Intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm

Abstract

The invention relates to the technical field of well completion data processing, in particular to an intelligent well completion multi-objective optimization method based on a multi-objective genetic algorithm. The method comprises the following steps: acquiring a complete well completion parameter set and a target function set of intelligent well completion; dimension reduction is conducted on the complete well completion parameter set, and a key well completion parameter set is obtained; constructing a dynamic penalty factor generation rule; constructing a multi-level proxy modeling system based on the key well completion parameter set and the target function set to obtain a hierarchical proxy model; constructing a target function evaluation system of multi-target optimization to obtain a fitness evaluation standard; and performing multi-population parallel genetic evolution based on the hierarchical agent model and the fitness evaluation standard to obtain a multi-objective optimization candidate solution set. According to the intelligent well completion parameter optimal configuration method, the efficient and stable intelligent well completion parameter optimal configuration method with the global optimization capability is realized by fusing dimension reduction, multi-target collaborative evaluation, parallel evolution and a dynamic constraint mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of well completion data processing, and particularly relates to an intelligent well completion multi-objective optimization method based on a multi-objective genetic algorithm. BACKGROUND

[0002] In the field of intelligent oil and gas well completion design, the selection and configuration of well completion parameters gradually develop towards multi-objective coordination and intelligent optimization. Traditional well completion schemes are mostly based on engineering experience and single-objective evaluation for parameter configuration, which is difficult to take into account multiple performance indicators such as productivity, cost, risk and system stability, and the parameter selection is limited by artificial subjective judgment, which easily leads to insufficient scheme optimization or unbalanced resource allocation. The existing intelligent well completion multi-objective optimization method has the following key technical bottlenecks:

[0003] Firstly, the dimension reduction and hierarchical modeling mechanism are not effectively integrated, resulting in bloated model structure or distorted response;

[0004] Secondly, the conventional genetic algorithm performs poorly in population diversity maintenance and constraint processing, lacks population coordination and adaptive evolution strategy, and is easily trapped in local optimum;

[0005] Thirdly, the traditional constraint mechanism mostly relies on static penalty functions, which cannot dynamically respond to the constraint violation degree in the optimization process, and it is difficult to guarantee the feasibility and stability of the optimization results. SUMMARY

[0006] Therefore, it is necessary to provide an intelligent well completion multi-objective optimization method based on a multi-objective genetic algorithm to solve at least one of the above technical problems.

[0007] To achieve the above-mentioned purpose, an intelligent well completion multi-objective optimization method based on a multi-objective genetic algorithm comprises the following steps:

[0008] Step S1: obtaining a well completion parameter set and a target function set of intelligent well completion; performing dimension reduction on the well completion parameter set to obtain a key well completion parameter set; and constructing a dynamic penalty factor generation rule;

[0009] Step S2: constructing a multi-level agent modeling system based on the key well completion parameter set and the target function set to obtain a hierarchical agent model; and constructing a target function evaluation system for multi-objective optimization to obtain an adaptability evaluation standard;

[0010] Step S3: performing multi-population parallel genetic evolution based on the hierarchical agent model and the adaptability evaluation standard to obtain a multi-objective optimization candidate solution set;

[0011] Step S4: performing constraint processing on the multi-objective optimization candidate solution set by using the dynamic penalty factor generation rule to obtain a feasible solution set;

[0012] Step S5: Pareto boundary analysis and multi-objective sorting are performed on the feasible solution set to obtain an optimal intelligent completion parameter configuration scheme.

[0013] The application significantly reduces the redundancy and computational complexity of the completion parameter space by introducing a strategy of combining dimensionality reduction and hierarchical modeling, effectively extracts key parameters while maintaining the integrity of physical meaning, and effectively alleviates the problems of model overfitting and unstable response caused by high-dimensional data; by constructing a hierarchical proxy modeling system, the nonlinear relationships between different levels of targets and parameters in the complex completion system are structured, the accuracy of target response and the modeling efficiency are improved; using a multi-objective collaborative evaluation standard, the balanced consideration of multiple targets such as production capacity, cost, risk control, etc. in the optimization process is ensured, and the systematicness and rationality of the optimization and selection are enhanced; combined with the multi-population parallel evolution mechanism and the adaptive solution interaction strategy, the global search ability and population diversity control ability of the population are significantly enhanced, effectively overcoming the local convergence problem caused by a single evolution path; the dynamic penalty factor mechanism adjusts the constraint processing strength in real time according to the deviation change in different stages, realizes the dynamic identification and flexible intervention of constraint violation, and improves the feasibility judgment and stability control ability of the solution; finally, through multi-objective non-dominated analysis and multi-level sorting strategy, the feasible solution set is optimized, the trade-off between targets is realized, and the optimal configuration scheme is quickly extracted, which significantly improves the scientificity, collaboration and practical application reliability of intelligent completion parameter design as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0014] Other features, objects, and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the drawings:

[0015] Figure 1 Figure 1 is a schematic diagram of the step flow of the intelligent completion multi-objective optimization method based on the multi-objective genetic algorithm of the application;

[0016] Figure 2 Figure 2 is a detailed step flow diagram of step S1 in the method; Figure 1

[0017] Figure 3 Figure 3 is a multi-population parallel genetic evolution process diagram of an embodiment of the application. DETAILED DESCRIPTION

[0018] The technical method of the application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0019] ​Further, the accompanying drawings are included to provide a further understanding of the present application, and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments of the present application and, together with the description, serve to explain the principles of the present application. In the drawings:

[0020] It should be understood that, although terms such as "first", "second", and so on can be used herein to describe various elements, the elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be referred to as a second element, and similarly a second element can be referred to as a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] To achieve the above object, there is provided Figures 1 to 3 The present application provides an intelligent completion multi-objective optimization method based on a multi-objective genetic algorithm, which comprises the following steps:

[0022] Step S1: obtaining a full set of completion parameters and a set of objective functions of intelligent completion; performing dimension reduction on the full set of completion parameters to obtain a set of key completion parameters; and constructing a dynamic penalty factor generation rule;

[0023] Step S2: constructing a multi-level agent modeling system based on the set of key completion parameters and the set of objective functions to obtain a hierarchical agent model; and constructing an objective function evaluation system for multi-objective optimization to obtain a fitness evaluation standard;

[0024] Step S3: performing multi-population parallel genetic evolution based on the hierarchical agent model and the fitness evaluation standard to obtain a multi-objective optimization candidate solution set;

[0025] Step S4: performing constraint processing on the multi-objective optimization candidate solution set by using the dynamic penalty factor generation rule to obtain a feasible solution set;

[0026] Step S5: performing Pareto boundary analysis and multi-objective sorting on the feasible solution set to obtain an optimal intelligent completion parameter configuration scheme.

[0027] Preferably, step S1 comprises the following steps:

[0028] Step S11: collecting multi-source intelligent completion related data to obtain a full set of completion parameters and a set of objective functions;

[0029] Step S12: Correlation analysis is performed on the complete set of completion parameters based on the objective function set, to obtain an initial characteristic importance ranking;

[0030] Step S13: Principal component analysis is performed on the initial characteristic importance ranking, to obtain a set of key principal components;

[0031] Step S14: Feature space mapping transformation is performed on the complete set of completion parameters based on the set of key principal components, to obtain a reduced dimension parameter representation;

[0032] Step S15: Hierarchical structure clustering division is performed on the reduced dimension parameter representation, to obtain a parameter hierarchical grouping structure;

[0033] Step S16: A high weight subset is selected based on the parameter hierarchical grouping structure and the objective function set, to constitute a set of key completion parameters;

[0034] Step S17: An initial constraint function family is constructed based on an adaptability evaluation standard, to obtain a constraint function definition set;

[0035] Step S18: Function deformation and weight adjustment are performed on the constraint function definition set, to obtain a dynamic penalty factor generation rule.

[0036] In the embodiment of the present application, through the deployment of multi-type data acquisition devices on the surface of the oil and gas well site, including wellbore pressure sensor, casing temperature sensor, flowmeter, production tester and formation physical property analysis equipment, at least 20 original well completion related parameter data including wellbore pressure, casing temperature, formation porosity, permeability, completion fluid type, injection and discharge capacity, packer position, perforation density, etc. are collected, and based on the operation result library and economic evaluation model of historical intelligent well completion projects, a target function set is constructed, which includes 4 types of indexes such as operation cost, productivity growth rate, reservoir damage control rate and safety stability. The obtained well completion parameter set and target function set are calculated by a double variable matrix calculation method based on Pearson correlation coefficient, the linear correlation coefficient between each well completion parameter and each target index is calculated in turn, and the absolute value is weighted and averaged to obtain the comprehensive correlation degree score of the parameter, which is arranged from high to low according to the score, and an initial feature importance ranking is formed. The top 10 well completion parameters in the ranking are selected as the principal component input by using the standard of feature value greater than 1, a covariance matrix is constructed, eigenvalue decomposition is performed, the first three principal component combinations with cumulative contribution rate not less than 85% are obtained, and a key principal component set is formed. The key principal component set is mapped to all parameters in the original well completion parameter set through linear combination, and a 3*20-dimensional mapping weight matrix is constructed to perform linear transformation to obtain a 3-dimensional reduced parameter representation. The reduced parameter representation is calculated by a clustering center iteration method based on Euclidean distance, the initial clustering number is set to 3, the maximum iteration number is set to 100, the intra-class variance is calculated after each iteration, and the convergence condition is judged, finally the reduced data is divided into 3 stable hierarchical structure groups, and the parameter hierarchical grouping structure is obtained. According to the average correlation degree score ranking of the well completion parameters in each group to the target function set, the top 2 well completion parameters in each group are selected to form a key well completion parameter set containing 6 parameters. Referring to the score limit and evaluation method of the target function in the fitness evaluation standard, combined with the engineering limit conditions summarized from the failure cases in the historical well completion operation, 6 initial constraint functions are constructed, including the maximum well completion fluid pressure not exceeding 40 MPa, the casing temperature not higher than 180℃, the minimum spacing between packer and oil layer upper boundary not less than 1 meter, etc., forming a constraint function definition set. The strength of each constraint function is adjusted by applying an exponential decay function, the corresponding penalty expression of each constraint is constructed, and according to the rule that the penalty strength increases linearly with the number of iterations, the function penalty weight is set to an initial value of 0.1 and a maximum dynamic range of 0.8, and the dynamic penalty factor generation rule is formed by gradually increasing the strength.

[0037] The application ensures the comprehensiveness of the well completion parameter information and the engineering adaptability of the target function system by introducing multi-source data acquisition and quantitative processing means, provides high-quality input basis for subsequent optimization; through correlation analysis between parameters and target functions, the system identifies key factors that significantly affect multi-target performance response, reducing the dimension burden of the feature space; the principal component extraction and mapping transformation method is adopted to improve the information expression efficiency and correlation compression ability between original parameters, significantly reducing the consumption of computing resources and enhancing the physical consistency of data expression; combined with hierarchical structure clustering division, the reduced parameters are grouped according to the structural characteristics, so that the structural coupling relationship between the parameters is effectively preserved in the optimization process, and the hierarchical clarity and interpretability of the subsequent modeling process are enhanced; through the key subset screening strategy, the target-oriented parameter extraction ability is further strengthened, and the parameter sensitivity difference between multi-targets is accurately mapped; at the same time, a constraint function definition system is constructed, so that the optimization process has comprehensive constraint response ability, and combined with the function form reconstruction and dynamic weight adjustment mechanism, a punishment control system that can automatically adjust with the evolution process is constructed, which significantly improves the discrimination ability and dynamic adaptability to non-feasible solutions, and provides rigorous constraint mechanism and fine parameter support for high-quality optimization in the whole process.

[0038] Preferably, the step S2 of constructing a multi-level agent modeling system based on the key well completion parameter set and the target function set comprises:

[0039] performing feature hierarchical classification on the key well completion parameter set to obtain well completion hierarchical structure data;

[0040] constructing each layer sub-model based on the well completion hierarchical structure data and the target function set to obtain an agent model substructure set;

[0041] performing model fusion on the agent model substructure set to obtain a unified hierarchical agent model.

[0042] In the embodiment of the present application, the key completion parameter set is input into the feature hierarchical classification module, and the parameter is layered by adopting the division mode based on the attribute scale and the action area double factors, wherein the attribute scale dimension includes three types of "continuous type", "discrete type" and "category type", the action area dimension is divided into three types of "wellbore section", "formation section" and "surface section", each parameter is marked in the two-dimensional coordinate, and then is grouped according to the clustering boundary condition to obtain the completion layering structure data, wherein the wellbore section includes parameters such as casing inner diameter and packer position, the formation section includes formation porosity and permeability, and the surface section includes injection displacement and completion fluid density, and the number of parameters in each section is limited to 2-4. Based on the above layering structure data and the target function set, a series of equidistant parameter combination samples are constructed in the value space of each parameter subset by adopting the finite difference approximation mode, and the corresponding target function response values are extracted by referring to the field experiment report, the operation curve and the engineering experience calculation formula to form the sample input and output pairs, and the input and output mapping structure of each layer is established respectively to form the proxy model substructure set. After obtaining a plurality of substructures, the substructures are fused. First, the outputs of the substructures are uniformly normalized by using the minimum-maximum linear normalization method to map the response values to the interval [0, 1]. Then, the normalized outputs of the substructures are summarized by constructing a three-dimensional response matrix (dimension: sample number x submodel number x target function number). Then, the cross-layer mapping operation is performed with the target function number as the index dimension, and the weighted average of the response values of each target function in different substructures is taken, and the weight is inversely proportional to the average absolute error of the corresponding substructure on the preliminary verification set. Finally, the weighted fusion results of all target functions are arranged and combined according to the sample serial number to form a unified response vector set, and the input and output pair relationship between the key completion parameter set and the unified response vector is established, thereby obtaining a unified hierarchical proxy model.

[0043] The present application can structure the completion parameters with significant differences in action area and attribute characteristics by introducing a feature hierarchical classification mechanism, so that the modeling process is more physically logical and interpretable, and the correlation expression ability between parameters is effectively improved. The respective establishment of each layer submodel makes the mapping relationship between local features and target functions more refined, avoids the decline of generalization ability caused by parameter interference in global modeling, and improves the response fitting precision and boundary behavior prediction ability. After the normalization and fusion of the results of multiple substructures, the response advantages provided by different levels of information are fully integrated, the robustness and response consistency of the overall fitting of the target function are enhanced, and the influence of abnormal input on the stability of the output is reduced. The final unified hierarchical proxy structure has hierarchical adaptability and overall convergence, can provide more accurate and balanced evaluation reference for different schemes in the genetic evolution stage, and significantly improves the screening efficiency and convergence speed of global solution in multi-objective optimization.

[0044] Preferably, the model fusion on the agent model substructure set comprises:

[0045] Performing output consistency analysis on the agent model substructure set to obtain model response normalized data;

[0046] Constructing a cross-layer fusion mapping structure based on the model response normalized data to obtain a structure mapping matrix;

[0047] Performing inter-layer information aggregation on the structure mapping matrix to obtain an aggregated output representation;

[0048] Performing joint regression modeling based on the aggregated output representation and a target function set to obtain a unified hierarchical agent model.

[0049] In the embodiment of the application, based on multiple well completion layer substructure agent models, first, the output prediction results of each substructure model on a unified test sample set (the number of samples is not less than 100, the sample dimension is equal to the number of key well completion parameters, and the number of key well completion parameters is not more than 20) are read in sequence to form an original response result matrix; the matrix is processed by standard deviation normalization by column, all model output results are normalized to the closed interval [0, 1] using the maximum and minimum value interval linear stretching method, and model response normalized data is obtained; according to the well completion parameter level information corresponding to each substructure model, a cross-layer fusion mapping structure is established, specifically: the model output results under the same level are linearly combined by row, and the model results across levels are fused by layering weighting, the weight is set based on the normalized weight value of the corresponding level target in the target function set (the normalized weight precision is set to three decimal places, and the total weight is equal to 1), thereby constructing a structure mapping matrix; the matrix row and column sum operation is performed on the structure mapping matrix in the order of level, the inter-layer information aggregation is performed, the aggregated output representation of each level is extracted, and a fusion feature set is formed; the fusion feature set is used as input, the multi-objective response value in the target function set is used as output, and a multiple linear regression method based on the least square error criterion is used to establish a joint regression expression, wherein the input-output correspondence is clear, each target function corresponds to a group of regression coefficient, and the regression residual error is controlled within 5%; and finally a unified hierarchical agent model is generated.

[0050] The application can effectively eliminate the response deviation caused by the non-uniform target response scale between each substructure, improve the comparability of each layer output under the unified evaluation system, realize the effective transmission and integration of information between different layers through the construction of the cross-layer fusion mapping structure, and improve the collaborative expression ability of each substructure in the target function prediction; the interlayer aggregation operation of the structure mapping matrix further strengthens the complementarity of the local response characteristics between different levels, so that the overall output has stronger global consistency while maintaining the difference in details; and finally, the aggregated output representation is associated with the target function set through the joint regression mode, so that the unified hierarchical agent model has the comprehensive mapping ability of the influence of multiple sources and multiple scale parameters, and significantly improves the stability, response continuity and prediction accuracy of the multi-objective evaluation under complex constraint conditions, and provides more reliable performance support for the subsequent optimal solution screening and evolution.

[0051] Preferably, the target function evaluation system of the multi-objective optimization constructed in step S2 comprises:

[0052] constructing a target function hierarchical structure based on the target function set to obtain target function weight mapping data;

[0053] normalizing the target function weight mapping data to obtain a standardized target weight set;

[0054] constructing a fitness evaluation function family for multi-objective optimization based on the standardized target weight set and the target function set to obtain a fitness evaluation standard.

[0055] In the embodiment of the present application, based on the target function set, the engineering significance of each target function in the intelligent completion operation is functionally decomposed, grouped and classified according to three dimensions of "production efficiency type", "operation cost type" and "risk control type", a three-layer nested target function hierarchical structure is constructed, and a multi-target hierarchical index table is formed; wherein the number of target functions under each dimension shall not exceed 5, and the target function naming remains consistent with the original data; the initial weight value of each target function is allocated according to the hierarchical structure, the initial allocation method is based on the engineering experience scoring table, the scoring uses a 1-9 integer table, the scoring results within the layer are compared by pair through AHP (analytic hierarchy process) to construct a comparison matrix and perform maximum eigenvalue consistency test, and target function weight mapping data is generated; the weight mapping data is normalized, the normalization method is to divide each function weight by the total sum of all function weights, to ensure that the sum of all target function weight values is 1, and the decimal precision is uniformly kept to three digits; after completing the construction of the standardized target weight set, based on the standardized weight and the target function set, an evaluation sub-function is defined for each target function in turn, the evaluation sub-function adopts a weighted linear scoring method, and the specific form is that the target response value is multiplied by the corresponding standardized weight coefficient to obtain the individual score value of each target function; the weighted sum of all score values is obtained to obtain the total evaluation score of a single completion parameter combination, and a fitness evaluation function set for multi-objective optimization is formed; the evaluation function set is used as a unified measurement standard for measuring the pros and cons of each solution in population evolution, and a fitness evaluation standard is formed.

[0056] By constructing the hierarchical structure of the target function, the present application can clearly determine the dependency relationship and importance between each optimization target, avoid the interference of multi-objective conflict on the optimization result, and enhance the organizational logic of the overall evaluation; the weight normalization processing ensures that each target is evaluated under a unified numerical scale, effectively prevents evaluation bias caused by the dimension difference of the original index, and improves the fairness of each target in the fitness calculation; the establishment of the fitness evaluation standard enables each optimization scheme to have a quantifiable evaluation basis in the iteration process, not only enhances the resolution of the pros and cons in the population evolution process, but also guarantees the dynamic response ability of the convergence path to the trade-off relationship of different targets, thereby improving the overall efficiency, stability and decision rationality of multi-objective optimization.

[0057] Preferably, step S3 comprises the following steps:

[0058] Step S31: constructing an initial multi-population genetic coding system based on the hierarchical agent model and the fitness evaluation standard, to obtain a population initialization parameter set;

[0059] Step S32: performing selection, crossover and mutation operations on each population in the population initialization parameter set in parallel, and combining the fitness evaluation standard to judge the pros and cons, to obtain the fitness data of the first generation of multi-objective population;

[0060] Step S33: adjusting the fitness evaluation standard based on the first generation multi-objective population fitness data to obtain an updated fitness function;

[0061] Step S34: performing an interactive transfer operation of solutions on each population in parallel evolution according to a preset migration frequency according to the updated fitness function to obtain population migration update data;

[0062] Step S35: performing a plurality of rounds of parallel evolution iterations based on the updated fitness penalty function and the population migration update data to obtain a multi-objective optimization candidate solution set.

[0063] In the embodiment of the application, based on the hierarchical agent model and the fitness evaluation standard, each parameter in the key completion parameter set is encoded using an integer coding method, the coding precision is set to three decimal places, and the coding range is strictly limited between the upper and lower limits of the parameter physical range. A multi-population genetic coding structure is constructed, wherein the number of parallel populations is fixed at 4, each population contains not less than 50 groups of individuals, and the population initialization process is generated by a uniform distribution sampling method to form a population initialization parameter set. Five rounds of individual selection, crossover and recombination, and disturbance transformation operations are performed on each parallel population, the specific process including: first, performing a roulette wheel selection on all individuals in the population, and the proportion of retained individuals is not less than 60%; then, performing a crossover and recombination operation on the selected individuals in pairs, the crossover method is single-point crossover, and the crossover probability is set to 0.9; then, applying a disturbance transformation operation to the individuals after the crossover, the disturbance method is Gaussian disturbance, the disturbance mean is 0, the disturbance standard deviation is set to 5% of the initial range of the parameter, and the transformation probability is set to 0.1. After completion, the total evaluation score of each transformed individual is calculated according to the fitness evaluation standard, and is classified and summarized according to the population number to form the first generation multi-objective population fitness data. Based on the above fitness data, the weights of each objective function in the original evaluation standard are adjusted once, the adjustment method is proportional offset adjustment, the offset amplitude is not more than ±0.03, and the basis is the variance data of the objective function score in the last generation. The weights of the objective functions with larger variance are adjusted upward, and the weights of the objective functions with smaller variance are adjusted downward to generate an updated fitness function. According to the updated fitness function and the uniform population migration mechanism, the four parallel populations are subjected to an interactive operation every 3 generations, the interactive method is to exchange the optimal solutions between the populations, the number of solutions exchanged each time is 5 groups, the results after the exchange are combined with the original population and sorted again to retain the first 50 groups to form population migration update data. Based on the updated fitness function and the population migration update data, a parallel evolution iteration process of not less than 20 rounds is performed, and the above selection, crossover, disturbance, evaluation and migration operations are repeated every round. Finally, a multi-objective optimization candidate solution set is obtained, in which the total value of the objective function score reaches a stable threshold (the score change rate is less than 1% for 5 consecutive generations).

[0064] The application can enhance the coverage of solution space, reduce the risk of falling into local optimum, and improve the global search ability by setting multiple populations for parallel evolution; the introduction of fitness evaluation criteria in genetic operations such as selection, crossover and mutation helps to improve the diversity of solutions and the evolution of population structure on the basis of preserving excellent individuals; the dynamic adjustment mechanism of the fitness evaluation criteria enables the optimization direction to be self-adapted according to the population performance of each generation, enhancing the response ability of the search strategy to the changes of the objective function; the solution transfer mechanism between populations at the preset migration frequency realizes the co-evolution of solutions, effectively promotes the sharing of excellent features between different evolution paths, and improves the convergence speed of solutions; the introduction of the updated fitness penalty function in multiple rounds of evolution iteration further strengthens the screening control of solutions that do not meet the constraint conditions, thereby improving the comprehensive quality of the candidate solution set in terms of multi-objective consideration and constraint satisfaction, and providing a solid foundation for finally selecting intelligent well completion parameter configuration with high practicability.

[0065] Preferably, step S32 comprises the following steps:

[0066] Step S321: performing individual selection operation on the population initialization parameter set to obtain a population selection result set;

[0067] Step S322: performing genetic crossover operation based on the population selection result set to obtain a crossover and recombination individual set;

[0068] Step S323: performing mutation and disturbance operation based on the crossover and recombination individual set to obtain a genetic variation individual set;

[0069] Step S324: performing individual good and bad judgment based on the genetic variation individual set and the fitness evaluation criteria to obtain individual fitness score data;

[0070] Step S325: performing constraint penalty correction based on the individual fitness score data and the dynamic penalty factor generation rule to obtain a modified fitness data set;

[0071] Step S326: grouping and summarizing the modified fitness data set according to the population label to obtain the first generation multi-objective population fitness data.

[0072] In the embodiment of the present application, for the population initialization parameter set, a roulette probability screening operation is performed on each population, the roulette weight is based on the score of the individual under the initial fitness evaluation standard, the reserved proportion is set to 60% of the number of original population individuals, the remaining individuals are discarded and do not participate in subsequent operations, and a population selection result set is obtained; the individuals in each selection result set are sequentially paired and grouped, and a single-point crossover operation is performed, specifically, a parameter at the same position of each pair of individuals is selected as a crossover point, the crossover position is uniformly extracted between the 3rd and 7th positions, a new crossover recombination individual set is generated, the fixed crossover probability in the crossover operation process is 0.9, and the number of individuals after crossover is consistent with the original population; the disturbance operation is performed on the crossover recombination individual set one by one, the disturbance mode is Gaussian disturbance, the disturbance mean is set to 0, the standard deviation is set to 5% of the original definition interval of the parameter, the disturbance probability is 0.1, and the parameter value exceeding the original physical boundary after disturbance is limited within the boundary range through the truncation operation, and a genetic variation individual set is obtained; using the fitness evaluation standard, the weighted score of each genetic variation individual is calculated, summed, and the individual total score data is formed, and then the individual fitness score data is formed; according to the dynamic penalty factor generation rule, the constraint condition test is performed on each genetic variation individual to obtain the corresponding constraint violation deviation value, and then the deviation value is matched with the preset penalty mapping relationship to generate a penalty correction value, and finally the original fitness score of each individual is reduced by the corresponding penalty correction value to form a modified fitness data set; the modified fitness data set is grouped according to the population number, and the scores of all individuals in each group of data are classified and summarized to form the first generation multi-objective population fitness data.

[0073] The individual selection operation can effectively retain individuals with high fitness in the current population, enhance the continuity of high-quality genes, and provide a stable foundation for subsequent population evolution; the genetic crossover operation promotes the combination and recombination of different characteristics in the population, expands the search space and accelerates the diversity expansion of the population in the solution domain; the mutation disturbance operation further increases the diversity level of the solution, strengthens the exploration ability of the population in the unknown area, and improves the innovation of the overall optimization; the fitness evaluation standard is introduced in the individual quality judgment link, each individual is assigned a clear evaluation value, which helps to build a unified evaluation scale and accurately identify potential optimal solutions; the dynamic penalty factor correction mechanism punishes and suppresses individuals that do not meet the constraint conditions, effectively guides the optimization process to avoid infeasible regions, and thus improves the overall constraint satisfaction of the population; finally, the population tags are used for grouping and summarizing, so that the fitness evaluation results have clear structure division characteristics, and provide stable and high-resolution population fitness basic information for subsequent collaboration of multi-population parallel evolution.

[0074] Preferably, step S4 comprises the following steps:

[0075] Step S41: based on the dynamic penalty factor generation rule, the violation degree of the multi-objective optimization candidate solution set is evaluated, and penalty scoring data is obtained;

[0076] Step S42: threshold judgment is performed on the penalty scoring data, and the out-of-limit solution is removed, and a feasible solution screening result is obtained.

[0077] Step S43: based on the feasible solution screening result, a candidate solution subset is extracted, and a feasible solution set is obtained.

[0078] In the embodiment of the application, for the multi-objective optimization candidate solution set, the corresponding key completion parameter combination of each solution is extracted, and all constraint condition parameters are extracted based on the constraint function definition set. The constraint condition checking operation is performed on each solution, which specifically includes limit constraint judgment (whether exceeding the physical upper and lower limits of the parameters), process boundary test (whether meeting the safety boundary of the completion process), and logic mutual exclusion test (whether violating the process dependent logic between parameters). Each type of constraint test result is quantized as a binary value of 0 or 1. Then, all constraint test results are input into the dynamic penalty factor generation rule, and a dynamic weight is assigned to each violation item. The dynamic weight is adjusted according to the current score fluctuation rate of the associated target function set. The score fluctuation rate calculation formula is: σ i / μ i , where σ i is the standard deviation of the score of target function i in the current population, μ i is the mean value. Finally, a violation deviation comprehensive weight value for each solution is formed. The value is combined with the original fitness score to construct a penalty scoring expression. The final penalty scoring value of each candidate solution is calculated using the expression, and the penalty scoring data is obtained. The threshold judgment operation is performed on the penalty scoring data. The maximum allowable value of the violation score is set to 10% of the total value of the target function score. The solution exceeding this proportion is considered as an out-of-limit solution, which is removed from the penalty scoring data. The remaining solutions that do not exceed the limit are marked as feasible solutions, and a feasible solution screening result is formed. Based on the feasible solution screening result, all solutions that pass the screening are extracted from the original multi-objective optimization candidate solution set, and a final feasible solution set is formed.

[0079] The application generates a rule of dynamic penalty factor for breach evaluation, so that the deviation degree of each candidate solution under the constraint condition is quantitatively expressed, thereby improving the accuracy of the feasibility determination of the solution; the mechanism implements differentiated scoring according to the breach amplitude of the solution, strengthens the distinguishing ability of the mild breach and the serious breach scheme, and improves the flexibility of the punishment intensity regulation; the threshold judgment standard is set to ensure that the screening process has a clear judgment basis, avoid the misjudgment risk caused by the fuzzy decision, and improve the stability and repeatability of the feasible solution screening; the elimination operation of the over-limit solution directly reduces the occupation of invalid solutions to subsequent computing resources, and improves the overall optimization efficiency; based on the screening result, a subset is extracted, and the feasible solution set finally formed not only guarantees the multi-objective response performance, but also strictly meets the physical constraints and engineering safety boundaries, thereby laying a solid foundation for subsequent optimal sorting and optimal scheme extraction.

[0080] Preferably, step S41 comprises the following steps:

[0081] Step S411: constraint condition extraction is performed on the multi-objective optimization candidate solution set to obtain solution set constraint feature data;

[0082] Step S412: constraint deviation analysis is performed based on the solution set constraint feature data and the target function set to obtain solution set breach deviation data;

[0083] Step S413: dynamic penalty strength mapping is performed on the solution set breach deviation data according to a dynamic penalty factor generation rule to obtain solution set penalty factor data;

[0084] Step S414: a breach scoring expression is constructed based on the solution set penalty factor data and the fitness evaluation standard to obtain a solution set penalty scoring function;

[0085] Step S415: the solution set penalty scoring function is calculated for the multi-objective optimization candidate solution set to obtain penalty scoring data.

[0086] In the embodiment of the application, for the multi-objective optimization candidate solution set, key completion parameter combinations are extracted one by one, and parameter-level constraint condition extraction operations are performed according to the constraint function definition set. The constraint types include three types of parameter physical limits, process safety threshold values and working condition logical dependency constraints. The physical limits are determined by comparing the upper and lower limit values to determine whether they are out of limits. The process safety threshold values are confirmed by comparing the field experience threshold range to determine whether they are out of limits. The logical dependency relationship is searched for conflict relationship through the defined exclusion or inclusion rule table. All results are summarized to form solution set constraint feature data. The data format is a two-dimensional matrix. Each row corresponds to a candidate solution, and each column is the test result value of a certain constraint. After binary processing, a breach matrix is formed. Joint analysis is performed on the solution set constraint feature data and the target function set. The deviation degree of each solution under each constraint dimension is calculated by statistical analysis. The deviation degree is represented by a normalized deviation degree. The formula is: (v i -bi ) / r i , where v i is the constraint value of the current solution, b i is the lower or upper limit boundary value of the constraint, r i is the standard deviation of the constraint in the historical data, the solution set violation deviation data is obtained, and the storage structure is a multi-dimensional deviation vector corresponding to the candidate solution number; according to the dynamic penalty factor generation rule, the mapping operation is performed on each item of the solution set violation deviation data, and the mapping mode adopts linear enhancement mapping, that is, each item of deviation is multiplied by the target function score fluctuation rate coefficient to which the current constraint belongs, and the score fluctuation rate calculation mode is σ i / μ i , where σ i and μ i are the standard deviation and mean of the target function j in the current candidate solution set, and the result is summarized to form the solution set penalty factor data, and the greater the value, the stronger the penalty; based on the solution set penalty factor data and the fitness evaluation standard, a violation scoring expression is defined, and the expression structure is F=F0-∑(w k ×p k ), where F is the final score value, F0 is the original fitness evaluation score, w k is the normalized weight of the kth target function, and p k is the constraint penalty factor value associated with the kth target function. The expression in the structure ensures that the score is negatively correlated with the constraint penalty strength, and finally forms the solution set penalty scoring function; the penalty scoring function is called for each group of solutions of the multi-objective optimization candidate solution set, and the penalty correction score value is calculated and output in turn, and the result forms the penalty scoring data.

[0087] The present application ensures that the constraint properties of each candidate solution are completely quantified before evaluation through constraint condition extraction, improves the clarity of the data structure and the pertinence of subsequent analysis; the constraint deviation analysis process clearly defines the deviation degree of each solution under different physical constraint indicators, providing an accurate basis for the subsequent penalty mechanism; the penalty factor data formed by associating the deviation data with the dynamic penalty factor generation rule makes the penalty strength proportional to the violation severity, enhancing the difference and adaptability of violation identification; the introduction of the fitness evaluation standard constructs a violation scoring expression, making the scoring function target-oriented and ensuring that a unified evaluation logic is formed between the response to constraint violation and the overall optimization goal; the finally output penalty scoring data has numerical continuity and comparability, and can effectively distinguish between boundary feasible solutions and obvious violation solutions in the screening stage, thereby improving the precision, stability and calculation efficiency of the entire feasibility evaluation system.

[0088] Preferably, step S5 comprises the following steps:

[0089] Step S51: performing target function response extraction on the feasible solution set to obtain a multi-objective response data set;

[0090] Step S52: non-dominated solution identification operation is performed based on the multi-objective response data set, to obtain a Pareto solution candidate set;

[0091] Step S53: the crowding distance of the Pareto solution candidate set is calculated, to obtain Pareto boundary level data;

[0092] Step S54: weighted sorting is performed based on the Pareto boundary level data and the standardized target weight set, to obtain a multi-objective preferred sequence;

[0093] Step S55: the first-position solution in the multi-objective preferred sequence is extracted, to obtain an optimal intelligent well completion parameter configuration scheme.

[0094] In the embodiment of the application, based on the feasible solution set, the key well completion parameter combinations are read one by one and input into the hierarchical agent model for target function response calculation, to obtain corresponding multi-objective response values, and a two-dimensional response matrix is constructed with the feasible solution number as an index, wherein each row represents a solution and each column corresponds to a target function value, forming a multi-objective response data set; according to the Pareto optimality determination standard, the data set is compared pair by pair, for any two solutions i and j, if solution i is not inferior to solution j in all target function values and is superior to solution j in at least one target, then it is determined that i dominates j, and all solutions that are not dominated by any other solution are marked as non-dominated solutions, and the set is output as a Pareto solution candidate set; according to the multi-objective response values in the Pareto solution candidate set, the crowding distance of each solution is calculated, the calculation method is as follows: each target function is arranged in ascending order according to the numerical value, and the position of each solution in the current target function is recorded, then the linear difference accumulation is performed on the adjacent distance of each solution in each target function, to obtain the total crowding distance value, the greater the value, the more sparse the solution is located in the area, and the calculation result forms the Pareto boundary level data; using the standardized target weight set, weighted sorting operation is performed on each solution in the Pareto boundary level data, the sorting score is a single-value index obtained by accumulating the corresponding weight after multiplying the target response value, and the crowding distance value is used as a secondary criterion, when the sorting score is consistent, the solution with a larger crowding distance is preferentially selected, and finally a multi-objective preferred sequence is formed; the first-position solution in the multi-objective preferred sequence is directly extracted, and the corresponding key well completion parameter combination is the optimal intelligent well completion parameter configuration scheme, which meets the constraint condition and has the optimal comprehensive performance under the multi-objective evaluation system.

[0095] The objective function response extraction process in the application ensures that the performance of each solution is fully quantified, and provides a stable input basis for subsequent identification of advantages and disadvantages; the non-dominated solution identification mechanism can filter out a solution set that cannot be strictly surpassed by other solutions on any target in the background of multiple target conflicts, and effectively preserve balance and diversity; the calculation of the crowding distance further depicts the distribution density of the solution in the target space, strengthens the difference in the structure of the solution set, and makes the subsequent ordering have higher discrimination; the introduction of the standardized target weight set for weighted ordering effectively integrates the priority information between multiple targets, and improves the rationality and pertinence of the ordering sequence in the application decision; the first solution of the final extraction ordering is taken as the optimal configuration scheme, which not only guarantees that it has a comprehensive advantage in multiple target dimensions, but also significantly improves the decision-making precision, landing feasibility and global optimality of intelligent well completion operation parameter configuration.

[0096] Therefore, from any viewpoint, the embodiments should be considered as exemplary and non-limiting, the scope of the application is not limited by the above description, and all changes falling within the meaning and scope of the equivalent elements of the application file are intended to be included in the application.

[0097] The above description is only a specific implementation of the application, enabling those skilled in the art to understand or implement the application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart well completion multi-objective optimization method based on a multi-objective genetic algorithm, characterized in that, Includes the following steps: Step S1: Obtain the complete set of completion parameters and the set of objective functions for intelligent well completion; reduce the dimensionality of the complete set of completion parameters to obtain the set of key completion parameters; construct the dynamic penalty factor generation rule; Step S2: Construct a multi-level surrogate modeling system based on the set of key completion parameters and the set of objective functions to obtain a hierarchical surrogate model; construct a multi-objective optimization objective function evaluation system to obtain a fitness evaluation standard; Step S3: Perform multi-population parallel genetic evolution based on the hierarchical surrogate model and fitness evaluation criteria to obtain a multi-objective optimization candidate solution set; Step S4: Use the dynamic penalty factor generation rule to constrain the candidate solution set of multi-objective optimization to obtain a feasible solution set; Step S5: Perform Pareto boundary analysis and multi-objective sorting on the feasible solution set to obtain the optimal intelligent well completion parameter configuration scheme.

2. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Collect multi-source intelligent well completion related data to obtain the complete set of well completion parameters and the set of objective functions; Step S12: Perform correlation analysis on the entire set of completion parameters based on the objective function set to obtain the initial feature importance ranking; Step S13: Perform principal component analysis on the initial feature importance ranking to obtain the key principal component set; Step S14: Perform feature space mapping transformation on the complete set of completion parameters based on the key principal component set to obtain a dimensionless parameter representation; Step S15: Perform hierarchical clustering on the dimensionality-reduced parameter representation to obtain the parameter hierarchical grouping structure; Step S16: Select a high-weight subset based on the parameter hierarchical grouping structure and objective function set to form a key completion parameter set; Step S17: Construct an initial family of constraint functions based on the fitness evaluation criteria to obtain the constraint function definition set; Step S18: Perform function transformation and weight adjustment on the constraint function definition set to obtain the dynamic penalty factor generation rule.

3. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S2 involves constructing a multi-level proxy modeling system based on the set of key completion parameters and the set of objective functions, including: The key completion parameter set is stratified and classified by feature to obtain completion layer structure data; Based on the well completion layer structure data and the objective function set, sub-models for each layer are constructed to obtain a set of surrogate model substructures. Model fusion is performed on the set of substructures of the proxy model to obtain a unified hierarchical proxy model.

4. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 3, characterized in that, Model fusion of the proxy model substructure set includes: Perform output consistency analysis on the set of substructures of the proxy model to obtain normalized model response data; A cross-layer fusion mapping structure is constructed based on the model response normalized data to obtain the structure mapping matrix; Inter-layer information aggregation is performed on the structure mapping matrix to obtain the aggregated output representation; By combining aggregated output representations with a set of objective functions to perform joint regression modeling, a unified hierarchical proxy model is obtained.

5. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S2 involves constructing an evaluation system for the objective function of multi-objective optimization, including: Based on the set of objective functions, a hierarchical structure of objective functions is constructed to obtain the objective function weight mapping data; The objective function weight mapping data is normalized to obtain a standardized objective weight set; A family of fitness evaluation functions for multi-objective optimization is constructed based on the standardized objective weight set and objective function set, and a fitness evaluation standard is obtained.

6. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Construct an initial multi-population genetic coding system based on the hierarchical surrogate model and fitness evaluation criteria to obtain the population initialization parameter set; Step S32: Perform selection, crossover, and mutation operations in parallel for each population in the population initialization parameter set, and judge their quality by combining the fitness evaluation criteria to obtain the fitness data of the first generation of multi-objective populations; Step S33: Adjust the fitness evaluation criteria based on the fitness data of the first-generation multi-objective population to obtain the updated fitness function; Step S34: Perform the solution exchange operation for each population in parallel evolution according to the preset migration frequency based on the updated fitness function to obtain population migration update data; Step S35: Perform multiple rounds of parallel evolutionary iterations based on the updated fitness penalty function and population migration update data to obtain a multi-objective optimization candidate solution set.

7. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 6, characterized in that, Step S32 includes the following steps: Step S321: Perform individual selection operation on the population initialization parameter set to obtain the population selection result set; Step S322: Perform genetic crossover operation based on the population selection result set to obtain the crossover and recombination individual set; Step S323: Perform mutation perturbation operation on the crossover and recombination individual set to obtain the genetically mutated individual set; Step S324: Based on the genetic variation individual set and fitness assessment criteria, judge the quality of individuals to obtain individual fitness score data; Step S325: Based on the individual fitness score data and the dynamic penalty factor generation rule, perform constraint penalty correction to obtain the corrected fitness dataset; Step S326: Group and summarize the corrected fitness dataset according to the population label to obtain the first generation multi-objective population fitness data.

8. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Evaluate the default degree of the candidate solution set for multi-objective optimization based on the dynamic penalty factor generation rule to obtain penalty score data; Step S42: Perform threshold judgment on the penalty scoring data, remove solutions that exceed the limit, and obtain the feasible solution screening results; Step S43: Extract a subset of candidate solutions based on the feasible solution screening results to obtain a feasible solution set.

9. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 8, characterized in that, Step S41 includes the following steps: Step S411: Extract constraints from the candidate solution set for multi-objective optimization to obtain constraint feature data of the solution set; Step S412: Perform constraint deviation analysis based on the solution set constraint feature data and the objective function set to obtain solution set default deviation data; Step S413: Perform dynamic penalty intensity mapping on the default deviation data of the solution set according to the dynamic penalty factor generation rule to obtain the penalty factor data of the solution set; Step S414: Construct a default scoring expression based on the solution set penalty factor data and fitness evaluation criteria to obtain the solution set penalty scoring function; Step S415: Calculate the solution set penalty scoring function for the candidate solution set of multi-objective optimization to obtain the penalty scoring data.

10. The intelligent well completion multi-objective optimization method based on multi-objective genetic algorithm according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Perform objective function response extraction on the feasible solution set to obtain a multi-objective response dataset; Step S52: Perform non-dominated solution identification operation based on the multi-objective response dataset to obtain a Pareto solution candidate set; Step S53: Calculate the crowding distance of the Pareto solution candidate set to obtain Pareto boundary level data; Step S54: Perform weighted sorting based on Pareto boundary level data and standardized target weight set to obtain multi-objective optimization sequence; Step S55: Extract the first and second solutions from the multi-objective optimization sequence to obtain the optimal intelligent well completion parameter configuration scheme.

Images