from the empirical verification paths retained after engineering constraint screening. Each empirical candidate drill bit corresponds to an empirical confidence score, which represents the matching reliability derived from historical successful case verification.

[0095] The merging strategy is determined based on the comparison between the theoretical confidence scores of each selected theoretical candidate drill bit and the empirical confidence scores of each selected empirical candidate drill bit. The comparison results include the numerical ranges of the theoretical and empirical confidence scores and their relative relative strengths and weaknesses. Following the determined merging strategy, the selected theoretical and empirical candidate drill bits are merged into a unified recommended drill bit set. During the merging process, the candidate drill bits from different sources are weighted, prioritized, or selected based on the merging strategy.

[0096] Based on the multiple recommended drill bits obtained after merging, a drill bit recommendation result is generated. The drill bit recommendation result includes a list of recommended drill bit models, as well as the recommendation reason and risk warning for each recommended drill bit. The recommendation reason explains the basis for the selection of the drill bit, and the risk warning is used to warn of the failure modes or operational precautions that the drill bit may encounter in the target well section.

[0097] By introducing a dynamic merging mechanism based on confidence comparison, the fusion method of the two types of candidate drill bits, namely theoretical recommendation and experience recommendation, can be flexibly adjusted according to the actual situation of their respective confidence levels. This avoids unreasonable decisions when using fixed weight fusion in scenarios with significant differences in confidence levels, and provides a differentiated merging basis for the generation of subsequent recommendation results.

[0098] In some embodiments, the above-mentioned determination of a merging strategy based on the comparison results of the theoretical confidence scores of each selected theoretical candidate drill bit and the empirical confidence scores of each selected empirical candidate drill bit, and merging the selected theoretical candidate drill bits and empirical candidate drill bits according to the merging strategy to obtain multiple recommended drill bits, may further include: if both the theoretical confidence scores of the selected theoretical candidate drill bits and the empirical confidence scores of the selected empirical candidate drill bits are higher than a first preset threshold, a collaborative fusion strategy is adopted to merge the theoretical candidate drill bits and the empirical candidate drill bits according to a preset ratio as recommended drill bits; if the theoretical confidence scores of the selected theoretical candidate drill bits and the empirical candidate drill bits are both higher than a first preset threshold, a collaborative fusion strategy is adopted to merge the theoretical candidate drill bits and the empirical candidate drill bits according to a preset ratio as recommended drill bits; if the theoretical confidence scores of the selected theoretical candidate drill bits are higher than the first preset threshold and the empirical confidence scores of the selected empirical candidate drill bits are higher than the .... If the empirical confidence level of the selected drill bit is lower than the second preset threshold, a theory-driven strategy is adopted, using theoretical candidate drill bits as the primary recommended drill bits and empirical candidate drill bits as the secondary recommended drill bits. If the theoretical confidence level of the selected theoretical candidate drill bits is lower than the second preset threshold and the empirical confidence level of the selected empirical candidate drill bits is higher than the first preset threshold, an experience-driven strategy is adopted, using empirical candidate drill bits as the primary recommended drill bits and theoretical candidate drill bits as the secondary recommended drill bits. If both the theoretical confidence level and the empirical confidence level of the selected theoretical candidate drill bits are lower than the second preset threshold, a rule-based recommendation strategy is adopted, generating recommended drill bits according to the IADC classification rules.

[0099] When there are theoretical candidate drill bits after screening and the theoretical confidence level of the theoretical candidate drill bit is higher than the first preset threshold, and at the same time there are empirical candidate drill bits after screening and the empirical confidence level of the empirical candidate drill bit is also higher than the first preset threshold, a collaborative fusion strategy is adopted to merge the theoretical candidate drill bits and the empirical candidate drill bits according to a preset ratio as recommended drill bits. The number of drill bits from theoretical sources and empirical sources in the merged recommended drill bit set is kept balanced.

[0100] When there are theoretical candidate drill bits after screening and the theoretical confidence level of the theoretical candidate drill bit is higher than the first preset threshold, while there are empirical candidate drill bits after screening but the empirical confidence level of the empirical candidate drill bit is lower than the second preset threshold, the theory-driven strategy is adopted, and the theoretical candidate drill bit is used as the main recommended drill bit, while the empirical candidate drill bit is only used as an auxiliary recommended drill bit and placed after the theoretical candidate drill bit in the recommendation list.

[0101] When there are empirical candidate drill bits after screening and the empirical confidence level of the empirical candidate drill bit is higher than the first preset threshold, while there are theoretical candidate drill bits after screening but the theoretical confidence level of the theoretical candidate drill bit is lower than the second preset threshold, the experience-driven strategy is adopted, and the empirical candidate drill bit is used as the main recommended drill bit, while the theoretical candidate drill bit is only used as an auxiliary recommended drill bit and placed after the empirical candidate drill bit in the recommendation list.

[0102] When the theoretical confidence score of both the theoretical candidate drill bits and the empirical confidence score of the empirical candidate drill bits are lower than the second preset threshold, a rule-based recommendation strategy is adopted. Dual-path candidate drill bits are discarded, and recommended drill bits are generated according to the IADC classification rules. The IADC classification rules provide basic recommendations based on the correspondence between drill bit type, formation hardness grade, and drill bit structural characteristics. The first and second preset thresholds are calibrated based on historical data, with the first preset threshold being higher than the second preset threshold. The two thresholds form the boundary between high-confidence, medium-confidence, and low-confidence intervals.

[0103] By dividing theoretical and empirical confidence into three intervals—high, medium, and low—and defining four corresponding fusion strategies, the merging mechanism can finely distinguish different combinations of dual-path confidence. In high-confidence collaborative scenarios, it fully utilizes the advantages of dual paths; in single-path high-confidence scenarios, it rationally allocates dominant positions; and in dual-path low-confidence scenarios, it promptly downgrades to rule-based recommendations, thus achieving adaptive matching between fusion strategies and data quality.

[0104] In some embodiments, generating a drill bit recommendation result based on the plurality of recommended drill bits may further include: calculating a plurality of drilling engineering parameters for each recommended drill bit; determining the weighting coefficient of each drilling engineering parameter based on the driller's demand information; determining the recommendation degree of the recommended drill bit based on the plurality of drilling engineering parameters and the weighting coefficient of each drilling engineering parameter; and generating a drill bit recommendation result based on the plurality of recommended drill bits and their corresponding recommendation degrees.

[0105] Drilling engineering parameters include the average mechanical drilling rate (MRR), median footage, cost per footage, reliability score, inventory availability score, and failure risk coefficient. The average MRR is obtained from the drill bit technical knowledge source, representing the historical average MRR of the recommended drill bit in the target well section and corresponding formation type. The median footage is obtained from the drill bit technical knowledge source, representing the 50th percentile of footage for the recommended drill bit in similar formation types. The cost per footage is calculated by dividing the drill bit purchase price by the median footage. The reliability score is obtained from the drill bit technical knowledge source, representing the statistical reliability score of the recommended drill bit in similar formation types. The inventory availability score is determined based on the current drill bit inventory status. The failure risk coefficient is calculated based on the failure mode probability distribution of the recommended drill bit in similar formation types from the drill bit technical knowledge source.

[0106] The weighting coefficients of each drilling engineering parameter are determined based on the driller's demand information, which is obtained from a driller demand knowledge source and includes the relative importance of performance indicators such as mechanical drilling rate, footage, cost, reliability, availability, and risk mitigation. For each recommended drill bit, multiple drilling engineering parameters and their weighting coefficients are used. Each parameter value is normalized, multiplied by its corresponding weighting coefficient, and summed to obtain the recommendation score for that drill bit. Based on multiple recommended drill bits and their corresponding recommendation scores, drill bit recommendations are generated in descending order of recommendation score. The recommendation results include the recommendation score value for each recommended drill bit and its corresponding detailed drilling engineering parameters.

[0107] By introducing a parameter system directly corresponding to drilling engineering KPIs for recommendation degree calculation, the recommendation ranking results have clear engineering semantics. The weight coefficients are dynamically determined according to the driller's needs, enabling the recommendation results to reflect differentiated objectives under different operating scenarios. The recommendation degree calculation process is transparent and traceable, providing a quantitative basis for the generation of recommendation reasons.

[0108] In some embodiments, multiple drilling engineering parameters are calculated for each recommended drill bit; based on driller demand information, the weighting coefficients of each drilling engineering parameter are determined, which may further include: for each recommended drill bit, obtaining the average mechanical drilling rate, median footage, cost per unit footage, reliability score, inventory availability score, and failure risk coefficient of the recommended drill bit in the formation type corresponding to the target well section, as multiple drilling engineering parameters; identifying the intent type of the target well drilling operation based on the driller demand information in the driller demand knowledge source; the intent type includes at least one of conventional development, exploration well evaluation, cost sensitive, emergency incident handling, or speed priority; and determining the weighting coefficients corresponding to the multiple drilling engineering parameters based on the intent type.

[0109] For each recommended drill bit, the average mechanical rate of penetration, median footage, cost per unit footage, reliability score, inventory availability score, and failure risk coefficient for that recommended drill bit in the target well section and corresponding formation type are obtained as multiple drilling engineering parameters. Based on the driller's requirements information in the driller's requirements knowledge source, the intent type of the drilling operation for the target well is identified. Intent types include at least one of conventional development, exploration well evaluation, cost-sensitive, emergency incident handling, or speed priority. Intent types are obtained by extracting intent classification tags from the driller's text instructions or reading the operation target field from a structured form.

[0110] Based on the identified intent types, weighting coefficients are determined for multiple drilling engineering parameters, with different combinations of weighting coefficients corresponding to different intent types. Under the conventional development intent, the weights for mechanical drilling rate and footage are moderate, while the cost weight is relatively low; under the exploration evaluation intent, the weights for reliability and risk are significantly increased; under the cost-sensitive intent, the cost per footage weight is the highest; under the emergency incident handling intent, the weights for inventory availability and reliability are increased; and under the speed-first intent, the mechanical drilling rate weight is the highest. The average mechanical drilling rate, median footage, cost per footage, reliability score, and failure risk coefficient of each recommended drill bit are multiplied by the weighting coefficients for their respective intent types, and then summed to obtain the overall recommendation score for that recommended drill bit. The failure risk coefficient is used as a penalty in the summation calculation with a negative value; that is, the higher the failure risk coefficient, the lower the overall recommendation score.

[0111] By directly mapping the intent type identification results in the driller's demand information to the weighted coefficient combination of drilling engineering parameters, the recommendation ranking mechanism can accurately respond to the core demands of different operation scenarios. Conventional development wells pursue a balance between efficiency and footage, exploration wells emphasize reliability and safety, cost-sensitive wells focus on economy, emergency accident handling focuses on availability and reliability, and speed-priority wells highlight mechanical drilling speed, thus achieving a deep alignment between the recommendation results and the operation objectives.

[0112] In some embodiments, determining the recommendation degree of a recommended drill bit based on multiple drilling engineering parameters of each recommended drill bit and the weighting coefficient of each drilling engineering parameter may further include: for each recommended drill bit, multiplying the mean mechanical drilling rate, median footage, cost per unit footage, reliability score, and failure risk coefficient by their respective weighting coefficients and then summing the results to obtain a comprehensive recommendation degree.

[0113] In some embodiments, determining the recommendation level of a recommended drill bit based on multiple drilling engineering parameters and weighting coefficients of each recommended drill bit may further include: for each recommended drill bit, multiplying the mean mechanical drilling rate, median footage, cost per unit footage, reliability score, and failure risk coefficient by their respective weighting coefficients and summing the results to obtain a basic recommendation level; correcting the basic recommendation level based on the inventory availability score of the recommended drill bit; and applying a decay coefficient to the basic recommendation level when the inventory availability score is lower than a preset threshold to obtain a comprehensive recommendation level for the recommended drill bit.

[0114] For each recommended drill bit, the mean mechanical drilling rate, median footage, cost per unit footage, reliability score, and failure risk coefficient are multiplied by their respective weighting coefficients and then summed to obtain the basic recommendation score. The mean mechanical drilling rate and median footage are normalized before calculation, the cost per unit footage is inverted and normalized before calculation, the reliability score is directly used as the normalized value, and the failure risk coefficient is used as a penalty term and normalized after being negative before calculation. The normalization range of each parameter is uniformly mapped to the interval between 0 and 1, ensuring comparability of parameters with different dimensions. The basic recommendation score is adjusted based on the drill bit's inventory availability score. The inventory availability score is determined based on the drill bit's inventory status: 1 for when in stock, 0.6 for when procurement is required and delivery is within the work window, and 0 for when there is no stock and delivery is outside the work window. When the inventory availability score falls below a preset threshold, a decay coefficient is applied to the basic recommendation score. The decay coefficient is positively correlated with the inventory availability score; the lower the inventory availability score, the greater the decay, thus obtaining the comprehensive recommendation score for the recommended drill bit. When the inventory availability score reaches or exceeds a preset threshold, the overall recommendation score equals the base recommendation score, without any decay. The overall recommendation scores of each recommended drill bit are sorted from highest to lowest to generate a final recommendation list, with the drill bit having the highest overall recommendation score in the list being the preferred choice.

[0115] By introducing an inventory availability score as a correction factor after the basic recommendation calculation, the final ranking of the recommendation results can reflect the actual supply of drill bits in real time, avoiding the inability to execute recommendations due to lack of inventory. The decay mechanism when the inventory availability score is below the threshold ensures that drill bits with inventory are selected first when the recommendation quality is similar, and high-quality drill bits that need to be purchased are still retained as alternatives when the recommendation quality differs significantly.

[0116] In some embodiments, when generating drill bit recommendation results based on multiple recommended drill bits, structured recommendation output content is constructed for each recommended drill bit.

[0117] The recommended output includes the recommended drill bit model, comprehensive score, score composition, recommendation rationale, footage prediction, risk warning, confidence level explanation, and alternative solutions. The comprehensive score uses the normalized value of the recommendation score calculation. The score composition includes formation matching degree, historical case support degree, demand matching degree, and reliability rating. Formation matching degree is taken from the contribution of the recommended drill bit in the theoretical confidence level or empirical confidence level. Historical case support degree is determined by combining the number and similarity of successful cases corresponding to experienced candidate drill bits. Demand matching degree is determined by the similarity between the engineering parameters of the recommended drill bit and the driller's demand vector. The reliability rating is taken from the reliability score in the drill bit technical knowledge source. The recommendation rationale generates a natural language description based on the dimension with the highest contribution in the score composition, and elaborates on the matching relationship between formation characteristics and drill bit performance characteristics. The footage prediction is presented in the form of 10th percentile, 50th percentile, and 90th percentile of footage. Each quantile is obtained from the historical footage distribution of the drill bit in the same formation type in the drill bit technical knowledge source. The risk warning is generated based on the failure mode probability distribution of the drill bit in similar formation types from the drill bit technical knowledge source. The mode with the highest failure probability is selected as the primary risk warning, along with monitoring recommendations. The confidence level description includes the overall confidence level value of this recommendation and the composition ratio of the confidence level sources. The respective weights of theoretical derivation and empirical verification are clearly marked in the confidence level source composition. The alternative solutions select 2 to 3 drill bit models with an overall score second only to the primary recommendation, and indicate the main performance differences between each alternative solution and the primary recommendation.

[0118] By constructing a structured recommendation output that includes scoring components, footage quantile predictions, confidence level explanations, and alternative solutions, drillers can fully understand the quantitative basis, expected effects, and risk boundaries of the recommendation results. The direct correspondence between the scoring components and the comprehensive score enhances the interpretability of the recommendation results. The footage quantile predictions replace single-point predictions, improving the engineering reference value. The confidence level explanations enable drillers to judge the reliability of the recommendations, and the alternative solutions provide decision-making flexibility, effectively improving the on-site adoption rate of the recommendation results.

[0119] In some embodiments, the above method may further include: acquiring evidence fragments in parallel through vector similarity retrieval and graph relation retrieval based on a drilling knowledge graph; the drilling knowledge graph uses formation, well section, drill bit model, historical cases, evaluation indicators, and risk events as graph nodes, and uses adaptation, applicability, effect, similarity, and constraint as edge relations; combining the retrieved evidence fragments into a domain context, inputting it into the large language model, and generating recommendation results with evidence chains; wherein, the evidence fragments include formation evidence, adjacent well case evidence, drill bit applicability boundaries, and risk warnings.

[0120] Specifically, a drilling knowledge graph is constructed, with formations, well sections, drill bit models, historical cases, evaluation indicators, and risk events as graph nodes, and "adapted to," "used for," "produces effects," "similar to," and "constrained by" as edge relationships, forming a searchable and traceable evidence network.

[0121] Upon receiving the driller's request and current well condition parameters, the system simultaneously performs vector similarity retrieval and graph relation retrieval to obtain evidence fragments related to the current formation, operational objectives, and drill bit parameters. These retrieved evidence fragments are organized into a domain context including formation evidence, adjacent well case evidence, drill bit applicability boundaries, and risk warnings. This context is then input into a large language model, enabling the model to generate candidate drill bits, recommendation rationales, applicability conditions, and risk descriptions under evidence constraints. The system validates the generated results from the large model against parameter boundaries, inventory status, cost constraints, and safety rules, filtering out recommendations that do not meet engineering constraints and retaining the evidence chain corresponding to the recommendation to improve the interpretability and auditability of the recommendations.

[0122] In some embodiments, the screening of the plurality of theoretical candidate drill bits and the plurality of empirical candidate drill bits based on engineering constraints includes at least one of the following methods: Method 1: Screen theoretical and empirical candidate drill bits that simultaneously meet the constraints of size matching, drill pressure safety, and formation type suitability. Method 2: Select theoretical and empirical candidate drill bits that meet at least one of the following constraints: size matching, drill pressure safety, and formation type adaptation. Method 3: Apply size matching constraints, drill pressure safety constraints, and formation type adaptation constraints in sequence according to preset priorities. After retaining theoretical and empirical candidate drill bits that meet the current constraints, continue to apply the next constraint for screening. Method 4: Use size matching constraints as hard constraints and drill pressure safety constraints and formation type adaptation constraints as soft constraints to screen theoretical and empirical candidate drill bits that meet size matching constraints and at least one of drill pressure safety constraints or formation type adaptation constraints. Method 5: Apply size matching constraints, drill pressure safety constraints, and formation type adaptation constraints to screen theoretical and empirical candidate drill bits respectively, obtain the screening results corresponding to each constraint, and select theoretical and empirical candidate drill bits that appear in at least two screening results as the screened candidate drill bits.

[0123] For Method 1, theoretical and empirical candidate drill bits that simultaneously meet the constraints of size matching, drilling pressure safety, and formation type compatibility are selected. Candidate drill bits that do not meet any of the constraints are directly excluded and do not proceed to the subsequent processing.

[0124] Method 2: Select candidate drill bits that meet at least one of the following constraints: size matching, drill pressure safety, and formation type adaptation. Only when a candidate drill bit does not meet all three constraints will it be excluded. This method is suitable for scenarios with sparse historical data or special formation conditions.

[0125] For Method 3: Size matching constraint, drill pressure safety constraint and formation type adaptation constraint are applied sequentially according to preset priorities. The preset priority sets size matching constraint as the highest priority, drill pressure safety constraint as the second highest priority, and formation type adaptation constraint as the lowest priority. The remaining candidate drill bits after three levels of screening are used as the screening results.

[0126] For method four: set the size matching constraint as a hard constraint and the drill pressure safety constraint and formation type adaptation constraint as soft constraints. Under the premise of satisfying the size matching constraint, the candidate drill bit only needs to satisfy either the drill pressure safety constraint or the formation type adaptation constraint to be retained.

[0127] For method five: apply size matching constraints, drill pressure safety constraints, and formation type adaptation constraints independently to screen candidate drill bits, obtain the screening results corresponding to each constraint, and select candidate drill bits that appear in at least two screening results as the screened candidate drill bits to achieve a balance between engineering safety and candidate sufficiency.

[0128] In some embodiments, generating drill bit recommendation results based on the plurality of recommended drill bits may further include: calculating multiple drilling engineering parameters for each recommended drill bit; determining the weighting coefficient of each drilling engineering parameter based on the driller's demand information and the context parameters of the current operation; the context parameters include well condition complexity, operation urgency, and inventory status; determining the recommendation degree of each recommended drill bit based on the multiple drilling engineering parameters and the weighting coefficient of each drilling engineering parameter; and generating drill bit recommendation results based on the plurality of recommended drill bits and their corresponding recommendation degrees.

[0129] When generating drill bit recommendations based on multiple recommended drill bits, the average mechanical drilling rate, median footage, cost per unit footage, reliability score, inventory availability score, and failure risk coefficient for each recommended drill bit in the target well section and corresponding formation type are obtained as multiple drilling engineering parameters.

[0130] Based on the driller's requirements and the context parameters of the current operation, the weighting coefficients of each drilling engineering parameter are determined. Context parameters include well condition complexity, operational urgency, and inventory status. Well condition complexity is determined comprehensively based on the wellbore trajectory of the target well section, formation pressure coefficient, and the degree of fracture development; operational urgency is determined based on drilling schedule constraints and the operational stage; and inventory status is determined based on the current drill bit material reserves. These context parameters are converted into adjustment factors for the weighting coefficients of each engineering parameter through a preset mapping function, ensuring that the weighting coefficients reflect the actual constraints of the current operational environment.

[0131] Based on multiple drilling engineering parameters and their weighting coefficients for each recommended drill bit, the parameter values ​​are normalized, multiplied by their corresponding weighting coefficients, and summed to obtain the recommendation score for that drill bit. Based on multiple recommended drill bits and their corresponding recommendation scores, drill bit recommendation results are generated in descending order of recommendation score. The recommendation results include the recommendation score value for each recommended drill bit and its corresponding detailed drilling engineering parameters.

[0132] In some embodiments, the above method may further include: collecting actual drilling performance feedback of the target well, quantifying the comprehensive reward value based on the feedback, and calculating the deviation of the current recommendation performance from the benchmark performance; attributing the contributions of formation information knowledge sources, historical record knowledge sources, drill bit technology knowledge sources, and driller demand knowledge sources, and calculating the contribution value of each knowledge source; the contribution value is jointly determined by the data quality of the knowledge source, its hit rate in the current recommendation, and the deviation; adjusting the fusion weight of each knowledge source using a gradient update method based on the difference between the contribution value of each knowledge source and the average contribution value, and truncating and normalizing the adjusted weights to obtain the updated fusion weights.

[0133] Specifically, actual drilling performance feedback for the target well is collected, including actual mechanical drilling speed, actual footage, cost per unit footage, wear level, safety incidents, and driller satisfaction. This on-site feedback is quantified into a comprehensive reward R: R = a × N(ROP) + b × N(M) - c × N(C) - d × Q + e × Sat; Where N() represents normalization, M is the actual footage advanced, C is the cost per unit footage advanced, Q is the risk deduction, and Sat is the driller's satisfaction; a, b, c, d, and e are the corresponding KPI weights. The deviation of the recommended performance from the baseline performance is calculated: Delta = R - R_0; Where R_0 is the historical benchmark value for similar operating conditions or the predicted value before the recommendation, and Delta is used to represent the magnitude of the increase or decrease of the current recommendation relative to the benchmark.

[0134] Contribution attribution is performed on four knowledge sources: formation information, historical cases, drill bit technical parameters, and driller requirements. g_i = q_i × s_i × Delta, i ∈ {F, H, B, P}; Where q_i represents the data quality or credibility of the i-th type of knowledge source, and s_i represents the hit rate or case support of the knowledge source in this recommendation.

[0135] When the on-site data is complete and there are no major safety anomalies requiring manual review, calculate the temporary weights: u_i = Clip(w_i + eta × (g_i - g_avg), w_min, w_max); Where w_i represents the weight of the i-th knowledge source before the update, eta is the learning rate, g_avg is the average contribution value of each knowledge source, w_min and w_max are the upper and lower bounds of the weights, and Clip represents the truncation function. The temporary weights are then normalized to obtain the updated fusion weights.

[0136] When data is missing, anomalies are severe, or feedback is not reliable, the system only marks the record in the database without directly adjusting the fusion weight. This mechanism ensures that weight adjustment no longer relies on manually set step sizes and penalty coefficients, but is entirely driven by on-site feedback data.

[0137] In some embodiments, the drill bit recommendation results include structured interpretable recommendation information for each recommended drill bit. This structured interpretable recommendation information includes: recommended drill bit model, comprehensive score, score composition, recommendation rationale, footage prediction, risk warning, confidence level description, and alternative solutions. The score composition includes formation matching degree and corresponding natural language description, historical case support degree and corresponding natural language description, demand matching degree and corresponding natural language description, and reliability rating and corresponding natural language description. The footage prediction includes a first footage prediction value, a second footage prediction value, and a third footage prediction value based on historical quantiles, as well as drill bit condition assessment suggestions. The risk warning includes failure mode risk warnings for the current formation and operation monitoring suggestions. The confidence level description includes the ratio of theoretical derivation weights to empirical case weights. The alternative solutions include at least one alternative drill bit model and a description of the differences between each alternative drill bit model and the recommended drill bit.

[0138] The overall score uses a normalized value from the recommendation score calculation. The score comprises formation matching degree, historical case support degree, demand matching degree, and reliability rating. Formation matching degree is derived from the contribution of the recommended drill bit in the theoretical or empirical confidence level. Historical case support degree is determined by combining the number and similarity of successful cases corresponding to the candidate drill bits. Demand matching degree is determined by the similarity between the engineering parameters of the recommended drill bit and the driller's demand vector. The reliability rating is derived from the reliability score in the drill bit's technical knowledge source. Each score is accompanied by explanatory text in natural language explaining the basis for the score.

[0139] The footage prediction is presented in the form of the 10th percentile, 50th percentile, and 90th percentile of the footage. Each percentile is obtained from the historical footage distribution of the drill bit in the same formation type in the drill bit technical knowledge source, and is accompanied by drill bit condition assessment suggestions, that is, the drill bit condition is assessed when the cumulative footage reaches the 50th percentile.

[0140] The risk warning is generated based on the failure mode probability distribution of the drill bit in similar formation types from the drill bit technical knowledge source. The mode with the highest failure probability is selected as the primary risk warning, along with monitoring recommendations. The confidence level description includes the overall confidence level value of this recommendation and the composition ratio of the confidence level sources. The respective weights of theoretical derivation and empirical verification are clearly marked in the confidence level source composition. The alternative solutions select 2 to 3 drill bit models with an overall score second only to the primary recommendation, and indicate the main performance differences between each alternative solution and the primary recommendation.

[0141] In some embodiments, prior to step S101, the method may further include: based on the formation information knowledge source and the drill bit technology knowledge source, generating multiple theoretical candidate drill bits and their corresponding theoretical confidence levels through the large language model according to a four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records, and output expectations.

[0142] In some embodiments, the above-mentioned generation of multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source, according to a four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records, and output expectations, through the large language model, may further include: extracting multiple geological entities and the correlation between these geological entities using a geological entity recognition model based on the geological logging report; the multiple geological entities include at least lithological entities, physical parameter entities, geological structure entities, and drilling characteristic entities; constructing a formation profile based on the multiple geological entities and the correlation between these geological entities; and constructing a formation information knowledge source based on the formation profile.

[0143] The geological entity recognition model can adopt a domain-adaptive pre-trained language model. After further pre-training on drilling geological corpus, the model is fine-tuned through corpus labeled with geological entity types and their relationships, and has the ability to identify drilling-specific entity types from unstructured geological logging reports.

[0144] The geological logging report includes lithological descriptions, physical property parameters, structural features, and records of drilling complexities, stored in the drilling database in natural language text format. The geological logging report is input into a geological entity recognition model, which extracts lithological entities, physical parameter entities, geological structural entities, and drilling characteristic entities. Lithological entities include rock type names such as limestone, sandstone, mudstone, flint, and dolomite. Physical parameter entities include quantitative parameter values ​​such as drillability level, abrasiveness index, porosity, and permeability. Geological structural entities include structural type names such as faults, fractures, interlayers, and unconformities. Drilling characteristic entities include drilling risk indicators such as lost circulation risk, collapse risk, and high-pressure zones.

[0145] The geological entity recognition model simultaneously extracts the relationships between various geological entities. These relationships include hierarchical relationships between lithology and physical parameters, induced relationships between formations and drilling risks, and inclusion relationships between depth ranges and lithological distribution. Based on the extracted geological entities and their relationships, a stratigraphic profile is constructed. This profile is presented in structured vector form, containing lithological feature components, physical property components, geological structure components, and drilling characteristic components. Each component is formed by concatenating the embedded vectors of the corresponding entities. A stratigraphic information knowledge source is then constructed based on the stratigraphic profile. This source stores the stratigraphic feature vectors corresponding to each depth range of the target well in key-value pairs for subsequent drill bit recommendation steps.

[0146] By automatically extracting standardized stratigraphic features from unstructured geological logging reports using a geological entity recognition model, the problem of geological data being scattered and stored in text form and difficult to be directly used for quantitative calculations is solved. The construction of stratigraphic profiles unifies heterogeneous geological descriptions into vectorized representations, providing a structured data foundation for stratigraphic information knowledge sources.

[0147] In some embodiments, the above-mentioned generation of multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source, according to a four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records and output expectations, through the large language model, may further include: constructing a historical record knowledge source based on the drill bit usage records of adjacent wells of the target well in the historical drilling database; each drilling case in the historical record knowledge source includes formation information, drill bit model, operation parameters and drilling results of the well.

[0148] The historical drilling database stores drill bit usage records for adjacent wells. Each record includes formation information, drill bit model, operating parameters, and drilling performance data collected during the drilling process. Drill bit usage records for adjacent wells of the target well are extracted from the historical drilling database, with each record constituting a drilling case. For each drilling case, formation information is extracted from the corresponding logging data, including the main lithology, drillability grade, abrasiveness index, and depth range of the well section. Drill bit model, drill bit type, drill bit size, and drill bit manufacturer information are extracted from the drill bit usage records for that case as the drill bit model field. Drilling parameters such as pressure on bit, rotational speed, displacement, and rate of penetration are extracted from the drilling engineering parameter records for that case as operating parameter fields. Actual footage, drill bit wear rating, and whether any downhole accidents occurred are extracted from the completion report for that case as drilling performance fields.

[0149] Formation information, drill bit type, operating parameters, and drilling results are integrated into structured historical records. Multiple historical records constitute a historical knowledge source. The historical knowledge source is indexed by formation fingerprints, which are formed by combining basin code, lithology class, drillability interval, and depth interval to support rapid location during subsequent similar case searches.

[0150] By extracting and integrating four types of information—formation, drill bit, parameters, and effects—from historical drilling databases to construct a structured historical knowledge source, drilling records scattered across different data tables are unified into a case format that can be searched based on similarity. Each case contains complete decision-making elements and result feedback, providing a standardized case library foundation for empirical verification paths.

[0151] In some embodiments, the above-mentioned generation of multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source, according to a four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records, and output expectations, through the large language model, may further include: obtaining multiple drill bit specification parameters according to the drill bit technical manual; the multiple drill bit specification parameters include drill bit type, cutting tooth size, tooth arrangement method, IADC classification code, and recommended operation parameter range; obtaining multiple drill bit performance parameters according to historical drill bit usage data; the multiple drill bit performance parameters include average mechanical drilling speed, footage quantile, and failure mode probability distribution; and constructing a drill bit technology knowledge source based on the multiple drill bit specification parameters and the multiple drill bit performance parameters.

[0152] Drill bit technical manuals contain specifications for various types of drill bits. From these manuals, you can obtain information such as drill bit type, cutting tooth size, tooth arrangement, IADC classification code, and recommended operating parameter ranges. This information constitutes the static parameters section of the drill bit. Drill bit types include PDC bits, roller cone bits, and diamond bits. Cutting tooth size is recorded in millimeters or inches. Tooth arrangement includes the number of blades, tooth density, and tooth size. IADC classification codes are based on the International Association of Drilling Contractors (IAD) standards, which classify and encode drill bit structural characteristics. Recommended operating parameter ranges include the manufacturer-recommended pressure on bit (PBT) and rotational speed (RPM) ranges.

[0153] Historical drill bit usage data is derived from actual usage records in the historical drilling database. The average mechanical rate of penetration (MRP), percentiles of footage, and probability distributions of failure modes are statistically analyzed for each drill bit model, categorized by formation type. The average MRP is calculated as the arithmetic mean of multiple usage records within the same formation type. Percentiles of footage include the 10th, 50th, and 90th percentiles, representing the low, median, and high values ​​of footage for that drill bit in the same formation, respectively. The probability distribution of failure modes is obtained based on the frequency of occurrence of various failure modes during historical use. Failure modes include tooth breakage, mud packing, torsion breakage, and premature wear.

[0154] A static parameter library for drill bits is constructed based on multiple drill bit specification parameters, and a dynamic profile of drill bits is constructed based on multiple drill bit performance parameters. The static parameter library and the dynamic profile of drill bits together constitute a source of drill bit technical knowledge. The dynamic profile of drill bits stores performance statistics data of each drill bit model in partitions according to formation type.

[0155] By integrating static specifications from drill bit technical manuals with dynamic performance parameters from historical usage data, a dual-dimensional drill bit technical knowledge source is constructed. Static parameters provide manufacturers with prior knowledge, while dynamic profiles provide performance statistics based on field verification. The footage quantiles in the dynamic profiles replace single-point prediction values, and the failure mode probability distribution provides data support for risk warnings, making the drill bit technical knowledge source both theoretically complete and engineering practical.

[0156] In some embodiments, the above-mentioned generation of multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source, according to a four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records, and output expectations, through the large language model, may further include: identifying the recommendation intent type and slot information based on the operation requirement information input by the driller through a natural language understanding model; the slot information is used to characterize the constraints of the target well drilling operation; generating driller requirement information based on multi-turn dialogues initiated for missing mandatory slots; and constructing a driller requirement knowledge source based on the driller requirement information.

[0157] The natural language understanding model employs a pre-trained language model fine-tuned with drilling-related corpora, enabling it to identify drilling-specific intentions and slot locations from unstructured text. The driller inputs operational requirements in natural language text form through the interactive interface, including descriptive statements such as "Recommend a drill bit for a directional well, with drilling speed as the priority" and "The drill bit is damaged and needs to be replaced."

[0158] The driller's input of operational requirements is fed into a natural language understanding model, which then identifies the recommended intent type and slot information. Recommendation intent types include initial recommendation, benchmarking optimization, fault replacement, cost accounting-oriented, and risk avoidance-oriented. Each intent type corresponds to a set of trigger words and required slots. Trigger words for initial recommendation include "recommend," "select," and "which drill bit to use," while required slots include formation description, well type, and target depth. Trigger words for benchmarking optimization include "can it be faster than the previous well?" and "is there a better one?", while required slots include reference well number or drill bit model and optimization direction. Trigger words for fault replacement include "drill bit failed," "early tripping," and "replace with a more durable one," while required slots include the cause or phenomenon of failure and description of remaining well sections.

[0159] Slot information is used to characterize the constraints of drilling operations at the target well, including specific constraints such as formation description, well type, target depth, cost constraints, drilling speed requirements, and special operating conditions. When a required slot is missing, a multi-turn dialogue is initiated through the interactive interface to proactively inquire about the missing required slot information from the driller. Optional slots are not inquired about if the recommended conditions are met. A driller requirement knowledge source is constructed based on the collected complete driller requirement information. This knowledge source is stored in the form of standardized requirement vectors, which contain weighted requirements for each performance indicator. The weights range from 0 to 1, and the sum of all weights is 1.

[0160] By using a natural language understanding model to identify the intent and fill in the slots of the text instructions input by the driller, unstructured work requirements are transformed into structured requirement vectors. A multi-turn dialogue mechanism completes the necessary information without disturbing the driller, enabling the driller's knowledge source to accurately reflect the differentiated goals on the work site and providing a weighting basis for subsequent KPI-oriented ranking.

[0161] In some embodiments, the above method may further include: adjusting the confidence levels of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source, and driller demand knowledge source according to a first cycle based on drill bit feedback data from the target well; updating the historical record knowledge source according to a second cycle based on drill bit feedback data from the target well; and fine-tuning the parameters of the large language model according to a third cycle based on drill bit feedback data from the target well; wherein the first cycle is shorter than the second cycle, and the second cycle is shorter than the third cycle.

[0162] Drill bit feedback data can be information on the actual drilling performance collected after the target well is completed, including the average actual mechanical drilling rate, actual footage, IADC wear rating when the drill bit is pulled out, whether recommended failure modes without warning occurred, and the driller's satisfaction score with the recommended results.

[0163] The first cycle can be set to be triggered after each well is completed. Based on the drill bit feedback data, the confidence of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source and driller demand knowledge source is adjusted. The confidence adjustment adopts a rule-based weight update method, and a small incremental adjustment is made according to the contribution accuracy of each knowledge source in this recommendation. After the adjustment, the sum of the confidence of all knowledge sources is normalized to 1.

[0164] The second cycle can be set as a condition-triggered update, which expands the historical knowledge source based on drill bit feedback data accumulated from multiple wells. The update content includes adding new drilling cases to the database, updating statistics in the drill bit dynamic profile, and adjusting the statistical parameters of formation zoning.

[0165] The third cycle can be set as periodic or conditionally triggered model fine-tuning, which efficiently fine-tunes the parameters of the large language model based on accumulated drill bit feedback data, updating the model's ability to recognize the formation-drill bit matching relationship. The three cycles correspond to real-time weight adjustment, periodic knowledge base expansion, and periodic model fine-tuning, respectively, forming a multi-timescale incremental update system from minutes to days.

[0166] By constructing a three-layer incremental update mechanism, continuous optimization can be achieved at different time scales based on field feedback data. Real-time weight adjustment responds to single-well feedback to achieve rapid correction. Regular knowledge base expansion accumulates multi-well data to achieve stable statistical updates. Periodic model fine-tuning enables the evolution of deep cognitive capabilities. The collaborative work of the three-layer mechanism solves the problem of the decline in recommendation quality over time in the existing system.

[0167] In some embodiments, the above-mentioned adjustment of the confidence levels of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source, and driller demand knowledge source based on the drill bit feedback data of the target well in a first cycle may further include: after the completion of each well, collecting the actual mechanical drilling rate, actual footage, drill bit wear rating, failure mode, and driller satisfaction of the well as drill bit feedback data; determining the contribution accuracy of each of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source, and driller demand knowledge source based on the drill bit feedback data; adjusting the confidence levels of each knowledge source based on the contribution accuracy, and implementing the following constraints: the confidence level of the formation information knowledge source is not lower than a preset lower limit; when the number of similar historical cases is lower than a preset threshold, the weight increase of the historical record knowledge source is reduced proportionally to the number of cases; when the feedback data comes from a block with a sample size lower than a preset threshold, the adjustment of all weights is reduced proportionally.

[0168] Based on drill bit feedback data, the contribution accuracy of each of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source, and driller's needs knowledge source is determined. One method for evaluating contribution accuracy is as follows: For the historical record knowledge source, the main adjacent well cases referenced in this recommendation are identified, the relative deviation between the predicted mechanical drilling rate and the actual mechanical drilling rate of these cases is calculated, and the contribution accuracy is obtained by subtracting this deviation from 1; for the formation information knowledge source, the degree of agreement between the actual formation drillability level and the preset level during drilling is evaluated, and the contribution accuracy is quantified based on the degree of agreement; for the drill bit technology knowledge source and the driller's needs knowledge source, a similar method is used to calculate the contribution accuracy based on the deviation between the actual and expected results.

[0169] Based on the calculated contribution accuracy of each knowledge source, the confidence level of each knowledge source is adjusted. The adjustment rule is that when the contribution accuracy of a knowledge source is higher than the high threshold, the confidence level of the knowledge source is slightly increased, and when the contribution accuracy is lower than the low threshold, the confidence level of the knowledge source is slightly decreased. After adjustment, the confidence levels of all knowledge sources are renormalized.

[0170] Three constraints are implemented during the adjustment process: the confidence level of the stratigraphic information knowledge source is not lower than the preset lower limit to ensure that the fundamental role of stratigraphic data in drill bit selection is not weakened; when the number of similar historical cases is lower than the preset threshold, the weight adjustment of the historical knowledge source is reduced proportionally to the number of cases to prevent over-reliance on historical paths in data-sparse scenarios; when the feedback data comes from blocks with a sample size lower than the preset threshold, all weight adjustments are reduced proportionally to prevent a single abnormal data point in a special block from having an excessive impact on the weight of the entire system.

[0171] The confidence level of knowledge sources is quantitatively adjusted by evaluating the accuracy of contributions, so that the fusion weights can be continuously optimized based on the actual performance of each knowledge source in historical recommendations. The three constraints prevent the weight adjustments from deviating from the rationality of the project from three dimensions: basic layer protection, data sparsity penalty, and cross-block anomaly protection, thus ensuring the stability and security of incremental updates.

[0172] In some embodiments, updating the historical knowledge source based on the drill bit feedback data of the target well according to the second cycle may further include: triggering a knowledge base update when a major failure is detected, the cumulative number of new formation types reaches a first preset number of wells, the cumulative number of new data for existing formation types reaches a second preset number of wells, or the number of days since the last update exceeds a preset number; managing newly generated drilling cases by formation fingerprint, wherein the formation fingerprint is determined according to basin code, lithology class, drillability interval, and depth interval; calculating the formation fingerprint similarity between new cases and existing cases, and merging and updating the statistics of existing partitions when the similarity is higher than a first preset threshold; adding new cases to the database when the similarity is between the first preset threshold and the second preset threshold; creating new partitions when the similarity is lower than the second preset threshold; and updating the mean mechanical drilling rate, footage quantile, and failure mode probability distribution of each drill bit in the drill bit technology knowledge source under each formation type based on the newly added cases.

[0173] The knowledge base update can be triggered by a combination of multiple conditions. An update is triggered when a major failure occurs, the cumulative number of new formation types reaches a first preset well count, the cumulative number of new data for existing formation types reaches a second preset well count, or the number of days since the last update exceeds a preset limit. Major failures can be events such as drill bit chipping, mud buildup, breakage, or premature wear leading to abnormal tripping; these events trigger an update immediately upon occurrence. The update is triggered when the cumulative number of new formation types reaches a first preset well count (e.g., 5 wells), allowing for updates with minimal data to prioritize covering unexplored formation types. The update is triggered when the cumulative number of new data for existing formation types reaches a second preset well count (e.g., 20 wells), allowing sufficient data accumulation before updating to avoid noise interference. A fallback update is triggered when the number of days since the last update exceeds a preset limit (e.g., 30 days), ensuring the knowledge base remains up-to-date.

[0174] After an update is triggered, newly generated drilling cases are managed by formation fingerprints. A formation fingerprint is formed by combining four elements: basin code, lithology class, drillability range, and depth range, serving as the formation identifier for each case. The similarity between the formation fingerprints of new and existing cases is calculated, weighted based on the matching degree of the four elements. When the similarity is higher than a first preset threshold, the new case is considered highly similar to the existing partition; no new case entry is added, but only the statistics of the existing partition are merged and updated, including the mean mechanical drilling rate, footage quantiles, and failure mode probability distribution. When the similarity is between the first and second preset thresholds, the new case is considered moderately similar to the existing partition and is added to the database as a new case, assigned to the corresponding formation fingerprint partition. When the similarity is lower than the second preset threshold, the new case is considered dissimilar to all existing partitions, and a new partition is created. Based on the drill bit usage data of newly added cases, the mean mechanical drilling speed, footage quantiles, and failure mode probability distribution of each drill bit in the drill bit technical knowledge source are updated under different formation types. The update method adopts incremental statistical method, which integrates new data on the basis of the original statistics.

[0175] The formation fingerprint partitioning management mechanism enables the refined expansion of historical knowledge sources. Similarity judgment ensures that cases are reasonably classified when entering the database to avoid data redundancy. Different processing methods for different similarity intervals keep the knowledge base structure clear. The incremental updates of drill bit dynamic profiles enable performance statistics to be continuously refined with the accumulation of data, providing richer and more reliable case support for experience verification paths.

[0176] In some embodiments, the above-mentioned parameter fine-tuning of the large language model based on the drill bit feedback data of the target well in a third cycle includes: triggering model fine-tuning when the number of days since the last fine-tuning exceeds a first preset number of days, the recommendation adoption rate is lower than a preset adoption rate threshold for a second consecutive preset number of days, the mechanical drilling rate prediction error exceeds a preset error threshold for a second consecutive preset number of days, or a new block is started; freezing the main parameters of the large language model and updating some trainable parameters; constructing a fine-tuning training dataset, wherein the fine-tuning training dataset is composed of newly accumulated data, historical data playback, and expert theoretical knowledge mixed in a preset ratio; and fine-tuning the large language model based on the fine-tuning training dataset.

[0177] The triggering conditions for fine-tuning the large language model can adopt a multi-condition combination mechanism. When the number of days since the last fine-tuning exceeds the first preset number of days, the recommendation adoption rate is lower than the preset adoption rate threshold for the second consecutive preset number of days, the mechanical drilling speed prediction error exceeds the preset error threshold for the second consecutive preset number of days, or drilling begins in a new block, the model fine-tuning operation is triggered.

[0178] The first preset number of days can be set, for example, to 90 days, as a baseline for periodic updates. The recommendation adoption rate threshold is set as the lower limit calibrated by the system based on historical performance; consistently falling below this threshold indicates a decline in recommendation quality. The mechanical drilling rate prediction error threshold is set as the upper limit calibrated by the system based on historical performance; consistently exceeding this threshold indicates a decline in prediction accuracy. Drilling in a new block triggers fine-tuning, enabling the model to quickly adapt to new formation characteristics. After fine-tuning is triggered, the main parameters of the large language model are frozen, and only some trainable parameters are updated. Efficient parameter fine-tuning methods such as LoRA or Adapter are used to learn patterns in the new data while ensuring that the model's core knowledge is not overwritten.

[0179] A fine-tuning training dataset can be constructed, consisting of newly accumulated data, historical data replay, and expert theoretical knowledge mixed in a preset ratio. Newly accumulated data accounts for 30% to learn the patterns of new formations and new drill bits; historical data replay accounts for 50% to prevent forgetting existing knowledge; and expert theoretical knowledge accounts for 20% to maintain the correctness of basic drilling theory and physical laws. Based on this constructed fine-tuning training dataset, the large language model is fine-tuned, with the learning rate set to one-tenth of the initial fine-tuning stage to prevent overfitting on a small amount of new data.

[0180] The multi-condition triggering mechanism ensures that model fine-tuning is performed in a timely manner when recommendation quality declines or new blocks appear. The efficient parameter fine-tuning method prevents catastrophic forgetting while updating the model's capabilities. The three-source mixing ratio of training data enables the model to simultaneously absorb new experience, consolidate old knowledge, and maintain theoretical correctness, thus realizing the continuous evolution of the large language model in the field of drilling engineering.

[0181] In some embodiments, the above method may further include: in the zero-data stage, drill bit recommendation is performed primarily based on theoretically recommended paths and secondarily on empirically recommended paths, wherein the weight of theoretically recommended paths is higher than that of empirically recommended paths. The theoretically recommended paths generate candidate drill bits based on drilling theory knowledge of large language models and migration experience of geologically similar blocks, while the empirically recommended paths generate candidate drill bits based on historical cases of geologically similar blocks. The zero-data stage is used to characterize that the number of completed wells in the target block where the target well is located is less than a first preset number of wells.

[0182] The zero-data phase is used to characterize scenarios where the number of completed wells in the target block where the target well is located is less than a first preset number of wells. The first preset number of wells can be set to, for example, 3 wells. In this phase, there is no or only a very small amount of historical drilling data available for reference within the target block. In the zero-data phase, a recommendation configuration is adopted that primarily uses theoretically recommended paths and secondarily uses empirically recommended paths. The weight of the theoretically recommended path can be set to, for example, 0.9, and the weight of the empirically recommended path can be set to, for example, 0.1.

[0183] The theoretical recommendation path generates candidate drill bits based on drilling theory knowledge from a large language model and migration experience from geologically similar blocks. The drilling theory knowledge of the large language model comes from the reasoning ability of the large language model, which has been fine-tuned by the FSOE quadruple instruction format, to match formation conditions with drill bit technical parameters. The migration experience from geologically similar blocks comes from historical successful cases in blocks with similar geological conditions selected based on formation similarity. These cases are weighted by similarity and used as supplementary basis for theoretical reasoning.

[0184] The experience-based recommendation path generates candidate drill bits based on historical cases from geologically similar blocks. By calculating the stratigraphic similarity between the target block and known blocks, blocks with similarity scores higher than a preset threshold are selected. Successful cases are retrieved from the historical knowledge sources of these blocks, and the drill bit models used in those cases are extracted as experience-based candidate drill bits. The recommendation results generated in the zero-data phase explicitly indicate insufficient data for new blocks and prioritize recommending drill bit models already in stock for the current block and validated in geologically similar blocks. The candidate drill bit set is automatically filtered to remove drill bits with reliability scores lower than the system's preset threshold.

[0185] By configuring weights in the zero-data phase and utilizing cross-block migration experience, the system can still provide theoretically based drill bit recommendations even under extreme conditions where there is no historical data for the target block. Migration experience from geologically similar blocks fills the gaps in the data of this block, and conservative recommendation strategies and clear labeling ensure the safety of exploration operations in new blocks.

[0186] In some embodiments, the above method may further include: during the data accumulation phase, as the number of completed wells increases, the weight of the theoretically recommended path is reduced and the weight of the empirically recommended path is gradually increased until the weights of the two paths transition to a preset ratio; the data accumulation phase is used to characterize that the number of completed wells in the target block where the target well is located is greater than or equal to a first preset number of wells and less than a second preset number of wells.

[0187] The data accumulation stage is used to characterize the scenario where the number of completed wells in the target block where the target well is located is greater than or equal to the first preset number of wells and less than the second preset number of wells. The first preset number of wells can be set to 3 wells for example, and the second preset number of wells can be set to 15 wells for example. In this stage, the target block has accumulated a certain amount of drilling data but has not yet reached the mature data volume.

[0188] During the data accumulation phase, as the number of completed wells increases, the weight of the theoretically recommended path is reduced while the weight of the empirically recommended path is gradually increased until the weights of both paths transition to a preset ratio. The weight adjustment adopts a linear, gradual approach: the weight of the theoretically recommended path decreases from zero data to a preset target value as the number of completed wells increases, while the weight of the empirically recommended path increases from zero data to a preset target value. The preset target value is set as the weight ratio under normal dual-path recommendation mode, and the sum of the weights of the theoretically recommended path and the empirically recommended path is always 1.

[0189] The asymptotic coefficient is dynamically determined based on data quality assessment. The asymptotic speed is accelerated when the data quality of the current block is high, and slowed down when the data quality is low. After each new well is completed, the drilling data of that well immediately triggers the second-level knowledge base update. The well case is partitioned and added to the database according to the formation fingerprint, and the statistics in the drill bit dynamic profile are updated to accelerate the accumulation of knowledge in the current block.

[0190] By gradually adjusting the weights of theoretical and empirical paths as the number of completed wells increases, a smooth transition from the cold start phase to the normal recommendation mode is achieved. This avoids fluctuations in recommendation quality caused by prematurely switching to the dual-path weighted mode when the amount of data is insufficient. The dynamic gradual coefficient ensures that the transition rate can adapt to the actual data quality, guaranteeing the stability of the transition process.

[0191] In some embodiments, the above method may further include: switching to a dual-path recommendation mode during the data maturity stage; the data maturity stage is used to characterize that the number of completed wells in the target block where the target well is located is greater than or equal to the second preset number of wells and meets the preset switching index.

[0192] The data maturity stage is used to characterize a scenario where the number of completed wells in the target block where the target well is located is greater than or equal to the second preset number of wells and meets the preset switching index. The second preset number of wells can be set to, for example, 15 wells. The preset switching index is used to ensure that the data in this block reaches a level that can support normal dual-path recommendation in terms of both quantity and quality.

[0193] During the data maturation stage, the system switches to a dual-path recommendation mode, removing the additional restrictions of the cold start stage. The theoretical recommendation path and the empirical recommendation path operate in the normal dual-path mode. Both paths generate candidate drill bits and evaluate confidence based on historical data accumulated in this block, no longer relying on migration experience from geologically similar blocks as the main reference.

[0194] After switching to the dual-path recommendation mode, the system simultaneously initiates a three-layer incremental update mechanism: real-time weight adjustment, periodic knowledge base expansion, and periodic model fine-tuning, continuously optimizing system performance based on the data in this block. Once the data maturity phase is complete, recommendations for other wells within the same block will directly run in the dual-path recommendation mode, eliminating the need for a repeated cold start process.

[0195] By setting both the number of wells drilled and preset switching indicators as the entry criteria for the data maturity stage, the system ensures that the data in this block has reached a sufficient quantity and quality before switching to the normal recommendation mode. After the switch is completed, the system can make full use of the historical data accumulated in this block for accurate recommendations, laying the data foundation for the subsequent launch of the three-layer incremental update mechanism.

[0196] In some embodiments, the preset switching indicators include: the proportion of the number of formation types covered by drilled wells to the total number of predicted formation types in the target block exceeds a preset coverage threshold; the average recommendation adoption rate of the most recent wells exceeds a preset adoption rate threshold; and the standard deviation of the mechanical drilling rate prediction error of the most recent wells is lower than a preset error threshold.

[0197] The preset switching criteria are used to determine whether a target block meets the conditions for switching from the data accumulation stage to the data maturity stage. All three criteria must be met simultaneously to trigger the stage switch. The first criterion is the formation type coverage criterion, which calculates the proportion of formation types covered by drilled wells to the total predicted formation types in the target block. When this proportion exceeds a preset coverage threshold, the switch is considered successful. For example, the preset coverage threshold can be set to 60%. This criterion ensures that the main formation types in this block have representative cases in historical data, avoiding a lack of experience support for key formation types that have not yet been touched. The second criterion is the recommendation adoption rate trend criterion, which calculates the average recommendation adoption rate of the most recent multiple wells. When this average exceeds a preset adoption rate threshold, the switch is considered successful. For example, the number of the most recent multiple wells can be set to 5. The preset adoption rate threshold is set according to the system's benchmark requirements for recommendation quality. This criterion ensures that the system's recommendation results in this block have been fully recognized by the field driller. The third indicator is the prediction error convergence indicator. It calculates the standard deviation of the mechanical drilling rate prediction error for the most recent multiple wells. If this standard deviation is lower than a preset error threshold, the system is considered satisfied. For example, the number of recent multiple wells can be set to 5. The prediction error standard deviation reflects the stability of the recommendation system's mechanical drilling rate prediction. A standard deviation below the threshold indicates that the system's understanding of the formation conditions in that block has become stable. If any of the three indicators is not satisfied, the system delays the switch, prompting the user with the specific unsatisfied indicator and the reason, until all three indicators are satisfied before performing the stage switch.

[0198] By setting three quantitative indicators—stratum type coverage, recommendation adoption rate trend, and prediction error convergence—as joint conditions for phase switching, the transition from the cold start phase to the normal recommendation mode has a clear quantitative basis. The coverage indicator ensures the breadth of knowledge, the adoption rate indicator ensures user acceptance, and the prediction error stability indicator ensures prediction reliability. Together, these three indicators ensure the sufficiency and rationality of the switching timing.

[0199] In some embodiments, the fine-tuning of the aforementioned large language model employs a four-level drilling corpus pyramid to construct a training dataset. This four-level drilling corpus pyramid includes: a basic theoretical corpus comprising drill bit typology technical manuals, formation lithology description specifications, and drilling engineering industry standards; a domain knowledge graph corpus comprising triplets extracted from historical drilling reports, consisting of formation conditions, drill bit selection, and drilling results, as well as counterfactual reasoning corpus of failure cases; a driller experience knowledge corpus comprising transcripts of expert interviews, drill bit evaluation records in daily field reports, and drill bit usage experiences in drilling team reports; and a cross-block transfer corpus comprising formation and drill bit adaptation cases from different regions and basins. Based on this four-level drilling corpus pyramid, the large language model is further pre-trained and subjected to supervised fine-tuning.

[0200] The four-level drilling corpus pyramid is used to construct the training dataset needed for fine-tuning large language models. The four levels correspond to different types of drilling domain knowledge sources.

[0201] The basic theoretical corpus includes the Drill Bit Typology Technical Manual, Formation Lithology Description Specifications, and Drilling Engineering Industry Standards. The Drill Bit Typology Technical Manual records the structural characteristics, applicable formation ranges, and recommended operating parameter values ​​for various types of drill bits. The Formation Lithology Description Specifications include lithology classification systems, drillability level classification standards, and abrasiveness index determination methods. The Drilling Engineering Industry Standards include API standards and oil and gas industry specifications. The basic theoretical corpus provides a foundational knowledge framework for drilling engineering for the large language model.

[0202] The domain knowledge graph corpus includes triplets composed of formation conditions, drill bit selection, and drilling results extracted from historical drilling reports, as well as counterfactual reasoning corpus of failure cases. The triplets present the correlation between formation characteristics, selected drill bit model, and actual drilling results in a structured form of "formation conditions-drill bit selection-drilling results". The counterfactual reasoning corpus adopts the descriptive format of "using drill bit Y under formation conditions X, failure W due to reason Z, drill bit V should be used instead", enabling the model to learn the correct decision path in failure scenarios.

[0203] The drilling experience and knowledge corpus includes transcripts of expert interviews, drill bit evaluation records in the daily field reports, and drill bit usage experiences in the drilling team's shift reports. This corpus records the practical experience and subjective judgments of field operators in unstructured natural language, covering field operation skills and coping strategies not recorded in theoretical manuals.

[0204] The cross-block migration corpus contains stratigraphic and drill bit adaptation cases from different regions and basins, covering drill bit usage records for multiple geological structural areas and various stratigraphic types, which is used to enhance the model's generalization ability under different geological conditions.

[0205] Based on the four-level drilling corpus pyramid, the large language model is further pre-trained and supervised fine-tuned. In the pre-training stage, basic theoretical corpus and domain knowledge graph corpus are used to adapt the model to the terminology system and knowledge structure of the drilling field. In the supervised fine-tuning stage, driller experience knowledge corpus and cross-block transfer corpus are used to enable the model to learn specific recommendation decision patterns.

[0206] By constructing a hierarchical, four-layered drilling corpus pyramid, the large language model can acquire drilling domain knowledge at different levels. Basic theoretical corpus lays the foundation of knowledge, domain knowledge graph corpus constructs structured knowledge connections, driller experience knowledge corpus integrates on-site practical wisdom, and cross-block transfer corpus enhances generalization ability. The synergistic effect of the four layers of corpus enables the model to possess the complete knowledge system required for drill bit recommendation.

[0207] In some embodiments, the method further includes: constructing a failure case library, wherein the failure case library extracts drill bit failure records from historical data, and each failure record is labeled with the failure cause type, formation conditions at the time of failure, and usage parameters; constructing positive and negative comparison training samples based on the failure case library, wherein the positive and negative comparison training samples include positive examples and negative examples, wherein the input of the positive examples is formation conditions, operating scenario, and adjacent well information, and the output is a recommended drill bit, with the actual result being success; the input of the negative examples is formation conditions, operating scenario, and adjacent well information, and the output is a recommended drill bit, with the actual result being failure, and additional annotations of failure cause analysis and the drill bit to be used; during the training of the large language model, the prediction accuracy of failure warning is given a higher weight than other prediction tasks.

[0208] The failure case database extracts drill bit failure records from historical drilling data. Each failure record includes the formation conditions at the time of failure, operating parameters, and the type of failure cause. Failure cause types include tooth breakage, mud packing, torsion breakage, and premature wear. Formation conditions include lithology type, drillability grade, abrasiveness index, and formation pressure. Operating parameters include drilling pressure, rotation speed, displacement, and drilling time.

[0209] A positive and negative comparison training sample is constructed based on a failure case database. Positive examples take formation conditions, operational scenario, and adjacent well information as input, and output a recommended drill bit. The actual result is success, defined as achieving the expected rate of drilling, reaching the target footage, having an acceptable drill bit wear rating, and no downhole accidents. Negative examples take formation conditions, operational scenario, and adjacent well information as input, and output a recommended drill bit. The actual result is failure. Negative examples are derived from real failure records in the failure case database, not randomly generated. Each negative example includes a failure cause analysis and a recommended drill bit. The failure cause analysis describes the specific mechanism leading to the failure, and the recommended drill bit indicates the correct drill bit model recommended under those formation conditions.

[0210] During the training of the large language model, the prediction accuracy of failure warning is given a higher weight than that of other prediction tasks. Specifically, this is achieved by multiplying the error term of the failure warning-related output in the loss function by an amplification factor, so that the model pays more attention to the ability to identify recommended combinations that may fail during training.

[0211] By constructing positive and negative contrast training samples based on real failure cases, the large language model can identify seemingly reasonable but actually ineffective recommended combinations under specific geological conditions. The high weighting of failure warnings reflects the principle of prioritizing drilling engineering safety. The negative samples are derived from real data rather than being randomly generated, ensuring the engineering authenticity of the training signals and enabling the model to have the inherent ability to avoid high-risk recommendations.

[0212] In some embodiments, the method further includes: calculating the formation similarity between the target block where the target well is located and the source block, wherein the formation similarity is determined based on lithological type matching degree, depth interval overlap degree, and geological structure matching degree; determining the credibility weight of the drilling data of the source block in fine-tuning based on the formation similarity, wherein when the formation similarity is higher than a first preset threshold, the credibility weight is a first weight value; when the formation similarity is between the first preset threshold and a second preset threshold, the credibility weight is a second weight value; when the formation similarity is lower than the second preset threshold, the credibility weight is a third weight value, wherein the first weight value is greater than the second weight value, and the second weight value is greater than the third weight value; and weighting and mixing the drilling data of the source block and the drilling data of the target block according to the credibility weight to construct a fine-tuning training dataset.

[0213] The lithological type matching degree is calculated based on the classification distance of the main lithological types between the target block and the source block. The depth interval overlap degree is calculated based on the overlap ratio between the target depth interval of the target well and the historical drilling depth interval of the source block. The geological structure matching degree is calculated based on the similarity of structural features such as fault development degree, fold morphology and stratum dip angle.

[0214] The reliability weight of drilling data from source blocks in fine-tuning is determined based on formation similarity. When the similarity is higher than a first preset threshold, the reliability weight is the first weight value, which can be set to, for example, 0.8 and 0.7. When the similarity is between the first and second preset thresholds, the reliability weight is the second weight value, which can be set to, for example, 0.5 and 0.4. When the similarity is lower than the second preset threshold, the reliability weight is the third weight value, which can be set to, for example, 0.1. The first weight value is greater than the second weight value, and the second weight value is greater than the third weight value. Data from high-similarity blocks contributes more training weight in fine-tuning, while data from low-similarity blocks contributes less weight. Based on the credibility weight, the drilling data of the source block and the drilling data of the target block are weighted and mixed to construct a fine-tuning training dataset. During the training process, cases of source blocks with high similarity are sampled at a higher frequency, while cases of source blocks with low similarity are sampled at a lower frequency. This allows the model to reasonably refer to the experience of neighboring blocks when learning, and to rely on the data of the block itself or theoretical derivation when there are significant differences in the formation.

[0215] By determining the credible weights of cross-block data based on stratigraphic similarity, the model fine-tuning can selectively utilize the empirical knowledge of blocks with different similarities. The high weights of high-similarity blocks enable the model to effectively transfer successful experiences from areas with similar geological conditions, while the low weights of low-similarity blocks prevent cases from unrelated areas from interfering with the model's judgment, thus solving the problem of insufficient training samples in the scenario of sparse new block data.

[0216] In some embodiments, the method further includes: determining multiple recommended drill bits based on the recommendation ranking results of each recommended drill bit; generating structured interpretable recommendation information for each recommended drill bit, wherein the structured interpretable recommendation information includes: recommended drill bit model, comprehensive score, score composition, recommendation reason, footage prediction, risk warning, confidence level description, and alternative solutions; wherein, the score composition includes formation matching degree and corresponding natural language description, historical case support degree and corresponding natural language description, demand matching degree and corresponding natural language description, and reliability rating and corresponding natural language description; the footage prediction includes a first footage prediction value, a second footage prediction value, and a third footage prediction value based on historical quantiles, as well as drill bit condition assessment suggestions; the risk warning includes failure mode risk warnings for the current formation and operation monitoring suggestions; the confidence level description includes the ratio information of theoretical derivation weight and empirical case weight; the alternative solutions include at least one alternative drill bit model and a description of the differences between each alternative drill bit model and the recommended drill bit; and generating drill bit recommendation results based on the multiple recommended drill bits and the structured interpretable recommendation information corresponding to each recommended drill bit.

[0217] Based on the recommendation ranking of each drill bit, multiple recommended drill bits are determined. The recommendation ranking is based on a KPI-driven comprehensive score, arranged from highest to lowest, with the drill bit having the highest comprehensive score being the first-choice recommendation. For each recommended drill bit, structured and interpretable recommendation information is generated. This information includes the recommended drill bit model, comprehensive score, score composition, recommendation rationale, footage prediction, risk warning, confidence level explanation, and alternative solutions. The recommended drill bit model is output in a standard naming format, including drill bit type, number of blades, cutting tooth characteristics, and dimensions. The comprehensive score is a normalized value of the recommendation calculation result, ranging from 0 to 1.

[0218] The scoring system comprises formation matching degree and corresponding natural language description, historical case support degree and corresponding natural language description, demand matching degree and corresponding natural language description, and reliability rating and corresponding natural language description. Each formation matching degree is derived from the degree of matching between the drill bit and formation conditions in the theoretical confidence level. The historical case support degree is determined by combining the number and similarity of successful cases corresponding to experienced candidate drill bits. The demand matching degree is determined by the similarity between the engineering parameters of the recommended drill bit and the driller's demand vector. The reliability rating is derived from the reliability score in the drill bit technical knowledge source. Each score is accompanied by an explanation of the basis for the score in natural language.

[0219] The footage prediction includes first, second, and third footage predictions based on historical quantiles, along with drill bit condition assessment recommendations. The first footage prediction is the 10th quantile, the second the 50th quantile, and the third the 90th quantile. The drill bit condition assessment recommendation is to evaluate the drill bit condition when the cumulative footage reaches the second footage prediction value. Risk warnings include failure mode risk warnings for the current formation and operational monitoring recommendations. Failure mode risk warnings are based on the failure mode probability distribution of the drill bit in similar formations from the drill bit's technical knowledge source, selecting the failure mode with the highest probability for warning. Operational monitoring recommendations provide specific monitoring parameters and operational suggestions for this failure mode. Confidence level information includes the weighting of theoretical derivation and empirical case studies, clarifying the respective contributions of theoretical reasoning and historical cases in this recommendation. Alternative options include at least one alternative drill bit model and a description of the differences between each alternative drill bit model and the recommended drill bit, comparing differences in drilling speed, wear resistance, and applicable scenarios. Based on multiple recommended drill bits and the structured, interpretable recommendation information corresponding to each recommended drill bit, a complete drill bit recommendation result is generated and output to the interactive interface for the driller to view.

[0220] By generating structured and interpretable recommendation information that includes scoring components, footage quantile predictions, confidence level explanations, and alternative solutions, drillers can fully understand the quantitative basis, expected effects, and risk boundaries of the recommendation results. The direct correspondence between the scoring components and the comprehensive score enhances the interpretability of the recommendation results. The footage quantile predictions replace single-point predictions, improving the engineering reference value. The confidence level explanations enable drillers to judge the reliability of the recommendations, and the alternative solutions provide decision-making flexibility, effectively improving the on-site adoption rate of the recommendation results.

[0221] In some embodiments, the above method may further include: constructing a drilling terminology thesaurus, wherein the thesaurus contains multiple semantically equivalent term groups; constructing positive and negative pairs based on the thesaurus, and fine-tuning the embedding model using a triplet loss function, so that semantically equivalent drilling terms are close in the embedding space, and semantically different drilling terms are far apart in the embedding space; the triplet loss function is: L_embed = max(0, d(a,p) - d(a,n) +margin), where a is the anchor vector, p is the positive example vector, n is the negative example vector, and margin is a preset interval threshold. By constructing a drilling terminology thesaurus containing multiple semantically equivalent term groups, constructing positive and negative pairs based on the thesaurus, and fine-tuning the embedding model using a triplet loss function, this loss function brings semantically equivalent drilling terms close in the embedding space and pushes away semantically different drilling terms by narrowing the distance between the anchor vector and the positive example vector and widening the distance between the anchor vector and the negative example vector. It solves the problem that semantically equivalent terms such as "hard formation" and "poor drillability" are too far apart in the embedding space in the general embedding model in the drilling field, so that the formation profile embedding vector can more accurately reflect the formation semantic information required for drill bit selection.

[0222] In some embodiments, the above method may further include: when both theoretical confidence and empirical confidence are higher than a first preset threshold, a collaborative fusion strategy is adopted to merge theoretical candidate drill bits and empirical candidate drill bits according to a 1:1 ratio of the weight corresponding to theoretical confidence to the weight corresponding to empirical confidence; when theoretical confidence is higher than the first preset threshold and empirical confidence is lower than a second preset threshold, a theory-dominated strategy is adopted to merge theoretical candidate drill bits and empirical candidate drill bits according to a 4:1 ratio of the weight corresponding to theoretical confidence to the weight corresponding to empirical confidence; when theoretical confidence is lower than the second preset threshold and empirical confidence is higher than the first preset threshold, an experience-dominated strategy is adopted to merge theoretical candidate drill bits and empirical candidate drill bits according to a 1:4 ratio of the weight corresponding to theoretical confidence to the weight corresponding to empirical confidence. When both theoretical and empirical confidence levels are above a first preset threshold, a collaborative fusion strategy is adopted, merging theoretical candidate drill bits and empirical candidate drill bits at a 1:1 ratio. When the theoretical confidence level is above the first preset threshold and the empirical confidence level is below a second preset threshold, a theory-dominated strategy is adopted, merging at a 4:1 ratio. When the theoretical confidence level is below the second preset threshold and the empirical confidence level is above the first preset threshold, an experience-dominated strategy is adopted, merging at a 1:4 ratio. This quantifies the fusion ratio in the technical solution into specific values, making the dual-path fusion strategy clearly engineering-operable.

[0223] In some embodiments, the above method may further include: calculating the recommendation adoption rate, mechanical drilling rate prediction error, and footage prediction error; calculating a comprehensive performance index based on the recommendation adoption rate, mechanical drilling rate prediction error, and footage prediction error, wherein the formula for calculating the comprehensive performance index is: E = α·adoption rate + β·(1-ROP prediction error) + γ·(1-footing prediction error), where α, β, and γ are preset weight coefficients, α+β+γ=1, the initial value of α is 0.4, the initial value of β is 0.3, and the initial value of γ is 0.3. This provides a quantitative evaluation basis for the triggering condition judgment in the three-layer incremental update mechanism, enabling the system to determine whether the recommendation quality has declined based on changes in the comprehensive performance index.

[0224] In some embodiments, the above method may further include: when a new drill bit model is added to the drill bit technology knowledge source, calculating the initial fusion weight of the new drill bit model, using the formula: w_new = θ1·T_novelty + θ2·V_degree + θ3·M_reliability, where θ1, θ2, and θ3 are preset weight coefficients, θ1+θ2+θ3=1, T_novelty is the novelty of the technology, V_degree is the degree of field verification, and M_reliability is the manufacturer's historical reliability. This allows the new drill bit to obtain a reasonable initial weight even in the absence of usage data in this block, reflecting a comprehensive consideration of technological novelty, degree of verification, and manufacturer reliability.

[0225] In some embodiments, the above method may further include: the calculation formula for the field validation degree is: V_degree = 1 - exp(-N_field / N_threshold), where N_field is the number of wells used in the field, and N_threshold is the number of validation baseline wells, wherein the number of validation baseline wells is 20. The field validation degree shows a saturation growth trend with the increase of the number of wells used. When the number of wells used is 0, the validation degree is 0; when the number of wells used is 20, the validation degree is approximately 0.63, reflecting the engineering understanding that the validation degree increases rapidly with a small amount of data, and tends to saturate after sufficient data.

[0226] In some embodiments, the above method may further include: adjusting the weights of the two paths using a linear asymptotic approach. The weight of the theoretically recommended path decreases with the number of completed wells according to the formula w_theory(n) = 0.9 - k·(n - 3), while the weight of the empirically recommended path increases with the number of completed wells according to the formula w_history(n) = 0.1 + k·(n - 3), where n is the number of completed wells and k is the asymptotic coefficient. By quantifying the linear asymptotic mechanism in the technical solution into a specific mathematical expression, the transition of weights between the theoretical and empirical paths has a clear mathematical pattern, and the introduction of the k coefficient allows the asymptotic rate to be dynamically adjusted according to data quality.

[0227] Figure 2 The overall architecture diagram of the drill bit recommendation system based on large model and multi-source knowledge fusion provided in the embodiments of this specification shows the relationship between the multi-source data input layer, data governance and standardization layer, multi-source knowledge fusion layer, model reasoning and recommendation decision layer, and recommendation result output.

[0228] Figure 3The flowchart of the drill bit recommendation method based on large model and multi-source knowledge fusion provided in the embodiments of this specification shows the processing flow from well condition and driller's requirements input, multi-source knowledge acquisition, unified modeling, candidate generation, dual verification, sorting output to feedback update.

[0229] Figure 4 This is a schematic diagram of multi-source heterogeneous drilling knowledge fusion provided in the embodiments of this specification. It shows the process of forming a fused knowledge vector after feature alignment, quality assessment and dynamic weight allocation of formation information knowledge source, historical record knowledge source, drill bit technology knowledge source and driller's requirements knowledge source.

[0230] Figure 5 The diagram illustrates the dual-path candidate drill bit generation mechanism based on theoretical recommendation and empirical verification provided in the embodiments of this specification, showing the relationship between the theoretical recommendation path, the empirical verification path, dual verification, and the generation of the fused candidate set.

[0231] Figure 6 The diagram illustrates the adaptive learning and knowledge update optimization mechanism provided in the embodiments of this specification, showing the closed-loop relationship between recommendation logs, on-site feedback, effect scoring, knowledge source contribution attribution, update determination, fusion weight update, knowledge base incremental update, and the next recommendation call.

[0232] Figure 7 The flowchart for knowledge retrieval enhancement provided in the embodiments of this specification illustrates the process of driller query parsing, vector similarity retrieval, graph relationship retrieval, evidence fragment combination, prompt word construction, domain large model generation, and result verification.

[0233] As can be seen from the technical solutions provided in the embodiments of this specification above, these embodiments can generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on formation information knowledge sources and drill bit technology knowledge sources through a large language model; generate multiple empirical candidate drill bits and their corresponding empirical confidence scores based on historical data knowledge sources and driller demand knowledge sources through similar case retrieval; screen the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints; and merge the screened theoretical candidate drill bits and empirical candidate drill bits according to the comparison results of the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and their corresponding recommendation reasons and risk warnings. By generating candidate drill bits through dual paths and combining them with engineering constraint screening, the comprehensiveness and engineering feasibility of the recommendation results are directly improved; the dynamic fusion of candidate drill bits from the two paths based on the confidence score comparison results avoids the failure of recommendations from a single path when data is sparse or theoretically insufficient, effectively enhancing the reliability and adaptability of the recommendation results.

[0234] The following is a specific embodiment of this specification: Taking the 8,000 m to 8,300 m section of a horizontal well in a certain block as an example, the geological logging report shows that the well section is mainly composed of interbedded hard and brittle sandstone and muddy sandstone, with local gravel and strong abrasiveness; the driller's input natural language requirements are "The formation in this well section is relatively hard, and we hope that the drill bit is more durable, the cost is controlled within 50,000 yuan, and frequent tripping in and out of the drill bit is avoided."

[0235] The system performs entity recognition and relation extraction on geological logging reports to obtain formation profiles such as lithology, hardness, abrasiveness, well depth, and drillability; it performs intent recognition and slot filling on the driller's needs to identify constraints such as "durability priority", "cost cap of 50,000 yuan", and "reduce tripping risk", and sets the weights of drilling speed, durability, cost and safety to 0.25, 0.35, 0.20 and 0.20 respectively.

[0236] The system retrieves historical cases from adjacent wells based on the current formation profile, filters out cases with formation similarity higher than a preset threshold, and calculates the average mechanical drilling rate, single bit footage, unit footage cost, and abnormal event records for each candidate drill bit in similar formations. At the same time, the large drilling model generates theoretically recommended parameter ranges based on formation hardness, abrasiveness, and operational requirements.

[0237] The theoretically recommended path outputs PDC drill bit A, PDC drill bit B, and roller cone drill bit C as candidates; the empirically verified path outputs PDC drill bit A and composite structure drill bit D as candidates. The system performs dual verification on the candidate drill bits, retaining PDC drill bit A, PDC drill bit B, and composite structure drill bit D that simultaneously meet the theoretical confidence level and historical case support level.

[0238] The system further calculates the semantic matching score, predicted drilling effect score, cost efficiency score, and risk assessment score between the candidate drill bits and the driller's requirements, resulting in a comprehensive score of 0.86 for PDC drill bit A, 0.81 for composite structure drill bit D, and 0.75 for PDC drill bit B. The final output identifies PDC drill bit A as the preferred option and composite structure drill bit D as an alternative, providing information on applicable formations, historical case support, expected footage, cost risks, and reasons for recommendation.

[0239] After the operation is completed, the system collects the actual drilling speed, single drill bit footage, cost, wear level, abnormal events, and driller satisfaction. The recommendation effect is written into the historical case library, and the subsequent fusion weight is adjusted according to the contribution of formation knowledge, historical case knowledge, drill bit technical knowledge, and demand knowledge in this recommendation.

[0240] Based on the above-described drill bit recommendation method, this specification also provides embodiments of a drill bit recommendation device. For example... Figure 8 As shown, the drill bit recommendation device 800 may specifically include the following modules: The first generation module 801 is used to generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the formation information knowledge source and the drill bit technology knowledge source through a large language model. The second generation module 802 is used to generate multiple candidate drill bits and their corresponding experience confidence levels based on historical knowledge sources and driller demand knowledge sources through similar case retrieval. The screening module 803 is used to screen the plurality of theoretical candidate drill bits and the plurality of empirical candidate drill bits based on engineering constraints; The merging module 804 is used to merge the screened theoretical candidate drill bits and the empirical candidate drill bits based on the comparison results of the theoretical confidence scores of each screened theoretical candidate drill bit and the empirical confidence scores of each screened empirical candidate drill bit, to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding recommendation reasons and risk warnings.

[0241] As can be seen from the technical solutions provided in the embodiments of this specification above, these embodiments can generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on formation information knowledge sources and drill bit technology knowledge sources through a large language model; generate multiple empirical candidate drill bits and their corresponding empirical confidence scores based on historical data knowledge sources and driller demand knowledge sources through similar case retrieval; screen the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints; and merge the screened theoretical candidate drill bits and empirical candidate drill bits according to the comparison results of the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and their corresponding recommendation reasons and risk warnings. By generating candidate drill bits through dual paths and combining them with engineering constraint screening, the comprehensiveness and engineering feasibility of the recommendation results are directly improved; the dynamic fusion of candidate drill bits from the two paths based on the confidence score comparison results avoids the failure of recommendations from a single path when data is sparse or theoretically insufficient, effectively enhancing the reliability and adaptability of the recommendation results.

[0242] This specification also provides a computer device for a drill bit recommendation method, including a processor and a memory for storing processor-executable instructions. Specifically, the processor can perform the following tasks according to the instructions: generating multiple theoretical candidate drill bits and their corresponding theoretical confidence scores using a large language model based on formation information knowledge sources and drill bit technology knowledge sources; generating multiple empirical candidate drill bits and their corresponding empirical confidence scores using similar case retrieval based on historical data knowledge sources and driller demand knowledge sources; filtering the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints; merging the filtered theoretical candidate drill bits and empirical candidate drill bits according to the comparison results of the theoretical confidence scores of each filtered theoretical candidate drill bit and the empirical confidence scores of each filtered empirical candidate drill bit to obtain the drill bit recommendation result for the target well; the drill bit recommendation result includes multiple recommended drill bits and their corresponding recommendation reasons and risk warnings.

[0243] To execute the above instructions more accurately, please refer to... Figure 9 As shown in the embodiments of this specification, another specific computer device 900 is also provided, wherein the computer device 900 includes a network communication port 901, a processor 902 and a memory 903, and the above structures are connected by internal cables so that the various structures can perform specific data interaction.

[0244] The processor 902 can specifically be used to: generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on formation information knowledge sources and drill bit technology knowledge sources through a large language model; generate multiple empirical candidate drill bits and their corresponding empirical confidence scores based on historical data knowledge sources and driller demand knowledge sources through similar case retrieval; screen the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints; and merge the screened theoretical candidate drill bits and empirical candidate drill bits according to the comparison results of the theoretical confidence scores of each screened theoretical candidate drill bit and the empirical confidence scores of each screened empirical candidate drill bit to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and their corresponding recommendation reasons and risk warnings.

[0245] The memory 903 can be used to store the corresponding instruction program.

[0246] In this embodiment, the network communication port 901 can be a virtual port bound to different communication protocols, thereby enabling the sending or receiving of different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. Furthermore, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM or CDMA; it can also be a Wi-Fi chip; or it can be a Bluetooth chip.

[0247] In this embodiment, the processor 902 can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers, etc. This specification is not limiting.

[0248] In this embodiment, the memory 903 includes volatile memory and non-volatile memory. The memory 903 can include multiple layers. In digital systems, anything that can store binary data can be a memory; in integrated circuits, a circuit with storage function but no physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0249] Furthermore, embodiments of this specification also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described... Figure 3 The instructions for the recommended drill bit method are shown.

[0250] Furthermore, embodiments of this specification provide a computer program product, including a computer program that, when executed by a processor, implements the above-described... Figure 1 The recommended method for drill bits is shown.

[0251] It should be understood that in the various embodiments of this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this specification.

[0252] It should also be understood that, in the embodiments of this specification, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this specification generally indicates that the preceding and following related objects have an "or" relationship.

[0253] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0254] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0255] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0256] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational tasks to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The task is a function specified in one or more boxes.

[0257] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A drill bit recommendation method, characterized in that, include: Based on the knowledge sources of formation information and drill bit technology, multiple theoretical candidate drill bits and their corresponding theoretical confidence scores are generated through a large language model. Based on historical knowledge sources and driller demand knowledge sources, multiple candidate drill bits with experience and corresponding experience confidence are generated through similar case retrieval. Based on engineering constraints, multiple theoretical candidate drill bits and multiple empirical candidate drill bits are screened. These engineering constraints include size matching constraints, drill pressure safety constraints, and formation type adaptation constraints. The size matching constraints characterize the matching relationship between the drill bit's outer diameter and the target wellbore size. The drill pressure safety constraints characterize the safety relationship between the recommended drill pressure and the drill string's buckling critical load. The formation type adaptation constraints characterize the adaptation relationship between the drill bit type and the formation drillability level. The screening of these multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints includes: screening theoretical and empirical candidate drill bits that simultaneously satisfy the size matching constraints, drill pressure safety constraints, and formation type adaptation constraints. Based on the comparison between the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening, the screened theoretical candidate drill bits and empirical candidate drill bits are merged to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding reasons for recommendation and risk warnings.

2. The method according to claim 1, characterized in that, The formation information knowledge source is used to characterize the lithological, physical, geological structural, and drilling characteristics of the formation in the target well; the historical knowledge source is used to characterize historical usage cases of drill bits in adjacent wells; and the drill bit technology knowledge source is used to characterize the specifications and parameters of the drill bit and its historical performance in the formation of the target well. The driller's requirements knowledge source is used to characterize the personalized requirements of the target well drilling operation; The method further includes: Based on formation information, historical drilling information, drill bit parameter information, and driller's operational needs information, a knowledge source for formation information, a knowledge source for historical records, a knowledge source for drill bit technology, and a knowledge source for driller's needs are constructed respectively.

3. The method according to claim 1, characterized in that, The knowledge source based on formation information and drill bit technology generates multiple theoretical candidate drill bits and their corresponding theoretical confidence scores through a large language model, including: Based on the formation information knowledge source and the drill bit technology knowledge source, and according to the four-element instruction format consisting of formation description, operation scenario, historical adjacent well operation records and output expectation, multiple theoretical candidate drill bits and their corresponding theoretical confidence scores are generated through the large language model.

4. The method according to claim 1, characterized in that, The method, based on historical knowledge sources and driller demand knowledge sources, generates multiple candidate drill bits with corresponding experience confidence levels through similar case retrieval, including: Based on the historical knowledge source and the driller's demand knowledge source, multiple successful cases of adjacent wells are retrieved for the target well according to formation similarity and operation similarity; the formation similarity is used to characterize the similarity between the target well and adjacent wells in terms of geological conditions; the operation similarity is used to characterize the similarity between the target well and adjacent wells in terms of engineering operation conditions. Based on the multiple successful cases of adjacent wells, multiple empirical candidate drill bits and the corresponding empirical confidence scores for each empirical candidate drill bit are generated.

5. The method according to claim 1, characterized in that, The process involves comparing the theoretical confidence scores of each selected theoretical candidate drill bit with the empirical confidence scores of each selected empirical candidate drill bit, then merging the selected theoretical and empirical candidate drill bits to obtain the drill bit recommendation results for the target well, including: Based on the comparison between the theoretical confidence scores of each theoretical candidate drill bit and the empirical confidence scores of each empirical candidate drill bit after screening, a merging strategy is determined. The theoretical and empirical candidate drill bits selected after being screened are merged according to the merging strategy to obtain multiple recommended drill bits; Based on the multiple recommended drill bits, a drill bit recommendation result is generated.

6. The method according to claim 5, characterized in that, The step of generating drill bit recommendation results based on the multiple recommended drill bits includes: Calculate multiple drilling engineering parameters for each recommended drill bit; Based on the driller's demand information, determine the weighting coefficients for each drilling engineering parameter; The recommendation level of each recommended drill bit is determined based on multiple drilling engineering parameters and the weighting coefficients of each drilling engineering parameter. Based on the multiple recommended drill bits and their corresponding recommendation scores, a drill bit recommendation result is generated.

7. The method according to claim 1, characterized in that, The method further includes: Based on the drill bit feedback data of the target well, the confidence levels of the formation information knowledge source, drill bit technology knowledge source, historical record knowledge source, and driller demand knowledge source are adjusted in the first cycle. Based on the drill bit feedback data from the target well, the historical knowledge source is updated in the second cycle; Based on the drill bit feedback data from the target well, the parameters of the large language model are fine-tuned in the third cycle. The first period is shorter than the second period, and the second period is shorter than the third period.

8. A drill bit recommendation device, characterized in that, include: The first generation module is used to generate multiple theoretical candidate drill bits and their corresponding theoretical confidence scores based on the knowledge sources of formation information and drill bit technology, through a large language model. The second generation module is used to generate multiple candidate drill bits and their corresponding experience confidence levels based on historical knowledge sources and driller demand knowledge sources through similar case retrieval. A screening module is used to screen multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints. The engineering constraints include size matching constraints, drill pressure safety constraints, and formation type adaptation constraints. The size matching constraints characterize the matching relationship between the drill bit's outer diameter and the target wellbore size. The drill pressure safety constraints characterize the safety relationship between the recommended drill pressure and the drill string's buckling critical load. The formation type adaptation constraints characterize the adaptation relationship between the drill bit type and the formation drillability level. The screening of the multiple theoretical candidate drill bits and multiple empirical candidate drill bits based on engineering constraints includes: screening theoretical candidate drill bits and empirical candidate drill bits that simultaneously satisfy the size matching constraints, drill pressure safety constraints, and formation type adaptation constraints. The merging module is used to merge the screened theoretical candidate drill bits and the empirical candidate drill bits based on the comparison results of the theoretical confidence scores of each screened theoretical candidate drill bit and the empirical confidence scores of each screened empirical candidate drill bit, to obtain the drill bit recommendation results for the target well; the drill bit recommendation results include multiple recommended drill bits and corresponding recommendation reasons and risk warnings.

9. A computer device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method of any one of claims 1-7.

Images