A multi-agent cooperative drilling optimization method

Abstract

The present application relates to a kind of multi-agent cooperative well drilling optimization method, it is related to well drilling engineering technical field, including: obtaining the drilling state parameter sequence of target well current single root period;According to historical well drilling experience data and drilling state parameter sequence, determine the constraint boundary of drilling control parameter in the drilling period of current single root period;Based on constraint boundary, according to the working condition experience data of multiple historical same well drilling period, with rock breaking efficiency, wellbore flow safety and pipe string dynamics safety as optimization target, the drilling control parameter in current single root period is optimized by multi-objective non-zero-sum game, and the drilling control parameter sequence of target well current single root period is obtained;According to the drilling experience data of multiple historical same well single root period, the drilling control parameter of current decision period in drilling control parameter sequence is fine-tuned, and the well drilling optimization instruction of target well current decision period is obtained.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of drilling engineering technology, specifically to a multi-agent collaborative drilling optimization method. Background Technology

[0002] The oil and gas drilling process is a complex nonlinear system composed of multiple physical subsystems, including wellbore hydraulics, drill string dynamics, and rock-breaking dynamics, which are coupled together. During drilling, reasonable parameter decisions are crucial for improving mechanical drilling speed, extending drill bit life, and ensuring wellbore stability.

[0003] However, existing drilling decision-making methods fail to effectively consider the multi-timescale coupling characteristics of the drilling process, and also fail to systematically utilize historical cross-well drilling experience data, resulting in insufficient decision rationality, low execution accuracy, and difficulty in balancing operational safety and drilling efficiency. Summary of the Invention

[0004] The purpose of the embodiments in this specification is to provide a multi-agent collaborative drilling optimization method to overcome the problems of insufficient decision-making rationality, low execution accuracy, and difficulty in balancing operational safety and drilling efficiency in existing methods.

[0005] To address the aforementioned technical problems, this specification provides, on the one hand, a multi-agent collaborative drilling optimization method, comprising: Obtain the drilling status parameter sequence for the current single cycle of the target well; Based on historical cross-well drilling experience data and drilling status parameter sequences, determine the constraint boundaries of drilling control parameters within the current single-well cycle; each drilling cycle includes multiple single-well cycles. Based on the constraint boundary and the working experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, the drilling control parameters in the current single cycle are optimized by multi-objective non-zero-sum game to obtain the drilling control parameter sequence of the target well in the current single cycle. Based on drilling experience data from multiple historical single-cycle drilling operations, the drilling control parameters for the current decision cycle in the drilling control parameter sequence are fine-tuned to obtain the drilling optimization instructions for the target well in the current decision cycle; each single-cycle includes multiple decision cycles.

[0006] Furthermore, embodiments of this specification provide a drilling optimization device, comprising: The acquisition module is used to acquire the drilling status parameter sequence of the target well in the current single cycle; The determination module is used to determine the constraint boundaries of drilling control parameters within the current single-cycle drilling cycle based on historical cross-well drilling experience data and drilling status parameter sequences; each drilling cycle includes multiple single-cycle drilling cycles. The optimization module is used to perform multi-objective non-zero-sum game optimization on the drilling control parameters in the current single cycle based on constraint boundaries and working experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, to obtain the drilling control parameter sequence of the target well in the current single cycle. The fine-tuning module is used to fine-tune the drilling control parameters in the current decision cycle of the drilling control parameter sequence based on drilling experience data from multiple historical single-cycle drilling trips, so as to obtain the drilling optimization instructions for the current decision cycle of the target well; each single cycle includes multiple decision cycles.

[0007] 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 aforementioned multi-agent collaborative drilling optimization method.

[0008] Furthermore, embodiments of this specification provide a computer storage medium storing computer program instructions, which, when executed, implement the aforementioned multi-agent collaborative drilling optimization method.

[0009] In another aspect, embodiments of this specification provide a computer program product, including a computer program that, when executed by a processor, implements the aforementioned multi-agent collaborative drilling optimization method.

[0010] As can be seen from the technical solutions provided in the embodiments of this specification above, the embodiments of this specification can obtain the drilling state parameter sequence of the current single cycle of the target well; determine the constraint boundary of the drilling control parameters within the drilling cycle in which the current single cycle is located based on historical drilling experience data of different wells and the drilling state parameter sequence; each drilling cycle includes multiple single cycles; based on the constraint boundary, and according to the working condition experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, perform multi-objective non-zero-sum game optimization on the drilling control parameters within the current single cycle to obtain the drilling control parameter sequence of the current single cycle of the target well; fine-tune the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on the drilling experience data of multiple historical single cycles of the same well to obtain the drilling optimization instruction for the current decision cycle of the target well; each single cycle includes multiple decision cycles. By determining the constraint boundaries of the drilling trip scale through cross-well experience data, and using the experience of historical drilling conditions in the same well to conduct multi-objective game optimization at the single-section scale, and then fine-tuning the execution scale based on the experience of single-section drilling in the same trip, an optimization decision-making link with multi-time scale progressive steps and cross-scale reuse of experience data is formed. This can achieve multi-objective collaborative optimization of rock breaking efficiency, wellbore flow safety and tubing dynamics safety while ensuring safety boundaries. At the same time, by continuously correcting the current operation through historical experience, the decision rationality and execution accuracy of drilling control parameters are effectively improved, taking into account both drilling operation safety and drilling efficiency. Attached Figure Description

[0011] 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.

[0012] Figure 1 This is a flowchart of a multi-agent collaborative drilling optimization method provided in the embodiments of this specification; Figure 2 The overall framework and long-sequence closed-loop diagram of the domain knowledge-guided multi-agent game-theoretic drilling long-sequence decision-making system provided in the embodiments of this specification; Figure 3 This is a schematic diagram illustrating the construction and retrieval of the drilling long-sequence decision knowledge base and the three-level memory provided in the embodiments of this specification; Figure 4 This is a schematic diagram of the triplet state evaluation system and hierarchical decision-making process provided in the embodiments of this specification; Figure 5 This is a schematic diagram of the collaborative decision-making process for multi-agent dynamic game provided in the embodiments of this specification; Figure 6 This is a schematic diagram of a three-step human-machine collaborative feedback conversion process provided in the embodiments of this specification; Figure 7This is a schematic diagram of the structural composition of a drilling optimization device provided in the embodiments of this specification; Figure 8 This is a schematic diagram of the structural composition of a computer device provided in the embodiments of this specification. Detailed Implementation

[0013] 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.

[0014] 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.

[0015] In some embodiments, drilling refers to the entire process of drilling a wellbore from the surface into the ground and completing well construction, encompassing multiple operations such as drilling, circulation and flushing, tripping, casing running, and cementing. Drilling specifically refers to the core stage in the drilling process where the drill bit breaks rocks at the bottom of the well, continuously deepening the wellbore. The efficiency and safety of the drilling process depend not only on the rock-breaking effect during the drilling stage but also on factors throughout the entire drilling cycle, such as wellbore cleaning and string stress. Drilling is the core sub-process of drilling; drilling includes but is not limited to drilling. Drilling provides the equipment, circulation, well control, and wellbore structure conditions for drilling, while drilling provides the wellbore extension and footage for drilling.

[0016] In some embodiments, the target well is the wellbore corresponding to the current drilling operation. The target well is the direct service object for drilling decision optimization. The target well is unique, distinguishing it from other wellbores that provide historical drilling data. Other wellbores include adjacent wells in the same block and drilled or abandoned wells in different blocks. Their drilling data serves as historical drilling experience data for different wells, providing a reference for knowledge transfer in optimizing drilling parameters for the target well. A connection is established between the target well and historical different wells through the similarity of formation feature vectors. The higher the similarity, the greater the reference value of the historical experience data for the current decision-making of the target well.

[0017] Historical drilling experience data is used to characterize drilling records accumulated during the historical drilling of adjacent wells in the same block and wells in different blocks that have been drilled or abandoned. These records can provide knowledge transfer references for target wells. The data is obtained by retrieving data such as drilling design, logging, measurement while drilling and logging, drill bit usage, and completion reports from the drilling information management system. After timestamp alignment, well depth normalization, outlier removal, missing value completion, and dimension unification, the data is divided into multiple working condition segments according to preset working condition segmentation rules. For each working condition segment, formation feature vectors, drilling strategy parameters, drilling response parameters, safety indicators, and operation result labels are extracted to form structured experience entries. Each structured experience item includes: formation characteristics, characterizing the lithology, drillability, compressive strength, abrasiveness, density, gamma ray, resistivity, porosity, and temperature of the formation segment; operating condition type, characterizing normal drilling, poor wellbore cleanliness, abnormal friction, buckling risk, drill bit mud packing, or drill string vibration; strategy success rate, characterizing the percentage of decisions made within the segment without safety exceedances; comprehensive efficiency score, characterizing the weighted score of mechanical drilling speed improvement, mechanical energy reduction, and risk reduction; and applicable boundaries, characterizing the allowable range of drilling pressure, rotational speed, displacement, and safety indicators. These experience items are correlated with the target well based on formation characteristic similarity; higher similarity indicates greater reference value.

[0018] In some embodiments, drilling condition parameters include a first state parameter characterizing the operating conditions of the drilling process and a second state parameter characterizing the cumulative effects of the drilling process. The first state parameter includes at least mechanical specific energy, torque fluctuation rate, rock drillability, mechanical drilling rate, bottom hole pressure, and vibration intensity. The second state parameter includes at least cumulative footage, drill bit wear, cumulative tubing fatigue, and tripping time.

[0019] Drilling status parameters characterize the real-time operating status and historical cumulative status of the drilling system during operation. The first status parameter is an instantaneous monitoring parameter, acquired or calculated in real-time through measurement-while-drilling tools and surface sensors. It includes at least mechanical energy specificity, torque fluctuation rate, rock drillability, rate of penetration, bottom hole pressure, and vibration intensity, reflecting the drilling system response at the current moment or within the current time window. The second status parameter is a cumulative statistical parameter, obtained by integrating or summing sensor data over time. It includes at least cumulative footage, bit wear, cumulative tubing fatigue, and trip duration, reflecting the historical cumulative effect from the start of the current trip to the present moment. The first and second status parameters together constitute the basic input data for game-theoretic optimization and constraint boundary calculation. The first status parameter drives the retrieval and matching of historical experience and the identification of operating conditions, while the second status parameter is used to perceive the long-term impact of historical decisions on bit life and tubing safety.

[0020] In some embodiments, drilling control parameters include at least drilling pressure, rotational speed, and displacement.

[0021] Drilling control parameters are actively adjustable operational variables in drilling operations, i.e., the decision objects to be optimized. Drilling control parameters include at least bit pressure, rotational speed, and flow rate. These three parameters are interdependent; changes in any one parameter will affect the optimization effect of the others. For example, increasing bit pressure can improve rock-breaking efficiency but will exacerbate drill bit wear and increase the risk of tubing buckling; increasing flow rate is beneficial for improving rock-carrying capacity and wellbore cleanliness but will increase pump pressure and the risk of lost circulation; increasing rotational speed can increase drilling speed but will exacerbate tubing vibration and fatigue damage. By adjusting the combination of these parameters, a dynamic coordination and trade-off can be achieved among the three optimization objectives of rock-breaking efficiency, wellbore flow safety, and tubing dynamic safety.

[0022] In some embodiments, drilling engineering parameters include at least well depth, formation drillability, mud density, formation pore pressure, and formation fracture pressure.

[0023] Drilling engineering parameters are input parameters that characterize the engineering environment and design conditions of the current drilling operation. They are used to support the constraint boundary reasoning of the physical model in the mechanistic knowledge base. Drilling engineering parameters include at least well depth, formation drillability, mud density, formation pore pressure, and formation fracture pressure. These parameters are derived from drilling design documents, well logging interpretation results, or data from adjacent wells, and are generally considered fixed values ​​within the same drilling cycle, not changing with optimization iterations. The difference between drilling engineering parameters and drilling status parameters is that drilling engineering parameters reflect the external environmental conditions and design boundaries faced by the drilling operation, serving as prerequisite constraints for decision-making; while drilling status parameters reflect the actual operational response of the drilling system under these environmental conditions, serving as feedback for decision-making. Both can be input into the physical model in the mechanistic knowledge base to calculate the constraint boundaries of drilling control parameters under the current operating conditions.

[0024] This specification provides an embodiment of a domain knowledge-guided multi-agent game-theoretic drilling long-sequence decision-making system, comprising: a knowledge and memory construction unit, a hierarchical decision-making framework unit, a multi-agent game-theoretic decision-making unit, and a causal reasoning human-machine collaboration unit.

[0025] The knowledge and memory building unit is configured to construct a long-sequence drilling decision knowledge base and a three-level memory base consisting of a single-well instant database, a trip drilling condition database, and a cross-well experience database. The long-sequence drilling decision knowledge base includes three types of knowledge: mechanistic knowledge, engineering experience knowledge, and risk boundary knowledge. Mechanistic knowledge is embedded in decision-making reasoning in the form of physical constraints; engineering experience knowledge stores historical well parameter optimization experience, drill bit selection rules, and operating condition identification patterns in the form of structured triples; and risk boundary knowledge stores safe pressure windows, tubing buckling limits, and equipment power limits as hard constraint boundaries. The single-well instant database... A rolling window mechanism is used to record the state-action-feedback triplet during the current drilling process of a single well. The drilling condition database stores summaries of each single well's condition extracted from the real-time single well database and accumulates to form a knowledge of drilling conditions. The cross-well experience database stores structured experience entries of successful decision-making cases and lessons learned from failures in different wells and blocks. Each experience entry is associated with a formation feature vector, historical successful decision-making strategies, comprehensive effect scores, and applicable boundaries. Each experience entry is also marked with the source well, applicable formation, and confidence level. The three-level memory is used to provide knowledge support and historical experience for subsequent decisions.

[0026] The hierarchical decision-making framework unit is configured to construct a ternary state evaluation system and a hierarchical decision-making framework for long-term drilling decisions. The ternary state evaluation system includes state variables, decision variables, and constraint variables. State variables include instantaneous state variables and cumulative state variables. Instantaneous state variables include at least mechanical specific energy, torque fluctuation rate, rock drillability, real-time drilling speed, bottom hole pressure, and vibration intensity. Cumulative state variables include at least cumulative footage, normalized bit wear, cumulative tubing fatigue wear, and the duration of the current drilling trip. Decision variables include drilling pressure adjustment, rotational speed adjustment, and displacement adjustment. Constraint variables include hard constraints and soft constraints. Hard constraints include at least the safe pressure window, critical drilling pressure for tubing buckling, and maximum permissible torque. Soft constraints include at least the upper limit of bit wear, the upper limit of cumulative tubing fatigue, and the pump pressure rating. The hierarchical decision-making framework includes strategic... The hierarchical decision-making framework comprises a planning layer, a collaborative decision-making layer, and an execution control layer. The strategic planning layer uses a single root sequence as the evaluation unit and outputs the parameter envelope intervals and risk boundaries for drilling pressure, rotational speed, and displacement based on the results retrieved from the three-level memory. The collaborative decision-making layer calls the multi-agent game module for combined optimization within a sliding time window, and the execution control layer makes continuous adjustments based on real-time monitoring data. The hierarchical decision-making framework unit is also configured to construct a two-way information transmission closed-loop mechanism between the strategic planning layer, the collaborative decision-making layer, and the execution control layer. Downward, the strategic planning layer transmits the target envelope interval and risk boundary to the collaborative decision-making layer, and the collaborative decision-making layer transmits parameter settings to the execution control layer. Upward, the execution control layer feeds back real-time execution deviations and constraint violations to the collaborative decision-making layer, and the collaborative decision-making layer feeds back the target achievement and strategy effectiveness assessment to the strategic planning layer to trigger dynamic corrections of the upper-level planning.

[0027] The multi-agent game decision-making unit is configured to construct a multi-agent dynamic game collaborative decision-making module. This module comprises three game agents: a rock-breaking efficiency agent, a wellbore flow safety agent, and a tubing string dynamics safety agent. The rock-breaking efficiency agent proposes strategies for increasing or decreasing drilling pressure and rotational speed based on mechanical specific energy, mechanical drilling speed, formation drillability, and drill bit wear. The wellbore flow safety agent proposes strategies for displacement and pressure boundaries based on bottom hole pressure, pump pressure, annular rock-carrying capacity, and safety pressure window. The tubing string dynamics safety agent proposes strategies for limiting drilling pressure and rotational speed based on torque, vibration, buckling margin, and fatigue accumulation. The three agents engage in a non-zero-sum dynamic game around the adjustments to drilling pressure, rotational speed, and displacement. The multi-agent game decision-making unit maps hard constraints to action space filtering conditions and soft constraints to penalty terms in the payoff functions of each agent, and then... The joint strategy of ε-Nash equilibrium under Pareto constraints is obtained by combining the recommended parameter range of the knowledge base with the retrieval results of the three-level memory. The multi-agent game decision unit is also configured to construct a long-sequence decision mechanism. By explicitly incorporating accumulated state variables into the state transition equation, rolling updates of sliding time windows, and hierarchical responses to sudden changes in operating conditions, each agent considers both immediate gains and long-term accumulated costs such as drill bit wear, tubing fatigue, and bottom hole pressure fluctuations in the payoff calculation of the current window, thereby achieving dynamic collaborative decision-making throughout the drilling cycle. The multi-agent game decision unit adopts a centralized training-distributed execution framework. The joint strategy is learned by using ε-Nash equilibrium under Pareto constraints as the convergence criterion. In the centralized training phase, the joint strategy is learned by sharing the global state, local observations of each agent, candidate actions, and joint reward signals through the central controller. In the distributed execution phase, each agent independently generates decisions based only on local observation information.

[0028] The causal reasoning human-machine collaboration unit is configured to construct an interpretable causal reasoning and human-machine collaboration mechanism for intelligent drilling decision-making. This unit includes a causal knowledge graph construction module, an interpretable decision reasoning module, a human-machine collaboration feedback and transformation module, and a trust assessment module. The causal knowledge graph construction module is configured to construct a causal knowledge graph for drilling decisions. This graph represents the causal relationships between drilling state variables, decision parameters, and engineering objectives using a directed graph structure. It is generated by integrating an explicit mechanism rule base and a data-driven pattern base. The explicit mechanism rule base extracts physical constraint causal relationships from rock mechanics, tubing dynamics, and hydraulics, while the data-driven pattern base mines implicit causal relationships from historical drilling data and expert experience. The decision-making logic module is configured to establish an interpretable decision-making reasoning mechanism based on causal inference, providing a synchronous causal explanation path for each decision output, forming a transparent reasoning chain from state perception to strategy selection. The human-machine collaborative feedback conversion module is configured to construct a three-step human-machine collaborative feedback conversion mechanism, which parses expert natural language feedback into structured triples and maps them to soft constraint tightening amounts or target weight coefficient adjustment amounts, and applies the corrected constraints or weights in the next decision window. After expert confirmation, the changes take effect permanently; otherwise, they are automatically rolled back, realizing online correction of expert knowledge. The trust assessment module is configured to score the trust level of each decision output, assisting on-site engineers in determining whether manual intervention is needed.

[0029] Corresponding to the above system, this specification provides a multi-agent collaborative drilling optimization method, referring to... Figure 1 As shown, the specific implementation may include the following steps: S101: Obtain the drilling status parameter sequence for the current single cycle of the target well.

[0030] In some embodiments, the current single-cycle monitoring data of the target well can be acquired during drilling.

[0031] Monitoring while drilling (MSW) data is used to characterize the raw signal records collected in real time by downhole tools and surface sensors during the current drilling process of a single well. This MSW data includes, but is not limited to, time-series signals from sensor channels such as drill pressure, torque, rotational speed, displacement, pump pressure, bottom hole pressure, vibration acceleration, temperature, and flow rate. The acquisition methods for MSW data include, but are not limited to: collecting near-bit engineering parameter signals using downhole measuring tools installed near the drill bit; collecting surface engineering parameter signals such as riser pressure, mud flow rate, and hook load using surface sensors; and transmitting and storing these signals to the drilling data acquisition and monitoring system at a set sampling frequency after analog-to-digital conversion, forming a time series of MSW data.

[0032] In some embodiments, the drilling status parameter sequence for the current single cycle of the target well can be extracted based on the monitoring while drilling data.

[0033] The drilling status parameter sequence is used to characterize the drilling system operation status of the target well at each time sampling point within the current single drilling cycle. This drilling status parameter sequence is obtained by preprocessing and feature extraction of the monitoring data while drilling. The extraction methods include, but are not limited to: denoising and filtering the monitoring data while drilling to remove high-frequency interference introduced by sensor noise and downhole vibration; arranging the filtered signals of each channel in chronological order to form a time series of each status parameter; for each time sampling point, extracting mechanical specific energy, torque fluctuation rate, rock drillability, mechanical drilling rate, bottom hole pressure, and vibration intensity from the preprocessed signal as instantaneous status parameters, and extracting cumulative footage, drill bit wear, cumulative tubing fatigue, and trip time as cumulative status parameters; combining the above parameters in chronological order of sampling time to form the drilling status parameter sequence, where each element corresponds to a multidimensional status parameter vector at a time sampling point.

[0034] S102: Based on historical drilling experience data and drilling status parameter sequences, determine the constraint boundaries of drilling control parameters within the current single-cycle drilling cycle; each drilling cycle includes multiple single-cycle drilling cycles.

[0035] In some embodiments, each drilling cycle includes multiple single-root cycles.

[0036] A trip drilling cycle is used to characterize the depth range and time span corresponding to the complete operation process of a drill bit from the bottom of the well to its tripping out of the well. A single drill pipe cycle is used to characterize the depth range and time span corresponding to the start of drilling for a single drill pipe. The inclusion relationship between a trip drilling cycle and a single drill pipe cycle is as follows: a trip drilling cycle covers the entire drilling depth from the starting depth to the ending depth of the trip. Within this depth range, it is divided into multiple consecutive single drill pipe cycles according to the length of a single drill pipe. The depth ranges of each single drill pipe cycle are contiguous and do not overlap, together forming the complete depth range of the trip drilling cycle.

[0037] In some embodiments, the constraint boundary includes an envelope of drilling control parameters and a risk boundary, wherein the envelope is used to characterize the adjustment range of drilling control parameters and the risk boundary is used to characterize the allowable limit of drilling control parameters.

[0038] The constraint boundaries include the envelope interval of the drilling control parameters and the risk boundary. The envelope interval characterizes the allowable range of adjustments for each drilling control parameter during a drilling cycle. This range is defined by upper and lower limits; parameter adjustments within this range do not require additional safety checks. The risk boundary characterizes the extreme threshold values ​​of each drilling control parameter under safety constraints. These extreme threshold values ​​are insurmountable hard boundary values; safety protection actions are triggered when any control parameter reaches or exceeds the risk boundary. Both the upper and lower limits of the envelope interval are within the corresponding risk boundary ranges, and the numerical range of the envelope interval is smaller than the numerical range defined by the risk boundary.

[0039] In some embodiments, the first formation characteristics of the drilling section in the current single cycle can be generated based on the drilling state parameter sequence.

[0040] The first formation feature is used to characterize the geological conditions and drillability of the well section drilled in the current single cycle. This first formation feature is generated by multi-dimensional fusion of rock drillability parameters, mechanical drilling rate parameters, and bottom hole pressure parameters from the drilling state parameter sequence. The generation method includes, but is not limited to: extracting the rock drillability value, mechanical drilling rate value, and bottom hole pressure value corresponding to each sampling point in the current single cycle from the drilling state parameter sequence; performing statistical processing on the above three types of parameter values ​​at each sampling point to obtain their respective average values ​​and rates of change; combining the average values ​​and rates of change into a multi-dimensional vector, and using this multi-dimensional vector as the first formation feature.

[0041] In some embodiments, the similarity between a first formation feature and a second formation feature associated with historical drilling experience data of different wells can be calculated.

[0042] The second formation feature is used to characterize the geological conditions and drillability attributes of historical out-of-bounds wells in the corresponding drilling intervals. Each historical out-of-bounds well drilling experience data pre-stores its associated second formation feature. Similarity calculation methods include, but are not limited to: normalizing the first formation feature and each second formation feature separately, ensuring that the feature values ​​of each dimension are within the same numerical range; calculating the difference between the normalized first formation feature and the second formation feature in each dimension, taking the square root of the weighted sum of the squared differences to obtain the weighted Euclidean distance value; using the reciprocal or negative correlation mapping value of this weighted Euclidean distance value as the similarity between the first formation feature and the second formation feature; the smaller the weighted Euclidean distance value, the higher the corresponding similarity. By calculating the similarity, historical reference cases closest to the current formation conditions can be selected from multiple historical out-of-bounds well drilling experience data.

[0043] Furthermore, a historical stratigraphic feature matrix can be constructed based on the secondary stratigraphic features associated with historical drilling experience data from various wells. Principal component analysis is then performed on the historical stratigraphic feature matrix to extract principal components whose cumulative variance contribution rate exceeds a threshold, thus constructing a projection space. The primary stratigraphic features and each secondary stratigraphic feature are then projected into the projection space to obtain the corresponding primary and secondary projected features. The similarity between the primary and secondary projected features is then calculated.

[0044] The historical stratigraphic feature matrix is ​​used to characterize the distribution structure of the second stratigraphic features corresponding to multiple historical drilling experience data in the original feature space. The contribution rate threshold is used to control the number of principal components selected; when the cumulative variance contribution rate reaches this threshold, the extraction of subsequent principal components is stopped. The specific implementation of principal component analysis includes, but is not limited to: calculating the covariance matrix of the historical stratigraphic feature matrix; performing eigenvalue decomposition on the covariance matrix; arranging the eigenvalues ​​in descending order; summing the proportion of each eigenvalue to the total sum of all eigenvalues ​​to obtain the cumulative variance contribution rate; selecting the first few eigenvalues ​​corresponding to the cumulative variance contribution rate reaching the contribution rate threshold and their corresponding eigenvectors; and using these eigenvectors to construct a projection matrix. Multiplying the first stratigraphic feature and each second stratigraphic feature by this projection matrix respectively, we obtain the dimensionality-reduced first projected feature and each second projected feature. Calculating the Euclidean distance or cosine distance between the first projected feature and each second projected feature, and using the negative correlation value of this distance as the similarity. Principal component analysis reduces the dimensionality of high-dimensional stratigraphic feature data to a low-dimensional space, while retaining the main feature information and removing redundant features and noise. This reduces the data dimension of similarity calculation, decreases the computational load of retrieval and matching, and improves the response speed in real-time decision-making scenarios.

[0045] Furthermore, based on the characteristics of the first formation and each of the second formations, the drillability sequence of the first formation distributed along the well depth direction within the current single well cycle, and the drillability sequence of the second formation distributed along the well depth direction from historical drilling experience data of different wells, can be determined respectively. The drillability sequences of the first formation and each of the second formations are then time-normalized and aligned, and the cumulative local distance between corresponding points along each alignment path is calculated. Based on the minimum value of the cumulative local distance, the similarity between the drillability sequences of the first formation and each of the second formations is determined.

[0046] The first formation drillability sequence is used to characterize the formation drillability change trajectory with well depth within the current single cycle; the second formation drillability sequence is used to characterize the formation drillability change trajectory with well depth within the corresponding drilling section of historical different wells. Time warping sequence alignment is achieved using a dynamic time warping algorithm. This algorithm does not require a one-to-one correspondence of sampling points in the depth direction between the two sequences, allowing for stretching or compression of the local depth axis during sequence matching, so that points with similar waveform characteristics in the two sequences can be aligned to the same path. Specific implementations of the dynamic time warping algorithm include, but are not limited to: using the first formation drillability sequence as the reference sequence and the second formation drillability sequence as the test sequence, constructing a local distance matrix between corresponding points in the two sequences; searching for the curved path with the minimum cumulative distance from the starting point to the ending point in this local distance matrix; the cumulative distance value on this curved path is the minimum local cumulative distance value; and using the negative correlation mapping value of this minimum value as the similarity between the sequences. By using a dynamic time warping algorithm to flexibly align the formation drillability sequence, it is possible to overcome the local stretching or compression effect in the depth dimension of the formation drillability sequence caused by differences in drill bit selection, drilling parameters, or local changes in geological structure between different wells, and avoid the mismatch problem caused by depth misalignment in the traditional point-by-point matching method.

[0047] It is also possible to calculate the intra-group similarity between the first stratigraphic feature and each second stratigraphic feature within multiple attribute groups; multiple attribute groups include lithological feature group, mechanical feature group, and pressure feature group. Based on the influence of each attribute group on the drilling response, the inter-group weights of each attribute group are determined. Similarity is then calculated based on the intra-group similarity and inter-group weights of each attribute group.

[0048] The lithological characteristic group is used to characterize the rock type and mineral composition of the formation; the mechanical characteristic group is used to characterize the strength, hardness, and fracturability of the formation; and the pressure characteristic group is used to characterize the pore pressure and fracture pressure of the formation. The calculation of intra-group similarity includes, but is not limited to: normalizing the feature values ​​of each dimension within the same attribute group, calculating the weighted Euclidean or Mahalanobis distance between the first formation feature and the second formation feature within the same attribute group, and using the negative correlation value of this distance as the intra-group similarity. The determination of inter-group weights includes, but is not limited to: determining the initial weight of each attribute group based on the average correlation coefficient between each feature of each dimension and the historical mechanical drilling rate; normalizing the initial weights of each attribute group so that the sum of the inter-group weights of each attribute group is 1. The calculation of the comprehensive similarity includes, but is not limited to: multiplying the intra-group similarity of each attribute group by its corresponding inter-group weight and then summing the results to obtain the weighted and fused comprehensive similarity. By dividing multidimensional formation features into three attribute groups—lithology, mechanics, and pressure—and performing similarity calculations within each group before weighted fusion, differentiated similarity measurement methods can be adopted for features of different attribute types. This allows the comprehensive similarity to more accurately reflect the applicability of historical experience under current working conditions and improves the reliability of cross-well knowledge transfer.

[0049] In some embodiments, historical drilling experience data from different wells that meet preset similarity conditions can be selected as matching experience data.

[0050] A preset similarity condition is used to filter historical cross-well experiences that are sufficiently similar to the current formation characteristics. This preset similarity condition is that the similarity value exceeds a preset similarity threshold, or is among the top N in a sorted order of similarity values, where N is a preset positive integer. Selection methods include, but are not limited to: comparing each calculated similarity value with the preset similarity threshold, retaining historical cross-well drilling experience data with similarity values ​​greater than or equal to the threshold as matching experience data; when no historical cross-well drilling experience data with similarity values ​​exceeding the threshold exists, selecting a preset number of historical cross-well drilling experience data as matching experience data, sorted from highest to lowest similarity value. By selecting matching experience data, historical experiences with low relevance to the current formation conditions are filtered out, reducing interference data in subsequent constraint boundary calculations.

[0051] In some embodiments, the constraint boundary of the drilling control parameters within the current single cycle can be generated based on the historical drilling control parameters and corresponding drilling efficiency in the matching experience data.

[0052] The historical drilling control parameters in the matching experience data include the drilling pressure, rotation speed and displacement values ​​corresponding to each matching experience, and the corresponding drilling efficiency includes the comprehensive efficiency evaluation value or mechanical drilling speed value corresponding to each matching experience. The generation of constraint boundaries includes, but is not limited to: weighting the drilling pressure (DPP) values ​​in each matched empirical data set according to the corresponding drilling efficiency to obtain a DPP center value; using this DPP center value as a reference, extending upwards and downwards by preset offsets to generate the upper and lower limits of the DPP envelope interval; weighting the rotational speed (RPS) values ​​in each matched empirical data set according to the corresponding drilling efficiency to obtain a RPS center value; using this RPS center value as a reference, extending upwards and downwards by preset offsets to generate the upper and lower limits of the RPS envelope interval; weighting the displacement (DPS) values ​​in each matched empirical data set according to the corresponding drilling efficiency to obtain a displacement center value; using this displacement center value as a reference, extending upwards and downwards by preset offsets to generate the upper and lower limits of the displacement envelope interval; using the minimum of the minimum values ​​of each control parameter in each matched empirical data set as the lower limit of the risk boundary, and using the maximum of the maximum values ​​of each control parameter in each matched empirical data set as the upper limit of the risk boundary. These constraint boundaries are used to limit the search range of each control parameter in the subsequent game optimization process, restricting the optimization space to a safe area consistent with historical successful experiences.

[0053] S103: Based on the constraint boundary and the working experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamic safety as optimization objectives, the drilling control parameters in the current single cycle are optimized by multi-objective non-zero-sum game to obtain the drilling control parameter sequence of the target well in the current single cycle.

[0054] In some embodiments, a first game agent, a second game agent, and a third game agent can be constructed respectively, corresponding to rock breaking efficiency, wellbore flow safety, and tubing dynamics safety.

[0055] Game agent refers to a computational entity that independently makes decisions and updates strategies during a dynamic game. Each game agent has an independent observation space, action space, and payoff function, used to output corresponding candidate actions based on the current state within each decision window. Rock breaking efficiency characterizes the efficiency level of the drill bit in breaking rock. It is obtained by comprehensively evaluating mechanical drilling rate, mechanical specific energy, and drill bit wear. Higher rock breaking efficiency indicates a larger volume of rock broken per unit time or a lower energy consumption per unit volume of rock broken. Wellbore flow safety characterizes the pressure balance and cutter-carrying cleanliness of the drilling fluid circulation system within the wellbore. It is obtained by comprehensively evaluating the position of the bottom hole pressure relative to the safe pressure window, annular return velocity, and cuttings bed height. Higher wellbore flow safety indicates more stable bottom hole pressure, more sufficient cutter-carrying capacity, and lower risk of well leakage and overflow. Tube string dynamic safety characterizes the drill string's ability to resist vibration, buckling, and fatigue failure during downhole movement. It is comprehensively evaluated by torque ripple rate, vibration intensity, buckling margin, and fatigue accumulation. Higher tube string dynamic safety indicates smoother drill string operation, lower buckling risk, and longer fatigue life. The construction methods for the first, second, and third game agents include, but are not limited to: setting a corresponding local observation space for each game agent, which is used to select a subset of state variables related to the corresponding optimization objective from the global state; setting a corresponding action space for each game agent, which is used to limit the types and range of actions that the agent can output; initializing a corresponding reward function for each game agent, which is used to calculate the agent's reward value based on the current state and output actions; and setting a corresponding policy network for each game agent, which is used to map local observations to specific action values ​​in the action space. The policy network is initialized randomly, and its parameters are updated during subsequent training phases. The local observation space of the first game agent includes at least mechanical specific energy, mechanical drilling rate, formation drillability, drill bit wear degree, and drilling pressure response. Its action space includes drilling pressure adjustment and rotation speed adjustment. The payoff function is positively correlated with mechanical drilling rate and negatively correlated with drill bit wear increment. The local observation space of the second game agent includes at least bottomhole pressure, equivalent circulation density, pump pressure, annular return velocity, cuttings bed height, and safe pressure window. Its action space includes displacement adjustment and pressure boundary constraints. The payoff function is positively correlated with bottomhole pressure stability and negatively correlated with wellbore risk. The local observation space of the third game agent includes at least torque, friction, axial load, vibration intensity, buckling margin, and fatigue accumulation. Its action space includes drilling pressure limit, rotation speed limit, and parameter change rate limit. The payoff function is positively correlated with vibration suppression effect and negatively correlated with fatigue accumulation.By constructing three independent game agents, each optimization objective has its own independent decision-making entity. In subsequent game processes, each agent can independently optimize its own strategy according to its own payoff function, thereby achieving dynamic coordination of multi-objective conflicts in a unified environment through policy interaction between agents.

[0056] In some embodiments, drilling state parameter sequences and operating condition experience data can be input into each game agent to obtain a first candidate action, a second candidate action, and a third candidate action; each candidate action includes an adjustment amount to the drilling control parameters.

[0057] Operating condition experience data is used to characterize the correspondence between drilling states and decision actions recorded under similar operating conditions during historical well trips. This operating condition experience data is stored in the trip operating condition database, containing operating condition feature vectors for each historical moment, along with their corresponding decision actions and profit records. Candidate actions are used to characterize the drilling control parameter adjustment schemes recommended by each game agent for its respective optimization objective in the current state. These adjustment schemes include one or more of the following: drilling pressure adjustment, rotational speed adjustment, and displacement adjustment. The candidate actions of different game agents are independent of each other in terms of the direction and magnitude of the adjustment values. The methods for inputting the drilling state parameter sequence and operating condition experience data into each game agent include, but are not limited to: extracting features from the drilling state parameter sequence along the time axis to obtain a multi-dimensional operating condition feature vector for the current single cycle; combining this multi-dimensional operating condition feature vector with the operating condition feature vectors for each historical moment stored in the operating condition experience data to form the input feature tensor for each game agent; and inputting the received input feature tensor into its respective policy network, which then outputs the corresponding candidate action after forward propagation calculation. By inputting state parameters and empirical data into each game agent, each agent can simultaneously refer to the current real-time state and historical experience when generating candidate actions, thereby improving the rationality and reliability of candidate actions.

[0058] The drilling state parameter sequence for the current single cycle can be input into the first, second, and third game agents respectively. The first game agent, based on a comparison between the current rate of drilling (RBD) and the average RBD under the same well and operating conditions in the empirical data, determines the adjustment direction and step size for the pressure on drill bit (PBD) and rotational speed (SSD) adjustments, generating a first candidate action. The second game agent, based on a comparison between the current bottom hole pressure and the safety pressure window boundary value in the empirical data, determines the adjustment direction and step size for the displacement adjustment, generating a second candidate action. The third game agent, based on a comparison between the current vibration intensity and the allowable vibration threshold in the empirical data, determines the limiting direction and amplitude for the PBD and SSD adjustments, generating a third candidate action.

[0059] The average mechanical drilling rate (MRR) under the same well and working conditions is used to characterize the statistical average level of the MMR during historical drilling under the same well and working conditions. This average MMR is obtained by arithmetically averaging all MMR records under the same well and working condition category in the working condition experience data. The safety pressure window boundary value is used to characterize the upper and lower limits of the safety pressure window. The upper limit is determined by the formation fracture pressure, and the lower limit is determined by the formation pore pressure. The permissible vibration threshold is used to characterize the maximum allowable vibration intensity of the tubing string under safe operating conditions. This permissible vibration threshold is derived from tubing string dynamic safety specifications or historical experience data. The comparison result is used to characterize the numerical relationship between the current measured value and the reference value. This comparison result includes three cases: the current value is higher than the reference value, the current value is lower than the reference value, and the current value is equal to the reference value. The adjustment direction characterizes the increasing or decreasing trend of control parameters. For drill pressure (DP) and rotational speed, when the current mechanical drilling rate (MRR) is lower than the average MRR, DP and rotational speed are increased to improve rock breaking efficiency; when the current MRR is higher than the average MRR, DP and rotational speed are maintained or slightly decreased to control drill bit wear. For displacement, when the current bottom hole pressure is lower than the center value of the safe pressure window, displacement is increased to increase the bottom hole pressure; when the current bottom hole pressure is higher than the center value of the safe pressure window, displacement is decreased to reduce the bottom hole pressure. The adjustment step size characterizes the magnitude of each adjustment. This adjustment step size is positively correlated with the difference between the current value and the reference value; the larger the difference, the larger the adjustment step size. The constraint direction characterizes the constrained trend of DP and rotational speed. This constraint direction is the direction in which DP and rotational speed are reduced when the vibration intensity exceeds the allowable vibration threshold. The constraint amplitude characterizes the amount by which DP and rotational speed are restricted. This constraint amplitude is positively correlated with the deviation of the current vibration intensity from the allowable vibration threshold; the larger the deviation, the larger the constraint amplitude. The first candidate action includes the adjustment amount of drilling pressure and rotation speed determined based on the above comparison results; the second candidate action includes the adjustment amount of displacement determined based on the above comparison results; and the third candidate action includes the adjustment amount of drilling pressure and rotation speed determined based on the above comparison results. By having each agent independently generate candidate actions based on the comparison results between the current measured values ​​and historical reference values, each candidate action directly reflects the degree of deviation of the current working condition from historical experience. This approach offers high computational efficiency and transparent decision-making logic, making it suitable for scenarios with relatively stable working conditions and sufficient historical experience data.

[0060] Each game agent can also calculate the similarity between the working characteristics corresponding to the current drilling state parameter sequence and the working characteristics at each historical moment in the working condition experience data. Multiple historical moments with working condition similarity exceeding a preset threshold are selected as a matching moment set. Each game agent can determine its optimal action preference in the current state based on the historical decision actions and corresponding historical rewards for each historical moment in the matching moment set. The optimal action preference includes action type and action magnitude. Each game agent generates corresponding candidate actions based on its own optimal action preference.

[0061] Operating condition features characterize the operational status pattern of the drilling system at a specific moment or within a time window. These features are composed of multi-dimensional drilling state parameters, including normalized feature vectors formed by parameters such as mechanical specific energy, torque fluctuation rate, bottom hole pressure, vibration intensity, and mechanical drilling rate. Operating condition similarity characterizes the closeness between current and historical operating condition features in the feature space. This similarity is obtained by calculating the negative correlation between the cosine similarity or Euclidean distance of the two feature vectors. The matching time set stores the indices of all historical time points where the operating condition similarity exceeds a preset threshold, along with their corresponding operating condition features, decision actions, and reward records. Historical decision actions characterize the actual control parameter adjustments performed at a historical time point; historical rewards characterize the reward value obtained after executing the historical decision action, calculated based on the reward function of each game agent. Optimal action preference characterizes the tendency of each player agent to maximize their payoff in the current state. This optimal action preference includes action type and action magnitude. Action type indicates which control parameters among drilling pressure, rotational speed, and displacement should be adjusted, while action magnitude indicates the specific adjustment amount for each control parameter. Methods for determining optimal action preference include, but are not limited to: sorting the historical payoffs of each historical moment in the matching time set from highest to lowest, selecting the historical decision actions corresponding to one or more historical moments with the highest historical payoffs as a candidate action set; taking a weighted average of the historical payoffs for each action in the candidate action set to obtain the action magnitude in the optimal action preference; and selecting the control parameter type with the highest frequency of occurrence from the control parameter adjustment objects of each action in the candidate action set as the action type of the optimal action preference. Each player agent combines the action type and action magnitude from their determined optimal action preference into corresponding candidate actions. By calculating the similarity of operating conditions, the historical moment closest to the current operating condition is identified, and the successful decision-making experience at that historical moment is transferred to the current decision. This improves the basis for generating candidate actions from a single threshold comparison to experience transfer based on overall operating condition matching, making it more adaptable and reasonable when operating conditions are complex and changeable.

[0062] Each game agent can generate its own initial action preferences based on the current drilling state parameter sequence and operational experience data. These initial action preferences include preferences for drilling pressure adjustment, rotation speed adjustment, and displacement adjustment. Each game agent can broadcast its generated initial action preferences to the other two game agents and receive initial action preferences broadcast by the other two game agents. Based on the initial action preferences received from the other two game agents, each game agent determines whether there is a directional conflict between its own initial action preference and the other initial action preferences. If a directional conflict exists, each game agent corrects the conflicting adjustment amounts in its own initial action preference according to the historical priority weights of each game objective in the operational experience data, generating corrected candidate actions. If no directional conflict exists, each game agent outputs its own initial action preference as a candidate action.

[0063] Initial action preferences characterize the initial action tendencies independently generated by each player agent without considering the influence of other agents. The generation of these initial action preferences includes, but is not limited to, the action values ​​output by each agent after forward propagation calculation of the current input based on its own policy network, including preferences for drilling pressure adjustment, rotational speed adjustment, and displacement adjustment. Broadcasting characterizes the communication behavior of sending its own generated initial action preference data to other players agents. This broadcasting is completed within the same decision window, ensuring that each agent can obtain the action preference information of other agents before generating candidate actions. Directional conflict characterizes the situation where two players agents give opposite adjustment directions for the same control parameter in their initial action preferences. Methods for determining whether a directional conflict exists include, but are not limited to: comparing the signs of the adjustment amounts of each control parameter in the initial action preferences of the first, second, and third players. If, for any control parameter, two or more agents have opposite adjustment signs, a directional conflict is determined to exist. Historical priority weights are used to characterize the relative importance of each game objective in historical decisions within the operational experience data. These weights are determined based on the average proportion of each game objective's payoff at each historical moment to the total payoff of all game objectives within the operational experience data. Methods for correcting conflicting adjustment amounts include, but are not limited to: for control parameters with directional conflicts, obtaining the absolute value of each player's adjustment amount for that control parameter; redistributing the absolute values ​​of the adjustment amounts based on the historical priority weights of each player, so that players with higher historical priority weights occupy a larger share of the adjustment amounts, while players with lower historical priority weights have their absolute adjustment amounts reduced accordingly; and unifying the sign of the adjustment amounts to the sign direction corresponding to the player with the highest historical priority weight. The corrected candidate actions characterize the final candidate actions output by each player after negotiation among them. These corrected candidate actions no longer conflict with each player in the adjustment direction of each control parameter. By introducing an action preference broadcasting and negotiation mechanism among agents during the candidate action generation stage, each agent can perceive the action intentions of other agents when generating candidate actions, and make early corrections to conflicting adjustment amounts, thereby reducing the number of conflicts in subsequent game iterations and improving game convergence efficiency.

[0064] In some embodiments, each candidate action can be modified based on constraint boundaries.

[0065] For each control parameter adjustment in each candidate action, obtain the current value of the control parameter; sum the adjustment amount with the current value to obtain the estimated execution value of the control parameter; determine whether the estimated execution value exceeds the envelope interval of the control parameter. If the estimated execution value exceeds the envelope interval but does not exceed the risk boundary, compress and correct the adjustment amount so that the corrected estimated execution value equals the boundary value in the envelope interval closest to the estimated execution value. That is, when the estimated execution value is higher than the upper limit of the envelope interval, reduce the adjustment amount to make the estimated execution value equal to the upper limit; when the estimated execution value is lower than the lower limit of the envelope interval, increase the adjustment amount to make the estimated execution value equal to the lower limit. If the estimated execution value exceeds the risk boundary, i.e., the estimated execution value is higher than the upper limit of the risk boundary or lower than the lower limit of the risk boundary, mark the candidate action as invalid and remove it, and the control parameter is not adjusted within the decision window. If the estimated execution value is within the envelope interval, do not correct the adjustment amount and maintain the original adjustment amount output. After the above correction, the adjustment of each control parameter in each candidate action satisfies the following: the estimated execution value after superimposing the current value does not exceed the risk boundary and is located as close as possible to the envelope interval.

[0066] In some embodiments, the rock-breaking efficiency gain, wellbore flow safety gain, and tubing dynamics safety gain corresponding to the corrected first candidate action, second candidate action, and third candidate action can be calculated respectively.

[0067] The rock-breaking efficiency gain characterizes the expected contribution of the first candidate action to the drill bit's rock-breaking efficiency. A higher gain value indicates a greater expected increase in mechanical drilling speed and a smaller increase in drill bit wear. The wellbore flow safety gain characterizes the expected improvement in the pressure balance and cutter cleanliness of the drilling fluid circulation system after executing the second candidate action. A higher gain value indicates that the expected bottomhole pressure is closer to the center of the safe pressure window and that the cutter carrying capacity is more sufficient. The tubing dynamics safety gain characterizes the expected improvement in the drill string's resistance to vibration, buckling, and fatigue failure during downhole movement after executing the third candidate action. A higher gain value indicates smoother expected tubing operation, lower buckling risk, and greater extension of fatigue life. The methods for calculating the three benefits include, but are not limited to: substituting the adjustment amounts of each control parameter in the modified first candidate action into the rock-breaking efficiency benefit function for calculation, and outputting a scalar value as the rock-breaking efficiency benefit; substituting the adjustment amounts of each control parameter in the modified second candidate action into the wellbore flow safety benefit function for calculation, and outputting a scalar value as the wellbore flow safety benefit; and substituting the adjustment amounts of each control parameter in the modified third candidate action into the tubing dynamics safety benefit function for calculation, and outputting a scalar value as the tubing dynamics safety benefit.

[0068] In some embodiments, each candidate action includes one or more of the following: drilling pressure adjustment, rotation speed adjustment, and displacement adjustment.

[0069] The drilling pressure adjustment amount characterizes the magnitude and direction of changes in the drilling pressure control parameter; a positive value indicates an increase in drilling pressure, and a negative value indicates a decrease. The rotational speed adjustment amount characterizes the magnitude and direction of changes in the rotational speed control parameter; a positive value indicates an increase in rotational speed, and a negative value indicates a decrease. The displacement adjustment amount characterizes the magnitude and direction of changes in the displacement control parameter; a positive value indicates an increase in displacement, and a negative value indicates a decrease. The candidate actions of different game agents may include different types of adjustment amounts. The first game agent's candidate actions are for adjusting drilling pressure and rotational speed, the second game agent's candidate actions are for adjusting displacement, and the third game agent's candidate actions are for adjusting drilling pressure and rotational speed.

[0070] Based on the drill pressure adjustment and rotation speed adjustment in the first candidate action, the expected mechanical drilling rate and expected drill bit wear increment corresponding to the first candidate action can be determined. Based on the expected mechanical drilling rate and drill bit wear increment, the drilling rate benefit and wear cost are calculated respectively. Based on the drilling rate benefit and wear cost, the rock breaking efficiency benefit is determined; the rock breaking efficiency benefit is positively correlated with the drilling rate benefit and negatively correlated with the wear cost.

[0071] The expected mechanical drilling rate (MRR) characterizes the predicted MRR value after implementing drill pressure (DP) and rotational speed (RPS) adjustments. This expected MRR is calculated by superimposing the DP and RPS adjustments onto the current DP and RPS values ​​and then substituting this result into the MRR prediction model. The expected drill bit wear increment characterizes the predicted increase in drill bit wear after implementing DP and RPS adjustments. This expected drill bit wear increment is calculated by substituting the superimposed DP and RPS values ​​into the drill bit wear rate model. Based on the expected MRR and expected drill bit wear increment, drilling speed gain and wear cost are calculated respectively. Drill speed gain characterizes the improvement effect of the expected MRR relative to the current MRR, and is positively correlated with the magnitude of the expected MRR. Wear cost characterizes the effect of the expected drill bit wear increment on drill bit lifespan, and is positively correlated with the magnitude of the expected drill bit wear increment. Based on the drilling speed gain and wear cost, the rock breaking efficiency gain is determined; this rock breaking efficiency gain is positively correlated with the drilling speed gain and negatively correlated with the wear cost. Methods for determining rock-breaking efficiency benefits include, but are not limited to, weighted differences between drilling speed benefits and wear costs. Specifically, the rock-breaking efficiency benefit equals the drilling speed benefit multiplied by a first weighting coefficient minus the wear cost multiplied by a second weighting coefficient, where both the first and second weighting coefficients are positive. By separately calculating the positive benefits from increased mechanical drilling speed and the negative costs from increased drill bit wear, the rock-breaking efficiency benefit comprehensively reflects the net benefit of candidate actions in achieving the rock-breaking efficiency optimization objective, avoiding a one-sided pursuit of high drilling speed while neglecting drill bit life consumption.

[0072] Based on the displacement adjustment in the second candidate action, the expected bottomhole pressure and expected annular return velocity corresponding to the second candidate action can be determined. The pressure deviation between the expected bottomhole pressure and the preset safe pressure window boundary is calculated. Based on the pressure deviation and the expected annular return velocity, the bottomhole pressure safety benefit and cutterhead carrying capacity benefit are calculated respectively; the bottomhole pressure safety benefit is negatively correlated with the pressure deviation. Based on the bottomhole pressure safety benefit and the cutterhead carrying capacity benefit, the wellbore flow safety benefit is determined.

[0073] The expected bottomhole pressure characterizes the predicted bottomhole pressure value after the displacement adjustment is implemented. This expected bottomhole pressure is calculated by superimposing the displacement adjustment amount onto the current displacement value and then substituting it into the hydraulic model. The hydraulic model includes calculations of pressure loss along the flow path and annular pressure distribution. The expected annular return velocity characterizes the predicted annular return velocity value after the displacement adjustment is implemented. This expected annular return velocity is calculated by dividing the superimposed displacement value by the annular cross-sectional area. The pressure deviation between the expected bottomhole pressure and the preset safe pressure window boundary is calculated. This pressure deviation characterizes the degree to which the expected bottomhole pressure deviates from the center value of the safe pressure window. The smaller the pressure deviation, the more stable the bottomhole pressure and the closer it is to the ideal operating state. Based on the pressure deviation and the expected annular return velocity, the bottom hole pressure safety benefit and the cutterhead carrying capacity benefit are calculated respectively. The bottom hole pressure safety benefit is negatively correlated with the pressure deviation, meaning that the smaller the pressure deviation, the higher the bottom hole pressure safety benefit. The cutterhead carrying capacity benefit characterizes the contribution of the expected annular return velocity to the cutterhead carrying capacity, and it is positively correlated with the magnitude of the expected annular return velocity. Based on the bottom hole pressure safety benefit and the cutterhead carrying capacity benefit, the wellbore flow safety benefit is determined. Methods for determining the wellbore flow safety benefit include, but are not limited to, weighted summation of the bottom hole pressure safety benefit and the cutterhead carrying capacity benefit, i.e., the wellbore flow safety benefit equals the bottom hole pressure safety benefit multiplied by a first weighting coefficient plus the cutterhead carrying capacity benefit multiplied by a second weighting coefficient, where both the first and second weighting coefficients are positive numbers. By simultaneously considering the benefits of both bottom hole pressure stability and cutterhead carrying capacity, the wellbore flow safety benefit comprehensively reflects the overall effect of candidate actions on the wellbore flow safety optimization objective, avoiding other safety risks caused by optimizing a single indicator.

[0074] Based on the drilling pressure and rotational speed adjustments in the third candidate action, the corresponding expected torque fluctuation, expected vibration intensity, expected buckling margin, and expected cumulative fatigue increment can be determined. Based on these factors, the torque stability benefit, vibration suppression benefit, buckling safety benefit, and fatigue life benefit can be calculated. The torque stability benefit is negatively correlated with the expected torque fluctuation, the vibration suppression benefit is negatively correlated with the expected vibration intensity, the buckling safety benefit is positively correlated with the expected buckling margin, and the fatigue life benefit is negatively correlated with the expected cumulative fatigue increment. Finally, the string dynamic safety benefit is determined based on these benefits.

[0075] Expected torque fluctuation characterizes the predicted torque fluctuation after implementing drill pressure and rotational speed adjustments. This expected torque fluctuation is calculated by substituting the superimposed drill pressure and rotational speed values ​​into the string dynamics model. The degree of torque fluctuation is characterized by the standard deviation or peak value of the torque signal. Expected vibration intensity characterizes the predicted drill string vibration severity after implementing drill pressure and rotational speed adjustments. This expected vibration intensity is calculated by substituting the superimposed drill pressure and rotational speed values ​​into the vibration model. Expected buckling margin characterizes the predicted buckling safety margin of the string after implementing drill pressure and rotational speed adjustments. This expected buckling margin is calculated by substituting the superimposed drill pressure values ​​into the string buckling model; a larger buckling margin indicates greater safety. Expected cumulative fatigue increment characterizes the predicted increase in string fatigue damage after implementing drill pressure and rotational speed adjustments. This expected cumulative fatigue increment is calculated by substituting the superimposed drill pressure and rotational speed values ​​and their rate of change into the fatigue accumulation model. Based on the expected torque fluctuation, expected vibration intensity, expected buckling margin, and expected cumulative fatigue increment, the torque stability gain, vibration suppression gain, buckling safety gain, and fatigue life gain are calculated respectively. The torque stability gain characterizes the contribution of candidate actions to reducing torque fluctuation; it is negatively correlated with the magnitude of the expected torque fluctuation, i.e., the smaller the expected torque fluctuation, the higher the torque stability gain. The vibration suppression gain characterizes the contribution of candidate actions to suppressing vibration; it is negatively correlated with the magnitude of the expected vibration intensity, i.e., the smaller the expected vibration intensity, the higher the vibration suppression gain. The buckling safety gain characterizes the contribution of candidate actions to maintaining the buckling safety of the tubing; it is positively correlated with the magnitude of the expected buckling margin, i.e., the larger the expected buckling margin, the higher the buckling safety gain. The fatigue life gain characterizes the contribution of candidate actions to extending the fatigue life of the tubing; it is negatively correlated with the magnitude of the expected cumulative fatigue increment, i.e., the smaller the expected cumulative fatigue increment, the higher the fatigue life gain. Based on the torque stability gain, vibration suppression gain, buckling safety gain, and fatigue life gain, the dynamic safety gain of the tubing is determined. The methods for determining the benefits of tubular dynamic safety include, but are not limited to: multiplying the benefits of torque stability, vibration suppression, buckling safety, and fatigue life by their respective weighting coefficients and then summing them to obtain the benefits of tubular dynamic safety. Each weighting coefficient is a positive number, and the relative magnitude of the weighting coefficients is determined based on the degree of influence of each factor on tubular safety. This embodiment, by simultaneously considering the benefits of four dimensions—torque stability, vibration suppression, buckling safety, and fatigue life—ensures that the benefits of tubular dynamic safety comprehensively reflect the overall effect of candidate actions on the tubular safety optimization objective, avoiding potential safety hazards to other tubular components caused by optimization of a single dimension.

[0076] In some embodiments, the modified first candidate action, second candidate action, and third candidate action can be combined into a candidate combined action.

[0077] The adjustment amounts for drilling pressure, rotational speed, and displacement in the first candidate action can be obtained and used as the first drilling pressure adjustment, first rotational speed adjustment, and first displacement adjustment, respectively. Similarly, the adjustment amounts for drilling pressure, rotational speed, and displacement in the second candidate action can be obtained and used as the second drilling pressure adjustment, second rotational speed adjustment, and second displacement adjustment, respectively. The adjustment amounts for drilling pressure, rotational speed, and displacement in the third candidate action can also be obtained and used as the third drilling pressure adjustment, third rotational speed adjustment, and third displacement adjustment, respectively. Based on the preset weight coefficients of each game agent, the first, second, and third drilling pressure adjustments can be weighted and summed to obtain the combined drilling pressure adjustment. The first, second, and third rotational speed adjustments can be weighted and summed to obtain the combined rotational speed adjustment. The first, second, and third displacement adjustment are weighted and summed to obtain the combined displacement adjustment. The drilling pressure combination adjustment amount, rotation speed combination adjustment amount, and displacement combination adjustment amount can be combined into candidate combination actions.

[0078] The drill pressure adjustment amounts for the first, second, and third candidate actions can be obtained separately, and their average value can be calculated as the average drill pressure adjustment amount. Similarly, the average rotational speed adjustment amount and the average displacement adjustment amount are obtained. Based on the current drill pressure value and the average drill pressure adjustment amount for the current single cycle, the drill pressure estimate, rotational speed estimate, and displacement estimate are calculated. It is then determined whether the drill pressure estimate, rotational speed estimate, and displacement estimate exceed the constraint boundaries. If the drill pressure estimate exceeds the upper limit of the drill pressure in the constraint boundaries, the average drill pressure adjustment amount is compressed to make the drill pressure estimate equal to the upper limit. If the drill pressure estimate is lower than the lower limit of the drill pressure in the constraint boundaries, the average drill pressure adjustment amount is compressed to make the drill pressure estimate equal to the lower limit. Similarly, the average rotational speed adjustment amount and the average displacement adjustment amount are compressed. Based on the compressed average drill pressure adjustment amount, average rotational speed adjustment amount, and average displacement adjustment amount, candidate combination actions are generated.

[0079] It can also calculate the similarity between the current drilling state parameter sequence and the historical working condition state parameter sequences in the working condition experience data, and select the historical working condition with the highest similarity as the most similar historical working condition. It can obtain the historical optimal rock-breaking efficiency benefit, historical optimal wellbore flow safety benefit, and historical optimal tubing dynamics safety benefit corresponding to the most similar historical working condition. Based on the relative magnitudes of the historical optimal rock-breaking efficiency benefit, historical optimal wellbore flow safety benefit, and historical optimal tubing dynamics safety benefit, the initial values ​​of the first, second, and third weight coefficients are determined, and the initial values ​​are positively correlated with each historical optimal benefit. It can obtain the historical decision accuracy of each game agent within the current decision-making cycle, and correct the initial values ​​of the first, second, and third weight coefficients based on the historical decision accuracy to generate corrected weight coefficients for each game agent. Based on the corrected weight coefficients of each game agent, the adjustment amounts of each control parameter in the first, second, and third candidate actions are weighted and summed to generate candidate combination actions.

[0080] In some embodiments, the consistency benefit corresponding to candidate combination actions can be calculated.

[0081] Candidate combined actions represent a complete drilling control parameter adjustment scheme formed by combining the control parameter adjustments from the first, second, and third candidate actions. This candidate combined action includes adjustments for drilling pressure, rotational speed, and displacement. Recommended parameter ranges represent the value ranges of each drilling control parameter recommended by the knowledge base under the current operating conditions. These recommended parameter ranges are derived from historical optimization strategies in the engineering experience knowledge base that match the current formation characteristics and operating condition type. They include recommended ranges for drilling pressure, rotational speed, and displacement, each with upper and lower limits. Consistency gains represent the degree of conformity between candidate combined actions and the recommended parameter ranges in the knowledge base. A higher consistency gain indicates that the candidate combined action is closer to the control parameter range recommended by the knowledge base, meaning that the candidate combined action has higher rationality and feasibility at the empirical level.

[0082] The recommended parameter range corresponding to the current single cycle can be determined. The deviation between each drilling control parameter in the candidate combination actions and the corresponding parameter in the recommended parameter range is calculated. Based on the deviation of each drilling control parameter, the matching degree between the candidate combination actions and the recommended parameter range is calculated. The consistency reward is determined based on the matching degree; the consistency reward is positively correlated with the matching degree.

[0083] The recommended parameter range is determined, but is not limited to, the following method: Based on the formation feature vector and working condition type of the current single cycle, a matching historical optimization strategy is retrieved from the engineering experience knowledge base. The value distribution range of each drilling control parameter recorded in the historical optimization strategy is obtained, and this value distribution range is used as the recommended parameter range. The deviation degree is used to characterize the distance between the estimated execution value after adding the combined adjustment amount of each drilling control parameter in the candidate combined action to the current value and the boundary value of the corresponding parameter in the recommended parameter range. Methods for calculating the degree of deviation include, but are not limited to: for each control parameter in the candidate combined action, the combined adjustment amount of the control parameter is superimposed with the current value to obtain the estimated execution value; the distance between the estimated execution value and the upper and lower limits of the recommended parameter range for the control parameter is calculated; when the estimated execution value is within the recommended parameter range, the degree of deviation is zero or negative; when the estimated execution value is higher than the upper limit of the recommended parameter range, the degree of deviation is the difference between the estimated execution value and the upper limit; when the estimated execution value is lower than the lower limit of the recommended parameter range, the degree of deviation is the difference between the lower limit and the estimated execution value. The matching degree is used to characterize the overall degree of conformity between the estimated execution values ​​of each control parameter in the candidate combined action and the recommended parameter range. A higher matching degree indicates that the estimated execution values ​​are more concentrated within the recommended parameter range. Methods for calculating the matching degree include, but are not limited to: normalizing the degree of deviation of each control parameter and taking the average value; the negative correlation value or reciprocal of this average value is used as the matching degree, i.e., the smaller the degree of deviation, the higher the matching degree; the larger the degree of deviation, the lower the matching degree. Consistency payoff is positively correlated with matching degree. Matching degree can be used as consistency payoff, or it can be calculated by multiplying the matching degree by a preset amplification factor. Consistency payoff reaches its maximum when the estimated execution values ​​of all control parameters in the candidate action combinations are within the recommended parameter range. When the estimated execution value of any control parameter deviates from the recommended parameter range, consistency payoff monotonically decreases with increasing deviation. Through the quantitative calculation of deviation and the comprehensive mapping of matching degree, consistency payoff can accurately quantify the degree of conformity between candidate action combinations and knowledge base experience. This numerical approach guides game optimization towards the safe region of empirical verification, while preserving the possibility of exploration outside the knowledge base's recommended range (achieved through continuously decreasing consistency payoff rather than binarization).

[0084] In some embodiments, the combined benefits can be determined based on rock-breaking efficiency benefits, wellbore flow safety benefits, tubing dynamics safety benefits, and consistency benefits.

[0085] The system can obtain preset weighting coefficients for rock breaking efficiency, wellbore flow safety, tubing dynamics safety, and consistency. The product of the rock breaking efficiency benefit and the rock breaking efficiency weighting coefficient is used as the rock breaking efficiency weighted value; the product of the wellbore flow safety benefit and the wellbore flow safety weighting coefficient is used as the wellbore flow safety weighted value; the product of the tubing dynamics safety benefit and the tubing dynamics safety weighting coefficient is used as the tubing dynamics safety weighted value; and the product of the consistency benefit and the consistency weighting coefficient is used as the consistency weighted value. The combined benefit is generated by adding these weighted values.

[0086] It is also possible to separately obtain the first safety threshold value corresponding to the wellbore flow safety benefit and the second safety threshold value corresponding to the tubing dynamics safety benefit. It is then determined whether the wellbore flow safety benefit is lower than the first safety threshold value, or whether the tubing dynamics safety benefit is lower than the second safety threshold value. If so, the combined benefit is set to a preset negative maximum value, thus excluding the current candidate combination action. If not, the product of the rock-breaking efficiency benefit, the wellbore flow safety benefit, the tubing dynamics safety benefit, and the consistency benefit is taken as the combined benefit.

[0087] In some embodiments, the drilling control parameter sequence for the current single cycle of the target well can be determined based on the combined benefits.

[0088] The adjustment amounts of each control parameter corresponding to the candidate combination action with the highest combined benefit in the current decision window can be superimposed on the current value to generate the drilling pressure value, rotation speed value and displacement value in the decision window, which are then arranged in chronological order to form a drilling control parameter sequence.

[0089] Furthermore, based on joint payoffs, it can be determined whether the combined actions satisfy the Nash equilibrium condition. The Nash equilibrium condition includes a first condition and a second condition. The first condition is that the payoff increase corresponding to each player agent changing its own candidate actions does not exceed a threshold. The second condition is that the candidate combined actions are not Pareto-dominated by other candidate combined actions. If satisfied, the combined actions are used as the drilling control parameter sequence for the current single cycle of the target well. If not satisfied, the strategies of each player agent are iteratively updated until the generated combined actions satisfy the Nash equilibrium condition.

[0090] The Nash equilibrium condition is used to characterize the criteria for determining whether a combination of actions reaches a stable state. This Nash equilibrium condition includes a first condition and a second condition. The first condition is that the increase in payoff for each player agent after individually changing its candidate actions does not exceed a threshold. This threshold characterizes the allowable tolerance for payoff increases and is a preset positive number. The calculation of the payoff increase after individually changing a player agent's candidate actions includes, but is not limited to: keeping the candidate actions of the other two players unchanged, exhaustively changing the candidate action of the current player agent within its action space or changing it according to multiple alternative actions output by the policy network, substituting the new combined action obtained after the change into the payoff function of the player agent to calculate the new payoff value, and subtracting the current payoff value from the new payoff value to obtain the payoff increase. The second condition is that the combined action is not Pareto-dominated by other candidate combined actions. Pareto dominance characterizes the situation where a combined action is strictly superior to another combined action in the payoff of at least one player agent and not inferior to that other combined action in the payoff of all other players agents. A combination of actions not being Pareto-dominated by other candidate combinations can be defined as follows: there are no other candidate combinations that can achieve a payoff no lower than the current combination for all players, and at least one player can achieve a payoff significantly higher than the current combination. Methods for determining whether a combination satisfies the Nash equilibrium condition include, but are not limited to: checking whether the first and second conditions are met separately; if both conditions are met, the combination is deemed to satisfy the Nash equilibrium condition. If the condition is met, the combination is output as the drilling control parameter sequence for the current single cycle of the target well. If the condition is not met, i.e., either the first or second condition is not met, an iterative update process begins: if the first condition is not met, the player with the largest payoff increase is selected, and its policy network parameters are updated according to the policy gradient direction, so that its next output candidate action can achieve a higher payoff value; if the second condition is not met, i.e., there are other candidate combinations that Pareto-dominate the current combination, the current combination is replaced with the Pareto-dominated combination, and the next round of equilibrium determination begins. Iterative updates continue until the generated combination simultaneously satisfies the first and second conditions, or the preset maximum number of iterations is reached. By setting Nash equilibrium conditions that include threshold constraints on profit increase and Pareto optimality constraints, the game convergence result can simultaneously satisfy individual rationality (no agent has a unilateral deviation motivation) and collective rationality (the combined actions are not Pareto-dominated by other options), ensuring that the final output drilling control parameter sequence reaches a stable and optimal state in the sense of multi-objective game.

[0091] S104: Based on drilling experience data from multiple historical single-cycle drilling operations, the drilling control parameters for the current decision cycle in the drilling control parameter sequence are fine-tuned to obtain the drilling optimization instructions for the current decision cycle of the target well; each single-cycle includes multiple decision cycles.

[0092] In some embodiments, each single root cycle includes multiple decision cycles. A decision cycle represents the smallest time unit for the execution control layer to adjust parameters and issue commands. The length of this decision cycle is fixed at the second or millisecond level, determined by the sampling rate of the drilling data acquisition and monitoring system and the response speed of the actuator. Within each decision cycle, the execution control layer completes one complete operation: acquiring drilling status parameters, calculating execution deviations, correcting control parameters, and issuing optimization commands. The inclusion relationship between a single root cycle and a decision cycle is as follows: the length of a single root cycle is equal to the sum of the lengths of all decision cycles within that single root cycle. Each decision cycle is consecutive on the time axis and does not overlap, collectively covering the complete time span of that single root cycle. This embodiment defines a many-to-one inclusion relationship between a single drilling cycle and a decision cycle, enabling the same decision-making process to be hierarchically decomposed at two different time granularities. The drilling control parameter sequence at the single drilling scale is decomposed into a sequence of optimized instructions executed point by point at the decision scale, ensuring logical consistency from macro planning to micro execution. At the same time, it provides a fixed-frequency instruction update mechanism for the execution control layer, enabling the optimized instructions to be dynamically fine-tuned on a decision cycle basis based on real-time feedback.

[0093] In some embodiments, the drilling status parameters of the target well in the current decision-making cycle can be compared with the corresponding drilling status parameters in the drilling experience data to obtain the execution deviation of the drilling control parameters.

[0094] Drilling experience data is used to characterize standard drilling state parameters recorded under similar working conditions in the same historical drilling cycle. This drilling experience data is stored in a single-cycle real-time database, containing the standard state parameter values ​​and their corresponding control parameter records for each historical decision cycle. Execution deviation is used to characterize the degree of deviation of the actual drilling state parameters in the current decision cycle from the standard drilling state parameters in the drilling experience data. This execution deviation is a vector quantity, including drill pressure execution deviation, rotational speed execution deviation, and displacement execution deviation. Methods for comparing drilling state parameters with the corresponding drilling state parameters in the drilling experience data include, but are not limited to: obtaining the measured values ​​of each drilling state parameter in the current decision cycle; obtaining the standard state parameter values ​​in the drilling experience data aligned with the current decision cycle; and subtracting the measured values ​​from the standard values ​​item by item according to parameter type to obtain the execution deviation of each control parameter. When the measured value is higher than the standard value, the execution deviation is positive, indicating that the current execution intensity is higher than the experience standard; when the measured value is lower than the standard value, the execution deviation is negative, indicating that the current execution intensity is lower than the experience standard; when the two are equal, the execution deviation is zero. By comparing the current actual drilling status with historical experience standards, the deviation information of the execution process can be quantitatively obtained, providing a quantitative basis for subsequent parameter correction and giving the correction operation a clear direction and adjustment range reference.

[0095] In some embodiments, the drilling control parameters in the current decision cycle of the drilling control parameter sequence can be corrected based on the execution deviation.

[0096] The proportional correction is calculated based on the execution deviation and a preset proportional coefficient. The execution deviations are accumulated to obtain the cumulative deviation, and the integral correction is calculated based on the cumulative deviation and a preset integral coefficient. The difference between the execution deviations of two adjacent control cycles is obtained, and the differential correction is calculated based on the difference and a preset differential coefficient. The proportional, integral, and differential corrections are added together to obtain the comprehensive correction. Based on the comprehensive correction, the drilling control parameters for the current control cycle in the drilling control parameter sequence are corrected to generate the corrected drilling control parameters.

[0097] The proportional gain is used to characterize the immediate response strength to the current execution deviation. This proportional gain is positive; the larger the proportional gain, the more drastic the response of the correction to the current deviation. The proportional correction is used to characterize the correction value directly calculated from the current execution deviation. This proportional correction is equal to the product of the execution deviation and the proportional gain. The cumulative deviation is used to characterize the time integral effect of the execution deviation from the start of the current decision cycle to the current time. This cumulative deviation is obtained by summing the execution deviations of each historical control cycle and is used to eliminate steady-state cumulative error. The integral gain is used to characterize the response strength to the cumulative deviation. This integral gain is positive; the larger the integral gain, the more drastic the response of the correction to the cumulative deviation. The integral correction is used to characterize the correction value calculated from the cumulative deviation. This integral correction is equal to the product of the cumulative deviation and the integral gain and is used to eliminate long-term steady-state deviations. The difference between the execution deviations of two adjacent control cycles characterizes the rate of change of the execution deviation. This difference is equal to the execution deviation of the current control cycle minus the execution deviation of the previous control cycle. The larger the difference, the faster the deviation changes. The differential coefficient characterizes the response strength to the rate of change of the executed deviation. This coefficient is positive; the larger the coefficient, the more drastic the response of the correction to the rate of change of the deviation. The differential correction value characterizes the correction calculated based on the rate of change of the executed deviation. This value is equal to the product of the difference in executed deviations between two adjacent control cycles and the differential coefficient, used to suppress rapid changes and oscillations in the deviation. The comprehensive correction value characterizes the total correction value obtained by combining the proportional, integral, and differential correction components. This value is equal to the sum of the proportional, integral, and differential correction values. These three components can be positive or negative; the sign and magnitude of the comprehensive correction value depend on the superposition of the components. Methods for correcting the drilling control parameters of the current control cycle include, but are not limited to: superimposing the comprehensive correction value onto the drilling control parameters of the current control cycle, i.e., the corrected drilling control parameters equal the current drilling control parameters plus the comprehensive correction value. When the comprehensive correction value is positive, the corresponding value is added to the current control parameters; when the comprehensive correction value is negative, the corresponding value is subtracted from the current control parameters. After generating the corrected drilling control parameters, they are output as the final execution parameters for the current control cycle. Through the comprehensive calculation of the three correction quantities, the correction operation simultaneously possesses the capabilities of immediate response, elimination of steady-state errors, and suppression of oscillations. This ensures the correction speed while avoiding over-correction and periodic fluctuations, thereby improving the stability and accuracy of the execution control.

[0098] In some embodiments, drilling optimization instructions for the target well in the current decision-making cycle can be generated based on the modified drilling control parameters.

[0099] The corrected drilling control parameters can be obtained, including corrected pressure on drill bit (PBD), rotational speed (SSD), and displacement rate (DMR). These values ​​are then converted into control signal formats corresponding to each actuator within the current control cycle of the target well, generating PBD control signals, SSD control signals, and DMR control signals. These signals are then encapsulated into a drilling optimization command for the current control cycle. This optimization command is sent to the drilling rig control system of the target well to control each actuator to perform drilling operations according to the corrected drilling control parameters.

[0100] Alternatively, the current drilling pressure, rotational speed, and displacement value of the target well in the current control cycle can be obtained. The corrected drilling pressure adjustment is added to the current drilling pressure value to obtain the target drilling pressure value. The corrected rotational speed adjustment is added to the current rotational speed value to obtain the target rotational speed value. The corrected displacement value is added to the current displacement value to obtain the target displacement value. Based on the target drilling pressure, target rotational speed, and target displacement value, drilling optimization instructions for the current control cycle are generated.

[0101] In some embodiments, determining the recommended parameter range corresponding to the current single drilling cycle may further include: retrieving matching historical optimization strategies from an engineering experience knowledge base based on the formation characteristics and operating conditions corresponding to the current single drilling cycle; the historical optimization strategies include the experience-recommended parameter range. Based on the drilling engineering parameters of the current single drilling cycle, and according to the physical model in the mechanism knowledge base, inferring the physical mechanism constraint boundaries of the drilling control parameters. Based on the safety thresholds stored in the risk boundary knowledge base, determining the safety constraint boundaries of the drilling control parameters. Based on the physical mechanism constraint boundaries and the safety constraint boundaries, correcting the experience-recommended parameter range to obtain the recommended parameter range.

[0102] Formation characteristics are used to characterize the geological conditions of the drilling section in the current single drilling cycle. These characteristics include rock drillability values, formation pressure coefficients, and lithology category codes. Operating condition characteristics characterize the current drilling operation mode, including type identifiers such as normal drilling, directional drilling, reaming, and tripping. An engineering experience knowledge base stores engineering experience data from multiple historical wells. Each set of engineering experience data records, in structured form, the actual control parameter combinations used in historical wells under specific formation conditions and operating condition types, and their corresponding drilling effects. Historical optimization strategies characterize control parameter combinations that have been validated as effective under conditions matching the current formation conditions and operating condition types in historical wells. These strategies include recommended parameter ranges with upper and lower limits. Drilling engineering parameters characterize the input parameters of the engineering environment and design conditions for the current drilling operation. These parameters include well depth, formation drillability, mud density, formation pore pressure, and formation fracture pressure. The mechanistic knowledge base stores mathematical models of the drilling process based on physical mechanisms. These physical models include hydraulic models, drill bit wear models, and tubing dynamics models. Physical mechanism constraint boundaries characterize the boundary values ​​of drilling control parameters calculated by the physical models based on current drilling engineering parameters within the physically feasible domain. These physical mechanism constraint boundaries dynamically change with variations in drilling engineering parameters. The risk boundary knowledge base stores pre-defined safety limits, including safe pressure window limits, tubing buckling limits, and equipment power limits. Each safety limit is a static constant that does not change with operating conditions. Safety constraint boundaries characterize the absolute safety boundaries of the drilling control parameters determined by the safety limits in the risk boundary knowledge base. These safety constraint boundaries are insurmountable hard boundary values. Methods for correcting the empirically recommended parameter range include, but are not limited to: using the narrower of the physical mechanism constraint boundary and the safety constraint boundary as the corrected upper and lower boundaries, and truncating the portion of the empirically recommended parameter range that exceeds this corrected boundary to the boundary value, so that the corrected recommended parameter range lies entirely within the range defined by the physical mechanism constraint boundary and the safety constraint boundary. This embodiment generates the recommended parameter range by integrating information from three dimensions: engineering experience knowledge, physical mechanism constraints, and safety thresholds. This ensures that the recommendation results simultaneously possess empirical rationality, physical feasibility, and safety assurance, avoiding recommendation bias or safety risks caused by relying on only a single information source.

[0103] In some embodiments, based on the formation characteristics and working condition characteristics corresponding to the current single drilling cycle, matching historical optimization strategies are retrieved from the engineering experience knowledge base. Specifically, this includes: generating a current formation feature vector based on the formation feature parameters and working condition type obtained from the current single drilling cycle; calculating the vector distance between the current formation feature vector and the historical formation feature vectors associated with each historical experience entry in the engineering experience knowledge base; selecting several historical experience entries with the smallest vector distance as a candidate matching set; using the reciprocal of the vector distance between each historical experience entry in the candidate matching set as a weight, weighted averaging the recommended drilling pressure, recommended rotation speed, and recommended displacement values ​​in each historical experience entry to obtain the center value of the experience recommended parameter interval; and generating the experience recommended parameter interval by using the minimum value of the same control parameter in each historical experience entry in the candidate matching set as the lower limit of the experience recommended parameter interval and the maximum value as the upper limit of the experience recommended parameter interval. The formation feature parameters are used to characterize the geological conditions of the current drilling section, and these parameters include rock drillability values, formation pressure coefficients, and lithology category codes. The working condition type characterizes the current drilling operation mode, including normal drilling, directional drilling, reaming, and tripping. The historical formation feature vector characterizes the geological condition attributes of the well section corresponding to each historical experience entry in the engineering experience knowledge base. This historical formation feature vector has the same dimension as the current formation feature vector. The vector distance characterizes the similarity between the current formation feature vector and the historical formation feature vector. This vector distance is calculated using Euclidean distance; a smaller vector distance indicates greater similarity between the two formation features. The candidate matching set stores several historical experience entries with the smallest vector distance; the size of this candidate matching set is a preset value. By calculating vector distance and using a weighted average, multiple similar historical experiences are retrieved from the engineering experience knowledge base and integrated. This allows the experience recommendation parameter range to comprehensively reflect successful experiences under multiple similar working conditions, avoiding the random bias caused by a single historical experience.

[0104] In some embodiments, based on the drilling engineering parameters of the current single-cycle drilling and the physical model in the mechanistic knowledge base, the physical mechanistic constraint boundaries of the drilling control parameters are inferred. Specifically, this includes: obtaining the drilling engineering parameters of the current single-cycle drilling, which include well depth, formation drillability, mud density, formation pore pressure, and formation fracture pressure; inputting the drilling engineering parameters into the hydraulic model in the mechanistic knowledge base to calculate the response curve of bottom hole pressure as a function of displacement under the current well depth conditions, and determining the physical feasible upper and lower limits of displacement; inputting the drilling engineering parameters into the drill bit wear model in the mechanistic knowledge base to calculate the drill bit wear rate under the current formation drillability and drill pressure conditions, and determining the physical feasible upper limit of drill pressure; inputting the drilling engineering parameters into the tubing dynamics model in the mechanistic knowledge base to calculate the critical buckling load of the tubing under the current well depth and drill pressure conditions, and determining the physical feasible upper limit of drill pressure and the physical feasible upper limit of rotational speed; and using the physical feasible upper and lower limits of each control parameter as the physical mechanistic constraint boundaries. The hydraulic model characterizes the physical relationship between displacement, mud density, well depth, and bottom hole pressure. This model is based on Bernoulli's equation and friction loss equation in fluid mechanics. The inputs are displacement, mud density, and well depth, and the output is the bottom hole pressure. The drill bit wear model characterizes the physical relationship between drill pressure, rotational speed, formation drillability, and drill bit wear rate. This model is based on the wear mechanism equation of PDC drill bit cutting teeth. The inputs are drill pressure, rotational speed, and formation drillability, and the output is the drill bit wear rate. The tubing string dynamics model characterizes the physical relationship between drill pressure, rotational speed, well depth, and the critical buckling load of the tubing string. This model is based on Euler's bar stability theory and the drill string vibration equation. The inputs are drill pressure, rotational speed, and well depth, and the output is the critical buckling load of the tubing string. The upper limit of physical feasibility characterizes the physically permissible maximum value of the control parameters calculated by the physical model; exceeding this value will lead to physical failure or equipment damage. The lower limit of physical feasibility characterizes the physically permissible minimum value of the control parameters calculated by the physical model; falling below this value will prevent the physical process from proceeding normally. By inputting the current drilling engineering parameters into multiple physical models to calculate the physical feasibility boundaries of each control parameter, the physical mechanism constraint boundaries can accurately reflect the real physical limitations under the current operating conditions, providing a reliable physical constraint basis for the subsequent correction of recommended parameter ranges.

[0105] In some embodiments, the safety constraint boundaries of drilling control parameters are determined based on safety thresholds stored in the risk boundary knowledge base. Specifically, this includes: reading the upper and lower limits of the safety pressure window, the tubing buckling limit, the maximum allowable torque of the equipment, and the rated power of the equipment from the risk boundary knowledge base; using the upper and lower limits of the safety pressure window as the safety constraint boundary for displacement, the tubing buckling limit as the upper limit of the safety constraint boundary for drilling pressure, and the maximum allowable torque of the equipment as the upper limit of the safety constraint boundary for rotational speed; when the upper limit of the safety constraint boundary for drilling pressure is lower than zero, setting the lower limit of the safety constraint boundary for drilling pressure to zero; when the upper limit of the safety constraint boundary for rotational speed is lower than zero, setting the lower limit of the safety constraint boundary for rotational speed to zero; and constructing the safety constraint boundaries using the upper and lower limits of the safety constraint boundaries for each control parameter. The upper limit of the safety pressure window is determined by the formation fracture pressure and is used to limit the maximum allowable value of the bottom hole pressure; the lower limit of the safety pressure window is determined by the formation pore pressure and is used to limit the minimum allowable value of the bottom hole pressure. The buckling limit of the tubing string is used to characterize the critical drill pressure value at which the tubing string will buckle and become unstable under axial load. This value is calculated based on buckling theory using the geometric and material parameters of the tubing string. The maximum permissible torque value of the equipment is used to characterize the maximum torque value that the surface drive equipment or downhole motor can safely output. This value is derived from the equipment's rated parameters. The rated power value of the equipment is used to characterize the maximum power value that the surface mud pump can safely provide, indirectly limiting the upper limit of the discharge capacity. Setting the lower limit of the drill pressure safety constraint boundary to zero indicates that the drill pressure cannot be negative; setting the lower limit of the rotational speed safety constraint boundary to zero indicates that the rotational speed cannot be negative. By reading static safety limits from the risk boundary knowledge base and constructing safety constraint boundaries for each control parameter accordingly, the safety constraint boundaries are made independent of changes in the current operating conditions, serving as a safety net in the decision-making process to ensure that the control parameters do not exceed the absolute safety limit under any circumstances.

[0106] In some embodiments, the empirically recommended parameter range is modified according to the physical mechanism constraint boundary and the safety constraint boundary to obtain a recommended parameter range. Specifically, this includes: taking the intersection of the physical mechanism constraint boundary and the safety constraint boundary parameter by parameter to obtain the hard constraint boundary for each drilling control parameter. The upper limit of this hard constraint boundary is the smaller of the upper limit of the physical mechanism constraint boundary and the upper limit of the safety constraint boundary, and the lower limit of the hard constraint boundary is the larger of the lower limit of the physical mechanism constraint boundary and the lower limit of the safety constraint boundary. The recommended drilling pressure range, recommended rotation speed range, and recommended displacement range in the empirically recommended parameter range are then compared with their corresponding hard constraint boundaries. The process involves several steps: First, if the upper limit of the recommended parameter range is higher than the upper limit of the hard constraint boundary, adjust the upper limit of the recommended range to match the upper limit of the hard constraint boundary. Second, if the lower limit of the recommended parameter range is lower than the lower limit of the hard constraint boundary, adjust the lower limit to match the lower limit of the hard constraint boundary. Third, if the entire recommended parameter range is within the hard constraint boundary, keep the recommended parameter range unchanged. Fourth, if there is no intersection between the recommended parameter range and the hard constraint boundary, use the hard constraint boundary as the recommended parameter range for that control parameter. Finally, combine the corrected recommended ranges of all control parameters into a single recommended parameter range. Intersection is performed parameter-by-parameter to ensure that the corrected recommended parameter range simultaneously meets both physical and safety constraints. The hard constraint boundary characterizes the final insurmountable boundary for each control parameter under both physical and safety limitations, and serves as the basis for correcting the recommended parameter range. Methods for truncating and correcting the recommended parameter range include, but are not limited to, checking whether the upper and lower limits of each recommended parameter range are within the hard constraint boundary; if they are, truncate to the corresponding endpoint of the hard constraint boundary. By taking the intersection method, the physical mechanism constraint boundary and the safety constraint boundary are merged into a hard constraint boundary. Then, the empirically recommended parameter range is truncated and corrected parameter by parameter using this hard constraint boundary, so that the final output recommended parameter range strictly meets the physical feasibility and safety assurance while retaining the rationality of the experience.

[0107] In some embodiments, the method further includes: acquiring drilling feedback data for each drilling cycle; calculating the drilling efficiency for each drilling cycle based on the drilling feedback data; comparing the drilling efficiency with the historical drilling efficiency of each well in the historical cross-well drilling experience data; if the comparison result is greater than a preset historical level threshold, associating and storing the drilling control parameter sequence and corresponding drilling efficiency for that drilling cycle to generate candidate drilling experience data; sending the candidate drilling experience data to an expert for verification; and, in response to the confirmation signal returned by the expert, writing the candidate drilling experience data into the historical cross-well drilling experience data and labeling it with well identifiers, formation characteristics, and confidence levels.

[0108] Drilling feedback data characterizes the actual drilling results achieved after a drilling cycle. This data includes the total footage, total pure drilling time, final bit wear at trip, and average mechanical rate of penetration (MRP) for each individual cycle within that cycle. Drilling efficiency characterizes the overall techno-economic performance of a drilling cycle. It is calculated by dividing the total footage by a weighted average of the total pure drilling time and bit wear. Higher overall efficiency indicates that the cycle achieved more footage and consumed less bit in the same amount of time. Historical cross-well drilling experience data stores cross-well experience entries from multiple historical wells, each associated with its own historical drilling efficiency value. A preset historical level threshold characterizes the threshold value at which the current drilling efficiency is considered significantly higher than the conventional level. This threshold is calculated by adding a standard deviation to the mean of the historical drilling efficiency of each well in the historical cross-well drilling experience data. Candidate drilling experience data is used to characterize the experience entries that, after efficiency screening, are eligible for inclusion in the cross-well experience database. This candidate drilling experience data includes the sequence of drilling pressure control parameters, rotation speed control parameters, and displacement control parameters for the current drilling cycle, along with the corresponding drilling efficiency values. The well identifier is used to uniquely identify the source well of the experience data. Formation features are used to characterize the geological condition attribute vector to which the experience data applies. The confidence score is used to quantitatively assess the reliability of the experience data. Through a dual verification mechanism of efficiency screening and expert verification, the data entries included in the cross-well experience database possess both statistical significance and professional reliability, preventing low-quality or anomalous data from contaminating the database and ensuring the accuracy of cross-well knowledge transfer.

[0109] In some embodiments, the method further includes: acquiring operating condition feedback data for each single drilling cycle; extracting operating condition features for that single drilling cycle based on the operating condition feedback data; associating and storing the operating condition features of that single drilling cycle with the operating condition features of each previously stored single drilling cycle to obtain operating condition cognitive data for the current drilling cycle; and updating historical operating condition experience data for the same well drilling cycles based on the operating condition cognitive data for the current drilling cycle, and marking the corresponding drilling cycle identifier and formation characteristics.

[0110] Operating condition feedback data is used to characterize a summary of drilling results after a single cycle is completed. This data includes the actual footage, average rate of penetration, average torque, average bottom hole pressure, bit wear increment, and operating condition type labels for that single cycle. Operating condition features are used to characterize a summary of the drilling process's operational mode within a single cycle. These features are extracted from the operating condition feedback data and include average formation drillability, average rate of penetration, bit wear increment, and operating condition type labels. Operating condition cognitive data is used to characterize a sequence of operating condition features from all completed single cycles within the current drilling cycle, arranged chronologically. This data accumulates and updates continuously as the drilling cycle progresses, reflecting the overall operating condition evolution trajectory of the current cycle. Historical operating condition experience data from the same well's drilling cycles stores the accumulated operating condition cognitive data for the same target well across historical drilling cycles. Each drilling cycle's operating condition experience data is associated with a corresponding drilling cycle identifier and formation characteristics. By writing the operational condition data into the historical operational condition experience database of the same well after the drilling cycle is completed, the strategic planning layer of subsequent drilling cycles can retrieve the operational condition evolution patterns of the well's historical drilling, realizing the reuse of experience and trend analysis between different drilling cycles of the same well.

[0111] In some embodiments, the method further includes: acquiring drilling feedback data for each decision cycle; constructing a drilling data sample for each decision cycle based on the drilling status parameters, drilling control parameters, and drilling feedback data; associating the drilling data sample with the drilling data samples of each decision cycle already stored within the current single cycle to form drilling experience data for the current single cycle; and updating the historical drilling experience data of the same single cycle based on the drilling experience data of the current single cycle, and marking the corresponding single cycle identifier.

[0112] The decision cycle represents the smallest time unit for the execution control layer to adjust parameters and issue commands. The length of this decision cycle is in the seconds or milliseconds range, and each single cycle contains multiple decision cycles. Drilling feedback data represents the actual observed drilling system response after executing drilling control parameters within a decision cycle. This drilling feedback data includes the actual mechanical drilling rate, actual torque, actual bottom hole pressure, and actual vibration intensity values ​​within that decision cycle. Drilling data samples represent a complete state-action-feedback data triplet within a decision cycle. These samples include the drilling state parameters, drilling control parameters, and drilling feedback data for that decision cycle. Drilling experience data for the current single cycle represents a sequence of data formed by arranging the drilling data samples of all decision cycles stored within the current single cycle in chronological order. Historical drilling experience data for the same single cycle stores the drilling experience data of historical single cycles within the same drilling cycle, with each single cycle's drilling experience data associated with a corresponding single cycle identifier. By storing state-action-feedback data in a single root real-time database at the end of each decision cycle and persisting it to the historical single root experience database at the end of the single root cycle, the execution control layer can retrieve historical data that has occurred in the current single root in real time in subsequent decision cycles for anomaly backtracking and parameter fine-tuning, while providing historical reference benchmarks for the execution control of subsequent single root cycles.

[0113] In some embodiments, the drilling state parameter sequence includes drilling state parameters for multiple decision cycles. Based on this, the method further includes: in the multi-objective non-zero-sum game optimization process, determining the second state parameter increment corresponding to the combined actions of each decision cycle based on the first and second state parameters of each decision cycle; and determining the second state parameters for the next decision cycle based on the second state parameters and the second state parameter increment.

[0114] The drilling state parameter sequence is a time-series set characterizing the drilling system operation status of the target well in each decision cycle within the current single cycle. Each element in this drilling state parameter sequence corresponds to a multi-dimensional state parameter vector for a decision cycle, and the elements are arranged in chronological order according to the decision cycle. The drilling state parameters for the decision cycle include first state parameters and second state parameters. The first state parameters characterize the instantaneous operating status within the current decision cycle and include mechanical specific energy, torque fluctuation rate, rock drillability, mechanical drilling rate, bottom hole pressure, and vibration intensity. The second state parameters characterize the historical cumulative effect from the start of this trip to the current decision cycle and include cumulative footage, drill bit wear, cumulative tubing fatigue, and trip duration.

[0115] The second state parameter increment is used to characterize the expected change of the second state parameter from its current value to the next decision cycle after executing the combined actions of the current decision cycle. Methods for determining the second state parameter increment include, but are not limited to: obtaining the mechanical drilling rate and bottom hole pressure values ​​from the first state parameters of the current decision cycle; obtaining the current drill bit wear and current tubing fatigue accumulation from the second state parameters of the current decision cycle; superimposing the drilling pressure adjustment, rotation speed adjustment, and displacement adjustment from the combined actions onto the current drilling pressure, rotation speed, and displacement values, respectively, to obtain the executed drilling pressure, executed rotation speed, and executed displacement values ​​for this decision cycle; and based on these executed drilling pressure, executed rotation speed, and executed displacement values, combined with the mechanical drilling rate and bottom hole pressure values ​​from the first state parameters of the current decision cycle... Using the drilling rate and bottom hole pressure values, calculate the drill bit wear rate and tubing fatigue accumulation rate within the decision cycle. Multiply the drill bit wear rate by the length of the decision cycle to obtain the drill bit wear increment, which is the first component of the second state parameter increment. Multiply the tubing fatigue accumulation rate by the length of the decision cycle to obtain the fatigue accumulation increment, which is the second component of the second state parameter increment. Use the footage increment of this decision cycle as the cumulative footage increment, which is the third component of the second state parameter increment. Use the length of the decision cycle as the increment of the drilling trip time, which is the fourth component of the second state parameter increment. The methods for determining the second state parameters for the next decision cycle include, but are not limited to: adding each component of the second state parameter in the current decision cycle to each component of the corresponding second state parameter increment to obtain the second state parameter for the next decision cycle. Specifically, the cumulative footage in the next decision cycle equals the current cumulative footage plus the cumulative footage increment; the drill bit wear degree in the next decision cycle equals the current drill bit wear degree plus the drill bit wear increment; the cumulative string fatigue in the next decision cycle equals the current cumulative string fatigue plus the cumulative fatigue increment; and the drilling time in the next decision cycle equals the current drilling time plus the length of the decision cycle. This recursive calculation is repeated in each decision cycle, ensuring the second state parameters are continuously updated as the decision cycle progresses. By calculating the second state parameter increment based on combined actions and recursively updating the second state parameters in each decision cycle, each game agent can obtain the updated cumulative state information in the payoff calculation of subsequent decision cycles. This allows them to perceive the long-term cumulative impact of historical decisions on drill bit life, string safety, and footage progress, achieving joint optimization of immediate actions and long-term costs in long-sequence decision-making.

[0116] In some embodiments, the method further includes: constructing a causal knowledge graph for drilling decisions, wherein the causal knowledge graph is a directed graph structure, and the nodes of the directed graph structure include drilling state variable nodes, drilling control parameter nodes, and drilling target variable nodes; the directed edges of the directed graph structure represent the causal influence of parent node variables on child node variables, and each directed edge has a corresponding edge weight, which is used to characterize the intensity and direction of the causal influence. Based on the directed graph structure, a causal influence path is determined from the drilling state variable node through intermediate nodes to the drilling target variable node. The decision parameters used in each decision are associated with the causal influence path to generate the causal explanation path corresponding to the current decision.

[0117] The drilling decision causal knowledge graph is used to represent the causal relationship network between variables during the drilling process. This knowledge graph is stored in the system using a directed graph data structure, supporting causal relationship querying, reasoning, and path tracing. Drilling state variable nodes represent observable state parameter variables during the drilling process, including nodes for mechanical specific energy, torque fluctuation rate, bottom hole pressure, vibration intensity, and drill bit wear. Drilling control parameter nodes represent actively adjustable operational variables during drilling operations, including nodes for drilling pressure, rotational speed, and displacement. Drilling target variable nodes represent optimization target variables for the drilling project, including nodes for mechanical drilling rate, cumulative footage, remaining drill bit life, and overall drilling efficiency. Directed edges represent the causal relationship between two variables, pointing from a causal variable node to an outcome variable node. The existence of a directed edge indicates that a change in the causal variable leads to a corresponding change in the outcome variable. Edge weights quantify the strength and direction of the causal impact. The absolute value of the edge weight represents the strength of the impact, and the sign of the edge weight represents the direction of the impact. A positive weight indicates that an increase in the causal variable also increases the result variable, while a negative weight indicates that an increase in the causal variable decreases the result variable. The causal impact path represents the complete causal transmission chain starting from the initial variable node, proceeding along directed edges, passing through one or more intermediate nodes, and finally reaching the target variable node. This causal impact path consists of a series of connected directed edges. The causal explanation path represents the decision-making reasoning process description generated by combining the adjustment of control parameters involved in the current decision with the corresponding causal impact path. This causal explanation path is output in text form, including the source of the state parameters used to base the decision, the intermediate causal links, and the final engineering objective. By constructing a causal knowledge graph and generating causal explanation paths, the decision-making process becomes transparent and traceable, transforming it from a black box. Field engineers can understand the reasoning logic and expected impact of each decision, increasing their trust in the intelligent decision-making system.

[0118] In some embodiments, the construction of a causal knowledge graph for drilling decisions may further include: acquiring an explicit mechanism rule base containing physical constraint causal relationships extracted from rock mechanics, tubing dynamics, and hydraulics, used to determine positive and negative causal relationships between drilling state variables, drilling control parameters, and drilling target variables; acquiring a data-driven pattern library containing implicit decision-making patterns mined from historical drilling data and expert experience, used to supplement causal relationships between variables not covered in the explicit mechanism rule base; merging the causal relationships in the explicit mechanism rule base and the data-driven pattern library to generate nodes and edges in a directed graph structure, and determining the edge weights of each edge based on the confidence level of each causal relationship.

[0119] The explicit mechanism rule base stores causal relationships determined based on physical mechanisms. These causal relationships originate from stress-strain constitutive relations in rock mechanics, vibration transmission equations in tubing dynamics, and pressure propagation equations in hydraulics. Positive causal relationships characterize the causal relationship where an increase in the causal variable leads to an increase in the outcome variable; negative causal relationships characterize the causal relationship where an increase in the causal variable leads to a decrease in the outcome variable. The data-driven pattern library stores causal relationships mined from historical drilling data and expert experience. These causal relationships are derived from the analysis of statistical correlations between variables in historical drilling data and the extraction of implicit decision rules from expert experience, covering causal relationships between variables not recorded in the explicit mechanism rule base. Confidence levels characterize the reliability of each causal relationship. The confidence levels of each causal relationship in the explicit mechanism rule base are determined based on the accuracy of the physical model, while the confidence levels of each causal relationship in the data-driven pattern library are determined based on a combination of statistical data volume, correlation coefficient significance, and expert scores. The edge weights are determined as follows: for causal relationships derived from the explicit mechanism rule base, the theoretical coefficients of the mechanism model are used as the baseline values ​​for the edge weights; for causal relationships derived from the data-driven pattern base, the regression coefficients or correlation coefficients between variables in historical data are used as the baseline values ​​for the edge weights. The confidence level of each causal relationship is multiplied by the corresponding baseline value to obtain the edge weight of each edge. By integrating causal relationships from both explicit mechanism rules and data-driven patterns to construct a knowledge graph, the knowledge graph simultaneously possesses the reliability of physical mechanisms and the comprehensiveness of data experience. Explicit mechanism rules ensure the physical rationality of causal relationships, while data-driven patterns supplement the engineering experience rules not covered by the mechanism model.

[0120] In some embodiments, the method further includes: acquiring drilling execution data for the current decision-making period; determining the confidence level of the current decision-making period based on the deviation between the drilling execution data and the expected drilling execution data; comparing the confidence level with a confidence level threshold; if the confidence level is less than a preset confidence level threshold, sending a manual intervention prompt or similar message to the field engineer's terminal; if the confidence level score is greater than or equal to the confidence level threshold, using the drilling control parameters of the current decision-making period as a reference for the next decision-making period.

[0121] Drilling execution data characterizes the actual observed drilling system response after executing drilling control parameters within the current decision-making cycle. This data includes actual mechanical drilling speed, actual torque, actual bottom hole pressure, and actual vibration intensity. Expected drilling execution data characterizes the drilling system response predicted by a mechanistic or empirical model before executing drilling control parameters. This data includes expected mechanical drilling speed, expected torque, expected bottom hole pressure, and expected vibration intensity. Deviation characterizes the degree of difference between actual and expected drilling execution data. This deviation is calculated as the weighted sum of squared relative errors of each parameter; a larger deviation indicates a greater deviation from expectations. Confidence level characterizes the credibility of the current decision result. This confidence level is negatively correlated with the deviation; a smaller deviation indicates a higher confidence level, and a larger deviation indicates a lower confidence level. Confidence threshold characterizes the minimum confidence threshold required to classify the current decision as credible; this threshold is a preset constant. Manual intervention prompts are used to represent early warning signals sent to field engineers, containing current decision information and deviation information. These prompts are displayed on the field engineer's terminal screen in text or graphic form, reminding the engineer to manually review or intervene in the current decision. By scoring the trust level of each decision output, field engineers can intervene promptly when the confidence level of the decision is low, preventing unreliable decisions from being executed and improving the safety and reliability of drilling operations. At the same time, using decision parameters with high confidence levels as subsequent reference inputs ensures that the execution control layer maintains control continuity between consecutive decision cycles and avoids parameter abrupt changes.

[0122] In some embodiments, the storage format of the above-mentioned single root instant library is represented as follows: M_t_single = {(s_{t-H+1}, a_{t-H+1}, r_{t-H+1}), ..., (s_t, a_t, r_t)}; Where M_t_single represents the single instant library at time step t; H represents the rolling window length, which is used to limit the number of sampling points stored in the single instant library; s_t represents the drilling state at time step t, which includes mechanical specific energy, torque fluctuation rate, rock drillability, mechanical drilling rate, bottom hole pressure, and vibration intensity; a_t represents the action executed at time step t, which includes the adjustment amount of drilling pressure, rotational speed, and displacement; r_t represents the instant reward at time step t, which is used to characterize the instant effect of the drilling system response after executing action a_t.

[0123] The single root instant library adopts a real-time rolling update mechanism. When each sampling point arrives, the oldest record on the timeline is discarded and the latest record is added to the library. When the execution control layer detects an abnormal parameter response, it backtracks the historical records stored in the single root instant library to identify the cause of the abnormality and determines whether to trigger the collaborative decision layer for re-optimization.

[0124] In some embodiments, the storage format of each single-root condition summary b_k in the above-mentioned drilling condition database is represented as follows: b_k = (D_k_range, _k_r, RO _k, ΔW_k_bit, operating condition type_k, π_k_effective); Where b_k represents the working condition summary of the kth single pipe; D_k_range represents the depth range corresponding to the kth drill pipe, which is limited by the starting well depth and the ending well depth; _k_r represents the average formation drillability during the drilling process of the k-th drill pipe; RO _k represents the average mechanical drilling speed during the drilling of the k-th drill pipe; ΔW_k_bit represents the drill bit wear increment generated during the drilling of the k-th drill pipe; working condition type_k represents the working condition category identified during the drilling of the k-th drill pipe; π_k_effective represents the optimal decision strategy during the drilling of the k-th drill pipe.

[0125] The strategic planning layer analyzes the formation drillability sequence of each single-run condition summary in the drilling condition database, analyzes the changing trend of formation drillability with well depth, and dynamically adjusts the decision target envelope interval of subsequent single runs based on this changing trend.

[0126] In some embodiments, the storage format of each experience entry e_n in the above-mentioned cross-well experience database is represented as follows: e_n = (stratum feature vector_n, working condition type_n, π_n_success, η_n_score, applicable boundary_n); Wherein, e_n represents the nth experience entry in the cross-well experience database; formation feature vector_n represents the geological condition attribute vector of the drilling section of the historical well corresponding to this experience entry; working condition type_n represents the working condition category corresponding to this experience entry; π_n_success represents the decision strategy recorded in this experience entry that has been verified as successful in historical data; η_n_score represents the comprehensive effect score of this experience entry, which is used to quantify the comprehensive performance of this decision strategy in historical wells; and applicable boundary_n represents the applicable formation conditions range of this experience entry, which is used to limit under what formation conditions this experience entry has reference value.

[0127] The cross-well experience database adopts a dual verification and update mechanism: after each drilling run is completed, the system automatically evaluates the overall efficiency of the run and selects decision-making experiences with overall efficiency higher than the historical average as candidate entries. These candidate entries are then officially written into the cross-well experience database after being verified by experts, and are marked with the source well identifier, applicable formation characteristics, and confidence level.

[0128] In some embodiments, the aforementioned state variables include two categories: immediate state variables (first state parameters) and cumulative state variables (second state parameters).

[0129] The instantaneous state variable is represented as: S_instant = {MSE, T_fluct, D_r, ROP_real, P_bh, Vib_level}; Wherein, S_instant represents the set of instantaneous state variables; MSE represents mechanical specific energy, which characterizes the energy consumed by the drill bit to break a unit volume of rock; T_fluct represents torque fluctuation rate, which characterizes the instantaneous fluctuation of the torque signal; D_r represents rock drillability, which characterizes the formation's resistance to drill bit breakage; ROP_real represents real-time mechanical drilling rate, which characterizes the drilling speed at the current moment; P_bh represents bottom hole pressure, which characterizes the real-time pressure value at the bottom of the well; and Vib_level represents vibration intensity, which characterizes the severity of drill string vibration.

[0130] The cumulative state variable is represented as: S_cumulative = {L_total, W_bit, E_fatigue, H_drill}; Wherein, S_cumulative represents the set of cumulative state variables; L_total represents the cumulative footage, which is used to characterize the total drilling depth from the start time to the current time in the current drilling cycle; W_bit represents the normalized drill bit wear degree, which is used to characterize the cumulative wear quantification value of the drill bit from the initial state to the current time, and this value takes a value between 0 and 1; E_fatigue represents the cumulative fatigue loss of the tubing string, which is used to characterize the cumulative fatigue damage value of the tubing string under alternating loads; H_drill represents the time taken for the current drilling cycle, which is used to characterize the cumulative time from the start time to the current time in the current drilling cycle.

[0131] The decision variables are represented as follows: A = {ΔWOB, ΔRPM, ΔQ}; Where A represents the set of decision variables; ΔWOB represents the drilling pressure adjustment amount, a positive value of which indicates an increase in drilling pressure and a negative value indicates a decrease in drilling pressure; ΔRPM represents the rotational speed adjustment amount, a positive value of which indicates an increase in rotational speed and a negative value indicates a decrease in rotational speed; ΔQ represents the displacement adjustment amount, a positive value of which indicates an increase in displacement and a negative value indicates a decrease in displacement.

[0132] The constraint variables include hard constraints and soft constraints; hard constraints include the safety pressure window [P_pore, P_frac], the critical drill bit pressure for string buckling WOB_max, and the maximum allowable torque T_max, where P_pore represents the formation pore pressure, P_frac represents the formation fracture pressure, and the lower limit of the safety pressure window is determined by P_pore, and the upper limit is determined by P_frac; soft constraints include the upper limit of drill bit wear W_bit,limit, the upper limit of string fatigue accumulation E_fatigue,limit, and the pump pressure rating P_pump,rated.

[0133] In some embodiments, the objective functions of the three agents are as follows: The local observations of the rock-breaking efficiency agent include mechanical specific energy, mechanical drilling speed, formation drillability, drill bit wear, and drilling pressure response. The action preferences of this rock-breaking efficiency agent are drilling pressure adjustment and rotation speed adjustment, and its payoff function is expressed as: J_rock = max E[Σ_{t=0}^{T} γ^t · (α_1 · ROP_t - α_2 · ΔW_bit,t)]; Where J_rock represents the reward function of the rock-breaking efficiency agent; E represents the mathematical expectation; T represents the decision time domain length; t represents the time step index; γ represents the discount factor, which is used to control the weight of future rewards relative to current rewards; ROP_t represents the mechanical drilling rate at time step t; ΔW_bit,t represents the drill bit wear increment at time step t; α_1 represents the weight coefficient of the mechanical drilling rate; and α_2 represents the weight coefficient of the drill bit wear increment.

[0134] The local observations of the wellbore flow safety agent include bottom hole pressure, equivalent circulation density, pump pressure, annular return velocity, cuttings bed height, and safety pressure window. The action preferences of this wellbore flow safety agent are displacement adjustment and pressure boundary constraints, and its payoff function is expressed as: J_fluid = max E[Σ_{t=0}^{T} γ^t · (β_1 · (σ_{P_bh,t})^{-1} - β_2 · Risk_wellbore,t) ]; Wherein, J_fluid represents the payoff function of the wellbore flow safety agent; σ_{P_bh,t} represents the bottom hole pressure variance at time step t, which is used to characterize the fluctuation of bottom hole pressure within time step t; Risk_wellbore,t represents the wellbore risk index at time step t, which is used to characterize the combined risk level of well leakage, overflow, and wellbore instability; β_1 represents the weighting coefficient of bottom hole pressure stability; and β_2 represents the weighting coefficient of wellbore risk.

[0135] The local observations of the tubing string dynamics safety agent include torque, friction, axial load, vibration intensity, buckling margin, and fatigue accumulation. The action preferences of this tubing string dynamics safety agent are drilling pressure limit, rotational speed limit, and parameter change rate limit. Its payoff function is expressed as: J_string = max E[Σ_{t=0}^{T} γ^t · (δ_1 · Vib_suppress,t + δ_2 ·L_fatigue,t) ]; Wherein, J_string represents the benefit function of the tubing dynamics safety agent; Vib_suppress,t represents the vibration suppression effect at time step t, which characterizes the degree of reduction in vibration intensity after the current action is performed; L_fatigue,t represents the tubing fatigue life extension effect at time step t, which characterizes the extension time of tubing fatigue life after the current action is performed; δ_1 represents the weighting coefficient of vibration suppression; and δ_2 represents the weighting coefficient of fatigue life extension.

[0136] The three agents share a unified drilling environment state, and their joint actions consist of adjustments to drilling pressure, rotation speed, and displacement. The same control variable may simultaneously increase the payoff of one agent and decrease the payoff of another agent, thus forming a non-zero-sum dynamic game relationship.

[0137] In some embodiments, the above-mentioned joint benefits are expressed as: r_t_joint = r_t_efficiency + λ_1 · r_t_safety + λ_2 · r_t_knowledge; Where r_t_joint represents the joint reward value at time step t; r_t_efficiency represents the efficiency reward at time step t; r_t_safety represents the safety constraint penalty at time step t; r_t_knowledge represents the knowledge consistency reward at time step t; λ_1 represents the weight coefficient of the safety constraint penalty; and λ_2 represents the weight coefficient of the knowledge consistency reward.

[0138] Efficiency rewards are represented as follows: r_t_efficiency = w_1 · ROP_t + w_2 · η_bit,t + w_3 · L_remaining,t; Where ROP_t represents the mechanical drilling rate at time step t; η_bit,t represents the drill bit efficiency index at time step t, which is used to characterize the rock breaking efficiency of the drill bit under the current wear state; L_remaining,t represents the expected remaining life of the drill bit at time step t; w_1, w_2, and w_3 represent the weighting coefficients of mechanical drilling rate, drill bit efficiency, and remaining life of the drill bit, respectively.

[0139] Safety constraint penalties are represented as follows: r_t_safety = - Σ_{c∈C_soft} max(0, g_c(s_t, a_t)); Where C_soft represents the set of soft constraints; g_c(s_t, a_t) represents the violation degree function of the c-th soft constraint when performing action a_t in state s_t; max(0, g_c(s_t, a_t)) represents the value taken as the penalty when the violation degree is positive, and zero as the penalty when the violation degree is negative or zero.

[0140] The knowledge consistency reward is represented as: r_t_knowledge = Sim(a_t, A_t_KB) + λ_explore · 1[r_t_efficiency > r_t_KB_expected]; Where a_t represents the action to be executed at time step t; A_t_KB represents the parameter range recommended by the knowledge base under the current conditions at time step t; Sim(·, ·) represents the continuous similarity function, which is used to calculate the degree of matching between the executed action and the parameter range recommended by the knowledge base; λ_explore represents the exploration reward coefficient; 1[·] is the indicator function, which takes the value of 1 when the condition in parentheses is true, and takes the value of 0 otherwise; r_t_KB_expected represents the expected efficiency value of the knowledge base under the current conditions at time step t; when the actual efficiency of the agent exceeds the expected efficiency of the knowledge base, an additional exploration reward is given to encourage the discovery of better strategies not recorded in the knowledge base.

[0141] In some embodiments, explicitly incorporating the accumulated state variables into the state transition equation includes: s_{cumulative}^{t+1} = s_{cumulative}^{t} + Δ_{cumulative}(s_{instant}^{t}, a^{t}); Wherein, s_{cumulative}^{t} represents the cumulative state variables at time step t, which include cumulative footage, drill bit wear, cumulative string fatigue, and time elapsed during drilling; s_{cumulative}^{t+1} represents the cumulative state variables at time step t+1; s_{instant}^{t} represents the instantaneous state variables at time step t, which include mechanical specific energy, torque fluctuation rate, rock drillability, mechanical drilling rate, bottom hole pressure, and vibration intensity; a^{t} represents the action executed at time step t; Δ_{cumulative}(·) represents the cumulative state increment function, which is used to calculate the change in the cumulative state based on the current instantaneous state and the executed action.

[0142] A dual-threshold abrupt change detection mechanism is adopted, which simultaneously monitors the mechanical specific energy change rate |ΔMSE / Δt|, the torque fluctuation rate jump variable ΔT_fluct, and the abnormal decrease ratio of mechanical drilling rate ROP_t / ROP_{tW}. When any of the monitored indicators exceeds the corresponding preset threshold, abrupt change diagnosis is triggered. Here, ΔMSE represents the mechanical specific energy change, Δt represents the time change, ΔT_fluct represents the torque fluctuation rate jump variable, which is used to characterize the abrupt change amplitude of torque fluctuation rate between adjacent time windows, ROP_t represents the mechanical drilling rate at time step t, ROP_{tW} represents the mechanical drilling rate at time step tW, and W represents the sliding window length. The response is tiered based on the type of mutation: for minor anomalies, only the parameter correction coefficients of the execution control layer are adjusted; for moderate anomalies, the collaborative decision-making layer is triggered to re-optimize the drilling control parameter sequence within the current sliding time window; for severe mutations, the strategic planning layer is simultaneously triggered to recalculate the constraint boundaries and re-retrieve the cross-well experience database.

[0143] In some embodiments, the above-mentioned causal knowledge graph for drilling decisions is represented by a directed graph G=(V, E), where G represents the causal knowledge graph, V represents the set of nodes, and E represents the set of directed edges.

[0144] The node set V = V_state ∪ V_action ∪ V_objective, where V_state represents the set of state variable nodes, V_action represents the set of decision parameter nodes, and V_objective represents the set of project objective nodes.

[0145] The set of state variable nodes is V_state = {MSE, T_fluct, P_bh, Vib_level, W_bit, …}, where MSE represents mechanical specific energy, T_fluct represents torque fluctuation rate, P_bh represents bottom hole pressure, Vib_level represents vibration intensity, and W_bit represents drill bit wear degree.

[0146] The decision parameter node set V_action = {WOB, RPM, Q}, where WOB represents drilling pressure, RPM represents rotational speed, and Q represents displacement.

[0147] The set of engineering target nodes is V_objective = {ROP, L_total, L_remaining, η_overall}, where ROP represents the mechanical drilling rate, L_total represents the cumulative footage, L_remaining represents the remaining life of the drill bit, and η_overall represents the overall drilling efficiency.

[0148] A directed edge (X_i → Y_j) ∈ E represents the causal influence of variable X_i on variable Y_j. The edge weight θ_ij represents the strength and direction of this causal influence. A positive weight indicates that when the causal variable increases, the result variable also increases, and a negative weight indicates that when the causal variable increases, the result variable decreases. The explicit mechanism rule base extracts physical constraint causal relationships from rock mechanics, tubing dynamics, and hydraulics. The data-driven pattern base mines implicit decision-making patterns from historical drilling data and expert experience.

[0149] Figure 2This is a schematic diagram of the overall framework and long-sequence closed-loop of a domain knowledge-guided multi-agent game-theoretic drilling long-sequence decision-making system provided in this embodiment of the invention. The diagram clearly shows the calling relationships between drilling site data, a drilling knowledge base, a three-level memory, a triplet state evaluation system, a strategic planning layer, a collaborative decision-making layer, an execution control layer, a multi-agent dynamic game module, and a causal explanation and human-machine collaboration module. The multi-agent dynamic game module consists of a rock-breaking efficiency agent, a wellbore flow safety agent, and a tubing dynamics safety agent. It outputs joint actions of drilling pressure, rotational speed, and displacement through hard constraint filtering, joint payoff calculation, and ε-Nash equilibrium and Pareto constraints. The execution results are then fed back to the three-level memory and upper-level planning to form a long-sequence closed-loop decision.

[0150] Figure 3 This is a schematic diagram illustrating the construction and retrieval of the drilling long-sequence decision-making knowledge base and three-level memory provided in this embodiment of the invention. The diagram shows that the knowledge base, derived from mechanistic models, historical well data, and expert rules, undergoes regularization, tripletization, and boundary extraction to obtain mechanistic knowledge, engineering experience knowledge, and risk boundary knowledge. This knowledge base serves as a knowledge service, providing recommended parameter ranges, action filtering conditions, and reward weights to the strategic planning layer, collaborative decision-making layer, and execution control layer. The three-level memory is progressively accumulated according to real-time data, single-well instant memory, drilling condition memory, and cross-well experience memory, respectively supporting anomaly backtracking in the execution control layer, sliding window optimization in the collaborative decision-making layer, and cross-well experience migration in the strategic planning layer.

[0151] Figure 4 This is a schematic diagram of the triplet state evaluation system and hierarchical decision-making process provided in this embodiment of the invention. The diagram organizes the immediate state variables, cumulative state variables, decision variables, and constraint variables into a triplet state evaluation system, and further illustrates the specific division of labor among the three decision-making layers: the strategic planning layer outputs parameter envelopes and risk boundaries at the drilling or global scale; the collaborative decision-making layer calls the multi-agent game module to obtain joint strategies at the single root or sliding window scale; and the execution control layer performs real-time tracking and disturbance correction at the second-level scale. The three layers achieve dynamic linkage through downlink transmission of objectives, boundaries, and setpoints, and uplink feedback of execution deviations, constraint violations, and strategy effects.

[0152] Figure 5This is a schematic diagram of the multi-agent dynamic game collaborative decision-making process provided in an embodiment of the present invention. The diagram clearly defines three game agents: a rock-breaking efficiency agent, a wellbore flow safety agent, and a tubing dynamics safety agent. These three agents form candidate actions or action preferences based on a unified global state, a knowledge base recommendation interval, and retrieval results from a three-level memory. Hard constraints are applied to the candidate actions, and the joint payoff, consisting of efficiency, safety, lifespan, and knowledge consistency terms, is calculated. The joint execution action, including drilling pressure, rotational speed, and displacement, is output based on ε-Nash equilibrium and Pareto constraints. If the equilibrium condition is not met, the agent's strategy is iteratively updated according to the payoff gradient.

[0153] Figure 6 This is a schematic diagram of the three-step human-machine collaborative feedback transformation process provided in an embodiment of the present invention. The diagram clearly depicts the online correction logic that officially takes effect after expert natural language feedback undergoes structured parsing, quantitative constraint mapping, decision window verification, and expert confirmation.

[0154] As can be seen from the technical solutions provided in the embodiments of this specification above, the embodiments of this specification can obtain the drilling state parameter sequence of the current single cycle of the target well; determine the constraint boundary of the drilling control parameters within the drilling cycle in which the current single cycle is located based on historical drilling experience data of different wells and the drilling state parameter sequence; each drilling cycle includes multiple single cycles; based on the constraint boundary, and according to the working condition experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, perform multi-objective non-zero-sum game optimization on the drilling control parameters within the current single cycle to obtain the drilling control parameter sequence of the current single cycle of the target well; fine-tune the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on the drilling experience data of multiple historical single cycles of the same well to obtain the drilling optimization instruction for the current decision cycle of the target well; each single cycle includes multiple decision cycles. By determining the constraint boundaries of the drilling trip scale through cross-well experience data, and using the experience of historical drilling conditions in the same well to conduct multi-objective game optimization at the single-section scale, and then fine-tuning the execution scale based on the experience of single-section drilling in the same trip, an optimization decision-making link with multi-time scale progressive steps and cross-scale reuse of experience data is formed. This can achieve multi-objective collaborative optimization of rock breaking efficiency, wellbore flow safety and tubing dynamics safety while ensuring safety boundaries. At the same time, by continuously correcting the current operation through historical experience, the decision rationality and execution accuracy of drilling control parameters are effectively improved, taking into account both drilling operation safety and drilling efficiency.

[0155] Based on the above-described multi-agent collaborative drilling optimization method, this specification also proposes embodiments of a drilling optimization device. For example... Figure 7 As shown, the drilling optimization device 700 may specifically include the following modules: The acquisition module 701 is used to acquire the drilling status parameter sequence of the target well in the current single cycle; The determination module 702 is used to determine the constraint boundaries of drilling control parameters within the current single-cycle drilling cycle based on historical cross-well drilling experience data and drilling status parameter sequences; each drilling cycle includes multiple single-cycle drilling cycles. The optimization module 703 is used to perform multi-objective non-zero-sum game optimization on the drilling control parameters in the current single cycle based on constraint boundaries and working experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamic safety as optimization objectives, to obtain the drilling control parameter sequence of the target well in the current single cycle. The fine-tuning module 704 is used to fine-tune the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on drilling experience data from multiple historical single-cycle drilling trips, so as to obtain the drilling optimization instructions for the current decision cycle of the target well; each single cycle includes multiple decision cycles.

[0156] As can be seen from the technical solutions provided in the embodiments of this specification above, the embodiments of this specification can obtain the drilling state parameter sequence of the current single cycle of the target well; determine the constraint boundary of the drilling control parameters within the drilling cycle in which the current single cycle is located based on historical drilling experience data of different wells and the drilling state parameter sequence; each drilling cycle includes multiple single cycles; based on the constraint boundary, and according to the working condition experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, perform multi-objective non-zero-sum game optimization on the drilling control parameters within the current single cycle to obtain the drilling control parameter sequence of the current single cycle of the target well; fine-tune the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on the drilling experience data of multiple historical single cycles of the same well to obtain the drilling optimization instruction for the current decision cycle of the target well; each single cycle includes multiple decision cycles. By determining the constraint boundaries of the drilling trip scale through cross-well experience data, and using the experience of historical drilling conditions in the same well to conduct multi-objective game optimization at the single-section scale, and then fine-tuning the execution scale based on the experience of single-section drilling in the same trip, an optimization decision-making link with multi-time scale progressive steps and cross-scale reuse of experience data is formed. This can achieve multi-objective collaborative optimization of rock breaking efficiency, wellbore flow safety and tubing dynamics safety while ensuring safety boundaries. At the same time, by continuously correcting the current operation through historical experience, the decision rationality and execution accuracy of drilling control parameters are effectively improved, taking into account both drilling operation safety and drilling efficiency.

[0157] This specification also provides a computer device for a multi-agent collaborative drilling optimization method, including a processor and a memory for storing processor-executable instructions. Specifically, the processor can perform the following tasks according to the instructions: acquiring the drilling state parameter sequence for the current single-cycle of the target well; determining the constraint boundary of the drilling control parameters within the current single-cycle trip based on historical drilling experience data from different wells and the drilling state parameter sequence; each trip includes multiple single-cycles; based on the constraint boundary and according to the working condition experience data from multiple historical trips of the same well, performing multi-objective non-zero-sum game optimization on the drilling control parameters within the current single-cycle, with rock breaking efficiency, wellbore flow safety, and tubing dynamics safety as optimization objectives, to obtain the drilling control parameter sequence for the current single-cycle of the target well; fine-tuning the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on drilling experience data from multiple historical trips of the same single cycle, to obtain the drilling optimization instructions for the current decision cycle of the target well; each single-cycle includes multiple decision cycles.

[0158] To execute the above instructions more accurately, please refer to... Figure 8 As shown in the embodiments of this specification, another specific computer device 800 is also provided, wherein the computer device 800 includes a network communication port 801, a processor 802 and a memory 803, and the above structures are connected by internal cables so that the various structures can perform specific data interaction.

[0159] The processor 802 can be specifically used to: acquire the drilling state parameter sequence of the current single-cycle of the target well; determine the constraint boundary of the drilling control parameters within the current single-cycle drilling period based on historical drilling experience data of different wells and the drilling state parameter sequence; each drilling period includes multiple single-cycles; based on the constraint boundary and according to the working condition experience data of multiple historical drilling periods of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, perform multi-objective non-zero-sum game optimization on the drilling control parameters within the current single-cycle to obtain the drilling control parameter sequence of the current single-cycle of the target well; fine-tune the drilling control parameters of the current decision cycle in the drilling control parameter sequence based on the drilling experience data of multiple historical single cycles of the same well to obtain the drilling optimization instruction for the current decision cycle of the target well; each single-cycle includes multiple decision cycles.

[0160] The memory 803 can be used to store the corresponding instruction program.

[0161] In this embodiment, the network communication port 801 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.

[0162] In this embodiment, the processor 802 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.

[0163] In this embodiment, the memory 803 includes volatile memory and non-volatile memory. The memory 803 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.

[0164] 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 multi-agent collaborative drilling optimization method are shown.

[0165] 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 multi-agent collaborative drilling optimization method is shown.

[0166] 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.

[0167] 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.

[0168] 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.

[0169] 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. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0170] 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.

[0171] 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.

[0172] 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 multi-agent collaborative drilling optimization method, characterized in that, include: Obtain the drilling status parameter sequence for the current single cycle of the target well; Based on historical cross-well drilling experience data and drilling status parameter sequences, determine the constraint boundaries of drilling control parameters within the current single-well cycle; each drilling cycle includes multiple single-well cycles. Based on the constraint boundary and the working experience data of multiple historical drilling cycles of the same well, with rock breaking efficiency, wellbore flow safety and tubing dynamics safety as optimization objectives, the drilling control parameters in the current single cycle are optimized by multi-objective non-zero-sum game to obtain the drilling control parameter sequence of the target well in the current single cycle. Based on drilling experience data from multiple historical single-cycle drilling operations, the drilling control parameters for the current decision cycle in the drilling control parameter sequence are fine-tuned to obtain the drilling optimization instructions for the target well in the current decision cycle; each single-cycle includes multiple decision cycles.

2. The method according to claim 1, characterized in that, The step of determining the constraint boundaries of drilling control parameters within the current single-well cycle based on historical cross-well drilling experience data and drilling status parameter sequences includes: Based on the drilling state parameter sequence, generate the first formation characteristics of the drilling section in the current single cycle; Calculate the similarity between the first formation features and the second formation features associated with historical drilling experience data from different wells; Historical drilling experience data of different wells that meet the preset similarity conditions are selected as matching experience data; Based on the historical drilling control parameters and corresponding drilling efficiency in the matching experience data, the constraint boundary of the drilling control parameters within the current single cycle is generated.

3. The method according to claim 1, characterized in that, Based on constraint boundaries and using operational experience data from multiple historical drilling cycles of the same well, the drilling control parameters for the current single cycle are optimized using a multi-objective non-zero-sum game approach, with rock-breaking efficiency, wellbore flow safety, and tubing dynamics safety as optimization objectives. This yields the drilling control parameter sequence for the current single cycle of the target well, including: First, second, and third game agents were constructed to correspond to rock breaking efficiency, wellbore flow safety, and tubing dynamics safety, respectively. The drilling state parameter sequence and operating condition experience data are input into each game agent to obtain the first candidate action, the second candidate action, and the third candidate action; each candidate action includes the adjustment amount of the drilling control parameters. Based on the constraint boundaries, each candidate action is modified; Calculate the rock-breaking efficiency benefit, wellbore flow safety benefit, and tubing dynamics safety benefit for the corrected first, second, and third candidate actions, respectively. The revised first candidate action, second candidate action, and third candidate action are combined into a candidate combination action; Calculate the consistency payoff corresponding to the candidate combination of actions; The combined benefits are determined based on the benefits of rock breaking efficiency, wellbore flow safety, tubing dynamics safety, and consistency. Based on the combined revenue, determine the drilling control parameter sequence for the current single cycle of the target well.

4. The method according to claim 3, characterized in that, The drilling control parameters include at least pressure on the drill bit, rotational speed, and displacement; each candidate action includes one or more of the following: pressure on the drill bit adjustment, rotational speed adjustment, and displacement adjustment. The calculation of the rock-breaking efficiency gain, wellbore flow safety gain, and tubing dynamics safety gain corresponding to the corrected first, second, and third candidate actions includes: Based on the drilling pressure adjustment and rotation speed adjustment in the first candidate action, determine the expected mechanical drilling speed and expected drill bit wear increment corresponding to the first candidate action; Calculate the drilling speed benefit and wear cost based on the expected mechanical drilling rate and the drill bit wear increment, respectively. The rock-breaking efficiency benefit is determined based on the drilling speed benefit and the wear cost; the rock-breaking efficiency benefit is positively correlated with the drilling speed benefit and negatively correlated with the wear cost.

5. The method according to claim 3, characterized in that, The drilling control parameters include at least pressure on the drill bit, rotational speed, and displacement; each candidate action includes one or more of the following: pressure on the drill bit adjustment, rotational speed adjustment, and displacement adjustment. The calculation of the rock-breaking efficiency gain, wellbore flow safety gain, and tubing dynamics safety gain corresponding to the corrected first candidate action, second candidate action, and third candidate action also includes: Based on the displacement adjustment in the second candidate action, determine the expected bottom hole pressure and expected annular return velocity corresponding to the second candidate action; Calculate the pressure deviation between the expected bottom-hole pressure and the preset safe pressure window boundary; Based on the pressure deviation and the expected annular return velocity, the bottom hole pressure safety benefit and the cutterhead carrying capacity benefit are calculated respectively; the bottom hole pressure safety benefit is negatively correlated with the pressure deviation. The wellbore flow safety benefits are determined based on the bottom hole pressure safety benefits and the rock-carrying capacity benefits.

6. The method according to claim 3, characterized in that, The drilling control parameters include at least pressure on the drill bit, rotational speed, and displacement; each candidate action includes one or more of the following: pressure on the drill bit adjustment, rotational speed adjustment, and displacement adjustment. The calculation of the rock-breaking efficiency gain, wellbore flow safety gain, and tubing dynamics safety gain corresponding to the corrected first candidate action, second candidate action, and third candidate action also includes: Based on the drilling pressure adjustment and rotation speed adjustment in the third candidate action, determine the corresponding expected torque fluctuation, expected vibration intensity, expected buckling margin, and expected fatigue cumulative increment. Based on the expected torque fluctuation, expected vibration intensity, expected buckling margin, and expected cumulative fatigue increment, the torque stability benefit, vibration suppression benefit, buckling safety benefit, and fatigue life benefit are calculated respectively. The torque stability benefit is negatively correlated with the expected torque fluctuation, the vibration suppression benefit is negatively correlated with the expected vibration intensity, the buckling safety benefit is positively correlated with the expected buckling margin, and the fatigue life benefit is negatively correlated with the expected cumulative fatigue increment. The dynamic safety benefits of the tubular column are determined based on the benefits of torque stability, vibration suppression, buckling safety, and fatigue life.

7. The method according to claim 3, characterized in that, The calculation of the consistency benefit corresponding to the candidate combination of actions includes: Determine the recommended parameter range corresponding to the current single period; Calculate the degree of deviation between each drilling control parameter in the candidate combination action and the corresponding parameter in the recommended parameter range; Based on the degree of deviation of each drilling control parameter, calculate the matching degree between the candidate combination of actions and the recommended parameter range; Consistency gains are determined based on the matching degree; consistency gains are positively correlated with matching degree.

8. The method according to claim 3, characterized in that, The step of determining the drilling control parameter sequence for the current single cycle of the target well based on the combined revenue includes: Based on joint payoffs, determine whether the combined actions satisfy the Nash equilibrium condition. The Nash equilibrium condition includes a first condition and a second condition. The first condition is that the payoff increase corresponding to each player agent changing its own candidate action does not exceed the payoff threshold. The second condition is that the candidate combined action is not Pareto-dominated by other candidate combined actions. If satisfied, the combined actions will be used as the drilling control parameter sequence for the current single cycle of the target well; If the conditions are not met, the strategies of each game agent are iteratively updated until the generated combination of actions satisfies the Nash equilibrium condition.

9. The method according to claim 1, characterized in that, The process involves fine-tuning the drilling control parameters for the current decision period within the drilling control parameter sequence based on drilling experience data from multiple historical single-cycle drilling runs, to obtain the drilling optimization instructions for the target well in the current decision period, including: The drilling status parameters of the target well in the current decision-making cycle are compared with the corresponding drilling status parameters in the drilling experience data to obtain the execution deviation of the drilling control parameters. Based on the execution deviation, the drilling control parameters in the current decision cycle of the drilling control parameter sequence are corrected; Based on the revised drilling control parameters, generate drilling optimization instructions for the target well in the current decision-making cycle.

10. The method according to claim 1, characterized in that, The method further includes: Obtain the rate of change of drilling status parameters in the current decision-making cycle; If the rate of change of drilling condition parameters is greater than the rate of change threshold, determine the level of sudden change in operating conditions for the current decision-making cycle; If the sudden change in operating conditions is a Level 1 sudden change, the drilling control parameters in the current decision cycle of the drilling control parameter sequence should be fine-tuned. If the change in operating conditions is a level 2 change, the drilling control parameters in the current single cycle will be re-optimized using a multi-objective non-zero-sum game. If the change in operating conditions is a level three change, the constraint boundaries of the drilling control parameters within the current single-root cycle should be redefined.

Images