A model preprocessing method and system for lithologic identification model

Abstract

The present invention provides a model preprocessing method and system for a lithologic identification model, the method comprising the following steps: obtaining parameter ranges of multiple hyperparameters in a pre-constructed lithologic identification model, constructing a first preset number of individuals based on the parameter ranges of the multiple hyperparameters, completing an initialization population, and performing a preset round of updates on the initialization population; including global optimization and local optimization in each update round; in the global optimization step, updating the position of the individual based on the distance between the individual and a target individual to obtain an updated position of the individual; in the local optimization step, re-optimizing the position of each individual in the global optimization population based on an individual influencing parameter group to complete local optimization; after the preset round of updates, screening the final individual by fitness, applying the hyperparameters of the individual to the lithologic identification model for model training, and performing lithologic identification after the training is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of lithology identification, and in particular to a model preprocessing method and system for a lithology identification model. Background Art

[0002] The use of machine learning models for lithology identification plays a crucial role in modern geological exploration and oil and gas resource development. Its importance is reflected in multiple dimensions, profoundly impacting the industry's efficiency, accuracy, and technological innovation. First, from an efficiency perspective, traditional lithology identification relies primarily on the experience and expertise of geologists. This process is often time-consuming and labor-intensive, and may be limited by a person's subjective judgment and knowledge. Machine learning models, on the other hand, can automatically extract features from large amounts of well logging data and, through algorithm optimization and learning, achieve rapid and accurate identification of different rock types. This not only greatly shortens the identification cycle and improves work efficiency, but also reduces the uncertainty caused by human factors, making the lithology identification results more objective and reliable.

[0003] Secondly, in terms of accuracy, machine learning models, trained on a large amount of well logging data with known lithology, are able to learn the complex relationships between rock types and logging responses, including subtle differences that are difficult to capture with traditional methods. This powerful learning capability enables machine learning models to demonstrate higher accuracy in lithology identification, especially in complex geological environments with the coexistence of multiple rock types. This not only helps geologists more accurately understand the distribution of subsurface rocks, but also provides a solid foundation for subsequent oil and gas resource assessment and development.

[0004] During the model construction phase, the model parameters need to be preliminarily determined. When processing data with a hierarchical structure and certain natural inspiration, most processing methods, due to the mismatch between parameter changes and the iterative process, make the results focus on emphasizing local search and ignoring global search, leading to local optimal problems, thereby reducing the robustness of the entire model, directly affecting the parallelism and efficiency of model execution, and slowing down the convergence speed of the model. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a model preprocessing method for a lithology identification model to eliminate or improve one or more defects in the prior art.

[0006] One aspect of the present invention provides a model preprocessing method for a lithology identification model, the method comprising the following steps:

[0007] Obtaining parameter ranges of multiple hyperparameters in a pre-built lithologic identification model, constructing a first preset number of individuals based on the parameter ranges of the multiple hyperparameters, completing the initialization of the population, and updating the initialization population for a preset number of rounds;

[0008] Each update round includes global optimization and local optimization;

[0009] In the global optimization step, the fitness of individuals in the current population is calculated, a second preset number of individuals are selected as target individuals based on the fitness of the individuals, and the positions of the individuals are updated based on the distances between the individuals and the target individuals to obtain updated positions of the individuals;

[0010] In the local optimization step, the individual influence parameter group of each individual after global optimization is calculated, wherein the individual influence parameter group includes at least one individual influence parameter, and the position of each individual in the global optimization group is optimized again based on the individual influence parameter group to complete the local optimization;

[0011] After a preset number of rounds of updates, the final individual is screened by fitness, and the lithology identification model uses the hyperparameters of the individual to perform model training, and performs lithology identification after the training is completed.

[0012] By adopting the above scheme, this scheme can pre-select the hyperparameters of the lithologic identification model and include two position optimization processes in each update round, namely global optimization and local optimization. In the process of global optimization, the target individual is determined by fitness, and the position of the target can be continuously iterated and tracked. By observing and analyzing the target, the purpose of accurately pursuing the target is achieved; in the process of local optimization, the individual influence parameters of the individual are introduced, and the influence of various aspects of the individual is taken into account. By appropriately adjusting these parameters, the state optimization process is realized, the behavioral efficiency of the population is improved, and the convergence speed is increased.

[0013] In some embodiments of the present invention, the global optimization step further comprises:

[0014] The position of the individual is updated based on the distance between the individual and the target individual, and the updated position of the individual is obtained as the first updated position;

[0015] Calculate the Gaussian distribution, calculate the second updated position of the individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, and use the second updated position of the individual as the final position of the individual in the global optimization step.

[0016] In some embodiments of the present invention, the position of the individual is updated based on the distance between the individual and the target individual. In the step of obtaining the updated position of the individual, the step length of each individual relative to a target individual is calculated based on the position of the individual and the position of the target individual, and the updated position of the individual is calculated based on the step length of the individual relative to each target individual.

[0017] In some embodiments of the present invention, in the steps of calculating the step length of each individual relative to a target individual based on the position of the individual and the position of the target individual, and calculating the updated position of the individual based on the step length of the individual relative to each target individual, the step length of the individual relative to a target individual is calculated based on the following formula:

[0018] ;

[0019] in, Indicates that individual 1 is relative to the target individual The step length, Indicates the target individual location, represents the coefficient vector of individual 1; Represents individual 1 and target individual distance.

[0020] The updated position of the individual is calculated based on the individual's step size relative to each target individual based on the following formula:

[0021] ;

[0022] in, represents the updated position of the individual, Represents the superposition value of the step length of the individual relative to each target individual, Indicates the number of target individuals.

[0023] In some embodiments of the present invention, in the step of calculating the second updated position of an individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, and using the second updated position of the individual as the final position of the individual in the global optimization step, the second updated position is calculated using the following formula:

[0024] ;

[0025] in, represents the second updated position, represents the first updated position, Indicates the preset maximum number of iterations, 、 、 and γ are preset constants, D represents the number of dimensions of the position vector, The value of one dimension in the position vector representing the first updated position, Indicates the position vector of the target individual corresponding to The value of the dimension, Represents the preset Gaussian mutation constant, Represents the position vector of the target individual; a position vector representing the first updated position; Represents a Gaussian distribution.

[0026] In some embodiments of the present invention, in the step of calculating the second updated position of an individual in a population based on the first updated position of the individual in the population and a Gaussian distribution, and taking the second updated position of the individual as the final position of the individual in the global optimization step, the second updated position of an individual is calculated based on each target individual, and the average of multiple second updated positions is calculated as the final position of the individual.

[0027] In some embodiments of the present invention, the individual influence parameters include a group attraction value, a dynamic inertia weight value, a group cohesion value, a target attraction value, and a hypothetical individual interference value. In the step of calculating the individual influence parameter group of each individual after global optimization, the group attraction value, the dynamic inertia weight value, the group cohesion value, the target attraction value, and the hypothetical individual interference value are calculated according to the following formula:

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] Among them, S, A, C, F and E represent the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and hypothetical individual interference value of an individual respectively; X represents the current position of the calculated individual, represents the current position of any individual j outside the individuals calculated in the current population; N represents the number of individuals outside the individuals calculated in the current population; represents the current speed of individual j; represents the attraction of the target individual to the calculated individual; Represents the interference value of the hypothetical individual on the calculated individual.

[0034] In some embodiments of the present invention, in the step of re-optimizing the position of each individual in the global optimization group based on the individual influence parameter group, the amount of movement of the individual is calculated based on the individual influence parameter group, and the target position of the individual is calculated based on the amount of movement of the individual.

[0035] In some embodiments of the present invention, in the step of calculating the movement amount of the individual's re-moval based on the individual influence parameter group, the movement amount of the individual's re-moval is calculated based on the following formula:

[0036] ;

[0037] in, represents the amount of movement of the individual again; s, a, c, f and e represent the calculation weights of the corresponding group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value; Indicates the distance between the individual’s current position and the final position of the previous round, Indicates correspondence The calculation weight of .

[0038] The second aspect of the present invention also provides a model preprocessing system for a lithologic identification model, which includes a computer device, wherein the computer device includes a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method described above.

[0039] The third aspect of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps implemented by the model preprocessing method of the aforementioned lithology identification model.

[0040] Additional advantages, objects, and features of the present invention will be described in part in the following description and will become apparent to those skilled in the art after studying the following or may be learned by practice of the present invention. The objects and other advantages of the present invention may be particularly pointed out and attained in the description and drawings.

[0041] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention.

[0043] Figure 1 A schematic diagram of an embodiment of a model preprocessing method for a lithology identification model of the present invention;

[0044] Figure 2 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 1;

[0045] Figure 3Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 2;

[0046] Figure 4 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 3;

[0047] Figure 5 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 4;

[0048] Figure 6 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 5;

[0049] Figure 7 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the Grey Wolf Algorithm (GWO) of the prior art using test function 6;

[0050] Figure 8 Schematic diagram showing the comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the grey wolf algorithm (GWO) of the prior art using test function 7;

[0051] Figure 9 Schematic diagram showing comparison of the results of the solution (GWO_FA) of experimental group 1 of the present invention and the Grey Wolf Algorithm (GWO) of the prior art using test function 8;

[0052] Figure 10 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 1;

[0053] Figure 11 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 2;

[0054] Figure 12 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 3;

[0055] Figure 13Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 4;

[0056] Figure 14 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 5;

[0057] Figure 15 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 6;

[0058] Figure 16 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 7;

[0059] Figure 17 Schematic diagram showing comparison of the results of the solution of experimental group 2 (GWO_FA_FWA) of the present invention, the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 8;

[0060] Figure 18 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 1;

[0061] Figure 19 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 2;

[0062] Figure 20 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the grey wolf algorithm (GWO) of the prior art using test function 3;

[0063] Figure 21Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 4;

[0064] Figure 22 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 5;

[0065] Figure 23 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 6;

[0066] Figure 24 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 7;

[0067] Figure 25 Schematic diagram showing comparison of the results of the solution of experimental group 3 (GWO_FA_FWA_DA) of the present invention, the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 1 (GWO_FA), and the Grey Wolf Algorithm (GWO) of the prior art using test function 8;

[0068] Figure 26 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 1;

[0069] Figure 27 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 2;

[0070] Figure 28 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 3;

[0071] Figure 29 Schematic diagram showing comparison of results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 4;

[0072] Figure 30 Schematic diagram showing comparison of results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 5;

[0073] Figure 31 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 6;

[0074] Figure 32 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 7;

[0075] Figure 33 Schematic diagram showing comparison of the results of the solution of experimental group 1 (GWO_FA), the solution of experimental group 2 (GWO_FA_FWA), the solution of experimental group 3 (GWO_FA_FWA_DA) and the solution of the prior art using test function 8;

[0076] Figure 34 This is a schematic diagram of the architecture of this solution;

[0077] Figure 35 This is a schematic diagram of the confusion matrix results of lithology identification in this scheme;

[0078] Figure 36 This is a curve diagram of the accuracy and loss value of the model in this solution during the training process. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0080] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0081] Introduction to existing technology:

[0082] The Gray Wolf Algorithm (GWA) demonstrates significant advantages in high-dimensional, complex spaces, particularly in identifying key factors and constructing structured models with clear mapping relationships. It has been widely applied in a variety of fields, including path planning, cluster analysis, feature selection, and power dispatch. In recent years, research on the GWA has continued to deepen, with many scholars expanding its theoretical framework and application scenarios. Furthermore, some researchers have combined optimization methods such as particle swarm optimization (PSO) and differential evolution (DE) to enhance the GWA's versatility and ability to escape local optima. In recent years, with the advancement of artificial intelligence and deep learning, the GWA has also been applied to neural network training, image processing, and complex engineering optimization problems, demonstrating strong adaptability and application value.

[0083] Despite its success in various fields, the Gray Wolf Algorithm (GWA) faces numerous challenges in practical applications. First, during the multidimensional feature selection process, the GWA fails to record historical data for each individual iteration, hindering its ability to obtain the global optimal solution. Furthermore, in task scheduling optimization, the uncertainty inherent in traditional combination strategies makes it difficult for the algorithm to strike a balance between precise and probabilistic solutions. Furthermore, the data processing process fails to simultaneously address the requirements of both homogeneous and heterogeneous data structures. These factors collectively lead to algorithmic optimization biases, impacting solution accuracy. Second, when solving complex optimization problems, the GWA's convergence speed and search efficiency still leave room for improvement. This is particularly true for high-dimensional and uncertain problems, where the algorithm is prone to falling into local optima, impacting the final optimization results. Furthermore, current research lacks systematic optimization methods for improving the balance between the GWA's global exploration and local exploitation capabilities.

[0084] To address these issues, this proposal proposes an improved Gray Wolf Algorithm (GWA) optimization strategy. First, the GWA's decision hierarchy and execution layering are leveraged to construct an overall strategy framework. The absorption parameter and partner list mechanism of the Firefly Photon Algorithm are introduced to record historical information, allowing for dynamic parameter adjustments during the search process. This ensures consistency between path exploration and target location, while also avoiding local optima. Second, to address the needs of discrete multi-processor task scheduling, the Fireworks Algorithm is integrated to enhance the algorithm's stability and flexibility during execution. Furthermore, by calculating the complexity of parameter configurations and optimizing crossover and mutation operations, the algorithm's effectiveness is improved, mitigating performance degradation caused by premature convergence. Finally, this paper combines the optimization processes of the Firefly Algorithm and the Dragonfly Algorithm, leveraging the synergy of these two metaheuristic algorithms to enhance the GWA's exploratory and exploitative capabilities. A comparative optimization method is introduced to further improve solution quality, accelerate convergence, and enhance search efficiency. Overall, this optimization strategy effectively addresses the shortcomings of the traditional GWA and provides a superior solution for complex high-dimensional problems.

[0085] It can be seen from this that there is a bottleneck problem in how to break through the Gray Wolf Algorithm in the optimization and solution process. Although the current optimization method has improved the computational performance of the algorithm to a certain extent, it is still difficult to find the optimal solution in high-dimensional complex problems, especially in key areas such as task scheduling, feature selection and resource allocation. The solution accuracy and adaptability of the existing algorithm still need to be improved. Therefore, the optimization strategy proposed in this scheme not only focuses on the balance between the search ability and development ability of the Gray Wolf Algorithm, but also combines the advantages of multiple metaheuristic algorithms to achieve a more intelligent and efficient solution process. In addition, the improved algorithm of this scheme can adapt to more complex optimization environments, including nonlinear, multi-constrained and dynamically changing optimization problems, providing theoretical support and practical basis for further expanding the application scenarios of the Gray Wolf Algorithm.

[0086] like Figure 1 As shown, the present invention proposes a model preprocessing method for a lithology identification model, the method comprising the following steps:

[0087] Step S100, obtaining parameter ranges of multiple hyperparameters in a pre-built lithologic identification model, constructing a first preset number of individuals based on the parameter ranges of the multiple hyperparameters, and completing the initialization of the population;

[0088] And update the initialized population for a preset number of rounds;

[0089] Each update round includes global optimization and local optimization;

[0090] Step S210, in the global optimization step, calculating the fitness of individuals in the current population, selecting a second preset number of individuals as target individuals based on the fitness of the individuals, and updating the positions of the individuals based on the distances between the individuals and the target individuals to obtain updated positions of the individuals;

[0091] Step S220, in the local optimization step, calculating the individual influence parameter group of each individual after global optimization, wherein the individual influence parameter group includes at least one individual influence parameter, and re-optimizing the position of each individual in the global optimization group based on the individual influence parameter group to complete the local optimization;

[0092] Step S300: After a preset number of rounds of updates, the final individual is screened by fitness, and the lithology recognition model is trained using the hyperparameters of the individual. After the training is completed, lithology recognition is performed.

[0093] The model of this solution adopts the BiLSTM model.

[0094] By adopting the above scheme, this scheme can pre-select the hyperparameters of the lithologic identification model and include two position optimization processes in each update round, namely global optimization and local optimization. In the process of global optimization, the target individual is determined by fitness, and the position of the target can be continuously iterated and tracked. By observing and analyzing the target, the purpose of accurately pursuing the target is achieved; in the process of local optimization, the individual influence parameters of the individual are introduced, and the influence of various aspects of the individual is taken into account. By appropriately adjusting these parameters, the state optimization process is realized, the behavioral efficiency of the population is improved, and the convergence speed is increased.

[0095] In some embodiments of the present invention, the global optimization step further comprises:

[0096] The position of the individual is updated based on the distance between the individual and the target individual, and the updated position of the individual is obtained as the first updated position;

[0097] Calculate the Gaussian distribution, calculate the second updated position of the individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, and use the second updated position of the individual as the final position of the individual in the global optimization step.

[0098] In some embodiments of the present invention, the position of the individual is updated based on the distance between the individual and the target individual. In the step of obtaining the updated position of the individual, the step length of each individual relative to a target individual is calculated based on the position of the individual and the position of the target individual, and the updated position of the individual is calculated based on the step length of the individual relative to each target individual.

[0099] In some embodiments of the present invention, in the steps of calculating the step length of each individual relative to a target individual based on the position of the individual and the position of the target individual, and calculating the updated position of the individual based on the step length of the individual relative to each target individual, the step length of the individual relative to a target individual is calculated based on the following formula:

[0100] ;

[0101] in, Indicates that individual 1 is relative to the target individual The step length, Indicates the target individual location, represents the coefficient vector of individual 1; Represents individual 1 and target individual distance.

[0102] The updated position of the individual is calculated based on the individual's step size relative to each target individual based on the following formula:

[0103] ;

[0104] in, represents the updated position of the individual, Represents the superposition value of the step length of the individual relative to each target individual, Indicates the number of target individuals.

[0105] In the specific implementation process, the following formula can be used to calculate the updated position of the individual:

[0106]

[0107] t represents the current iteration round.

[0108] The calculation method can also be used:

[0109] ;

[0110] ;

[0111] .

[0112] Specifically, This method simulates the exploration behavior of gray wolves when searching for prey. By introducing randomness, this method forces individuals to deviate from their current target position, thereby performing a global search and preventing the algorithm from falling into a local optimal solution too early. Indicates attraction. Every individual has its own unique attraction. Indicates and The distance between locations i and j. Observation coefficient parameter It is important in determining the convergence speed and behavior of the FA algorithm, as it affects the variation of attraction and, in theory, its value ; but in most applications, usually from 0.01 to 100, = 1 and ∈[0, 1].

[0113] During implementation, traditional algorithms lack detailed descriptions of individual execution or state changes, resulting in poor performance in distinguishing individual effectiveness. To address this issue, this paper introduces the Fireworks algorithm to address the shortcomings of traditional algorithms in expressing natural laws. This algorithm can comprehensively express individual mutations, individual strengths, the range of changes in individual iteration positions, the state of movement, as well as mapping laws and selection strategies.

[0114] First, based on the population initialization situation, indicators such as the intensity of state change, the threshold of state change, and the vector value of displacement are proposed. A good fitness value produces the optimal number of individuals for benign iterative calculation. Secondly, based on the fitness value, the effectiveness and reliability of the population state change are determined. Then, an optimal range of state change is selected to accommodate individuals with weaker strength as much as possible, avoid the "premature aging" phenomenon of individuals, and maximize the capabilities of individuals with weaker strength. Finally, the state change intensity and state change threshold are calculated to implement vector operations on individual displacements, as shown in the following formula:

[0115]

[0116] Represents the value of the position vector of individual i in the kth dimension. Indicates that after the population state changes i times in the kth dimension. Represents 0 and A random number between .

[0117] This paper uses Gaussian distribution to express pre-formed Gaussian mutations of individuals in any dimension in the population; in addition, the modular operation rule is used to map individuals that cross the boundary back to the optimal range of the population, thereby further improving the diversity of the population. The formula is as follows:

[0118]

[0119]

[0120]

[0121] In each iteration of the gray wolf algorithm, a portion of gray wolf individuals are selected for Gaussian mutation operation to achieve Gaussian mutation. The formula is as follows:

[0122]

[0123] .

[0124] Represents the position vector of the Gaussian mutation individuals generated by population i in the kth dimension. n follows a Gaussian distribution with mean 1 and variance 1. and They represent the upper and lower bounds of individual i in the kth dimension respectively.

[0125] In some embodiments of the present invention, in the step of calculating the second updated position of the individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, and using the second updated position of the individual as the final position of the individual in the step of global optimization,

[0126] In order to accurately express the behavior of the algorithm after determining the target position, this solution expresses the process of approaching the target by tracking the moving state of the positioning target. Based on the perspective of the group, we gradually discover and use the parameters of the individual to optimize the static and dynamic behavior of the group. We analyze the coverage of the algorithm and the movement path of the group to express the implementation process of tracking the target at this stage. The linear decrease from 2 to 0 is expressed as follows:

[0127] ;

[0128] in, represents the second updated position, represents the first updated position, Indicates the preset maximum number of iterations, 、 、 and γ are preset constants, D represents the number of dimensions of the position vector, The value of one dimension in the position vector representing the first updated position, Indicates the position vector of the target individual corresponding to The value of the dimension, Represents the preset Gaussian mutation constant, Represents the position vector of the target individual; a position vector representing the first updated position; Represents a Gaussian distribution.

[0129] In some embodiments of the present invention, in the step of calculating the second updated position of an individual in a population based on the first updated position of the individual in the population and a Gaussian distribution, and taking the second updated position of the individual as the final position of the individual in the global optimization step, the second updated position of an individual is calculated based on each target individual, and the average of multiple second updated positions is calculated as the final position of the individual.

[0130] This scheme introduces the concept of group cohesion, and avoids conflicts between adjacent individuals by separating the group from other individuals. At the same time, the speed of each individual is considered from the perspective of group behavior, and the central attraction of the group is calculated by the group's vector to represent the cohesion of the population. The attraction of the objective function and the interference ability to the target are calculated. To this end, the influence range of each individual's group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value is set as the field, and by appropriately adjusting these parameters, the state optimization process is realized to improve the behavioral efficiency of the population.

[0131] In some embodiments of the present invention, the individual influence parameters include a group attraction value, a dynamic inertia weight value, a group cohesion value, a target attraction value, and a hypothetical individual interference value. In the step of calculating the individual influence parameter group of each individual after global optimization, the group attraction value, the dynamic inertia weight value, the group cohesion value, the target attraction value, and the hypothetical individual interference value are calculated according to the following formula:

[0132] ;

[0133] ;

[0134] ;

[0135] ;

[0136] ;

[0137] Among them, S, A, C, F and E represent the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and hypothetical individual interference value of an individual respectively; X represents the current position of the calculated individual, represents the current position of any individual j outside the individuals calculated in the current population; N represents the number of individuals outside the individuals calculated in the current population; represents the current speed of individual j; represents the attraction of the target individual to the calculated individual; Represents the interference value of the hypothetical individual on the calculated individual.

[0138] In some embodiments of the present invention, in the step of re-optimizing the position of each individual in the global optimization group based on the individual influence parameter group, the amount of movement of the individual is calculated based on the individual influence parameter group, and the target position of the individual is calculated based on the amount of movement of the individual.

[0139] In some embodiments of the present invention, in the step of calculating the movement amount of the individual's re-moval based on the individual influence parameter group, the movement amount of the individual's re-moval is calculated based on the following formula:

[0140] ;

[0141] in, represents the amount of movement of the individual again; s, a, c, f and e represent the calculation weights of the corresponding group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value; Indicates the distance between the individual’s current position and the final position of the previous round, Indicates correspondence The calculation weight of .

[0142] Experimental example

[0143] First, this project improves the Gray Wolf Algorithm by incorporating the firefly algorithm's description of the target's inherent attractiveness. This aims to force individuals to deviate from their current target position, preventing the algorithm from prematurely falling into a local optimum and thus entering the global search phase. For this experimental process, data generation and simulation were completed. The experimental environment was Windows 11 and MATLAB R2023b.

[0144] This solution uses a variety of test functions, as shown in Table 1 below:

[0145] Table 1

[0146]

[0147] Experimental Group 1

[0148] The protocol for experimental group 1 was:

[0149] Obtaining parameter ranges of multiple hyperparameters in a pre-built lithologic identification model, constructing a first preset number of individuals based on the parameter ranges of the multiple hyperparameters, completing the initialization of the population, and updating the initialization population for a preset number of rounds;

[0150] Each update round includes global optimization and local optimization;

[0151] In the global optimization step, the fitness of individuals in the current population is calculated, a second preset number of individuals are selected as target individuals based on the fitness of the individuals, and the positions of the individuals are updated based on the distances between the individuals and the target individuals to obtain updated positions of the individuals;

[0152] Specifically, the position of the individual is updated based on the distance between the individual and the target individual, and the updated position of the individual is obtained as the first updated position. The step length of each individual relative to a target individual is calculated based on the position of the individual and the position of the target individual, and the updated position of the individual is calculated based on the step length of the individual relative to each target individual.

[0153] The step length of an individual relative to a target individual is calculated based on the following formula:

[0154] ;

[0155] in, Indicates that individual 1 is relative to the target individual The step length, Indicates the target individual location, represents the coefficient vector of individual 1; Represents individual 1 and target individual distance.

[0156] The updated position of the individual is calculated based on the individual's step size relative to each target individual based on the following formula:

[0157] ;

[0158] in, represents the updated position of the individual, Represents the superposition value of the step length of the individual relative to each target individual, Indicates the number of target individuals;

[0159] After a preset number of rounds of updates, the final individual is screened by fitness, and the lithology identification model uses the hyperparameters of the individual to perform model training, and performs lithology identification after the training is completed.

[0160] Experimental Group 2

[0161] The solution of experimental group 2 adds the calculation of Gaussian distribution to the global optimization step of the solution of experimental group 1, calculates the second updated position of the individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, and uses the second updated position of the individual as the final position of the individual in the global optimization step;

[0162] The second updated position is calculated using the following formula:

[0163] ;

[0164] in, represents the second updated position, represents the first updated position, Indicates the preset maximum number of iterations, 、 、 and γ are preset constants, D represents the number of dimensions of the position vector, The value of one dimension in the position vector representing the first updated position, Indicates the position vector of the target individual corresponding to The value of the dimension, Represents the preset Gaussian mutation constant, Represents the position vector of the target individual; a position vector representing the first updated position; Represents a Gaussian distribution.

[0165] Experimental Group 3

[0166] The solution of Experimental Group 3 is based on Experimental Group 2. In the local optimization step, the individual influence parameter group of each individual after global optimization is calculated. The individual influence parameter group includes at least one individual influence parameter. Based on the individual influence parameter group, the position of each individual in the global optimization group is further optimized to complete the local optimization.

[0167] The individual influence parameters include group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value. In the step of calculating the individual influence parameter group of each individual after global optimization, the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value are calculated according to the following formula:

[0168] ;

[0169] ;

[0170] ;

[0171] ;

[0172] ;

[0173] Among them, S, A, C, F and E represent the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and hypothetical individual interference value of an individual respectively; X represents the current position of the calculated individual, represents the current position of any individual j outside the individuals calculated in the current population; N represents the number of individuals outside the individuals calculated in the current population; represents the current speed of individual j; represents the attraction of the target individual to the calculated individual; represents the interference value of the hypothetical individual on the calculated individual;

[0174] The amount of movement of the individual is calculated based on the individual influence parameter group, and the target position of the individual is calculated based on the amount of movement of the individual. The amount of movement of the individual is calculated based on the following formula:

[0175] ;

[0176] in, represents the amount of movement of the individual again; s, a, c, f and e represent the calculation weights of the corresponding group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value; Indicates the distance between the individual’s current position and the final position of the previous round, Indicates correspondence The calculation weight of .

[0177] Secondly, we use a random function to generate data and set the population size to 50. The upper and lower bounds of the variables are set to [10, 10] and [-10, -10] respectively. We simulate the fitness of the algorithm and the results are as follows: Figures 2 to 9 As shown in the figure, the introduction of the attraction parameter from the firefly algorithm improves the exploration behavior of the gray wolf algorithm when searching for prey, thereby improving the calculation strategy of the distance variable (GWO_FA). The horizontal axis of the image is the number of iterations, which increases from left to right. The vertical axis is the objective function value. As the number of iterations increases, the position variable gradually converges to the optimal value.

[0178] In summary, it can be seen that the improved algorithm GWO_FA not only decreases faster in the initial stage, but also reaches a lower fitness value within a smaller number of iterations, indicating that its optimization efficiency and effect are better than the original GWO algorithm.

[0179] Test function 、 and They all represent the input values ​​of each test function.

[0180] Figure 2 and Figure 3 Both are the performance of the algorithm on the unimodal function, but Figure 2 The search space of the corresponding function is larger than Figure 3Corresponding functions. During the iteration process, the algorithm's fitness values ​​gradually decrease and eventually tend to the minimum value. Therefore, the algorithm performs well on unimodal functions in different search spaces.

[0181] Figure 4 、 Figure 5 The performance of the algorithm on unimodal and multimodal functions respectively. Figure 4 It shows a very fast convergence speed and reaches a near-optimal fitness value within the first 5 iterations. Figure 5 In the , GWO_FA algorithm also shows fast convergence characteristics, but compared with the unimodal function, the number of iterations required to converge to the final stable value is slightly more, and it shows a certain volatility. This is because the multimodal function has multiple local optimal solutions, and the algorithm needs more iterations to avoid falling into the local optimum when exploring the solution space. Figure 5 The final fitness value stabilizes at around -7.89, which shows that the GWO_FA algorithm can also effectively find the global optimal solution on multimodal functions.

[0182] Figure 6 、 Figure 7 is the performance of the algorithm on different multimodal functions, Figure 6 The function needs to find the global minimum among multiple local minimums, which requires the algorithm to have good global search capabilities. Figure 7 The algorithm also needs to handle the complexity of exponential and cosine functions. Figure 6 、 Figure 7 From the results, we can also see that the GWO_FA algorithm has better global search capabilities than the GWO algorithm, and performs better in handling the complexity of exponential and cosine functions.

[0183] Figure 8 、 Figure 9 The performance of the algorithm on different multimodal functions. Figure 8 The function is a single variable function, and Figure 9 is a multivariable function. In both figures, the GWO_FA algorithm shows fast convergence characteristics, and the fitness value drops rapidly in the first few iterations.

[0184] Next, based on the improved GWO_FA algorithm, we continue to introduce the fireworks algorithm to perform Gaussian mutations on individuals to expand the scope of state changes and avoid the phenomenon of "premature aging" of individuals. We simulate this process. During the experiment, the population size, upper and lower bounds of the variables remain unchanged. The probability of Gaussian mutation is set to 0.1, the initial standard deviation of the Gaussian distribution is set to 0.5, and a random function is used to generate data. The individual position variables are simulated. The results are as follows Figures 10 to 17As shown in the simulation results: the circular curve is the unimproved Grey Wolf Algorithm (GWO); the asterisk curve is the Gaussian mutation process of the Fireworks Algorithm, which aims to perform Gaussian mutation operations on individuals in any dimension of the Grey Wolf Algorithm to further improve the population diversity (GWO_FA_FWA).

[0185] As can be seen from the figure, GWO_FA_FWA generally outperforms GWO in terms of convergence speed, especially on multi-peak functions. Furthermore, GWO_FA_FWA converges very quickly in the early stages of the iteration, quickly approaching the optimal solution and performing well on various objective functions. Although the fitness value of GWO_FA gradually decreases during the iterations, it ultimately surpasses that of GWO_FA_FWA. This indicates that the initial improved function GWO_FA did not find the optimal value in this experiment, failing to effectively explore the solution space during the search process, and exhibiting algorithmic instability.

[0186] In summary, the improved algorithm GWO_FA_FWA is more effective in dealing with complex problems and has good versatility and adaptability.

[0187] exist Figure 10 、 Figure 11 In the presented example, the GWO_FA_FWA algorithm quickly reduces the fitness value in the initial iteration and maintains the lowest fitness value throughout the entire process, showing the best convergence performance.

[0188] according to Figure 12 、 Figure 13 ,GWO_FA_FWA algorithm converges faster and achieves better results than the previous GWO_FA on unimodal and multimodal functions.

[0189] analyze Figure 14 From the curve, we can see that the fitness values ​​of the three algorithms all decrease rapidly in the initial iteration, but the GWO_FA algorithm eventually stabilizes at a higher fitness value. Compared with the GWO_FA_FWA algorithm, GWO_FA lacks an effective global search mechanism, which makes it difficult to escape from the local optimal solution. Therefore, the algorithm has poor stability and low final convergence performance. However, the GWO_FA_FWA algorithm has enhanced global search capabilities due to the introduction of the Gaussian mutation strategy, so it is Figure 14 、 15 In both cases, a stable state closer to the global optimal solution is achieved within a smaller number of iterations.

[0190] exist Figure 16 、 17In the experiment, the GWO_FA algorithm remains stable at a relatively high fitness value, which further confirms the conclusion drawn above, that is, the GWO_FA algorithm has low efficiency in global search when the function involved in the optimization problem has multiple local minima, the search space is large, or the dimension is high.

[0191] Finally, the dragonfly algorithm was introduced to consider the individual movement from the perspective of group behavior. The individual attributes were adjusted through parameters such as (s, a, c, f, and e), thereby improving the efficiency of group behavior. During the experiment, the population size, upper and lower bounds of the variables remained unchanged, the Gaussian mutation probability, and the initial standard deviation of the Gaussian distribution remained unchanged. A random function was used to generate data and simulate the individual position variables. The results are as follows Figures 18 to 25 As shown in the simulation results, the circular curve is the unimproved Grey Wolf Algorithm (GWO); and combined with Figure 1 The simulation results show that the asterisk curve is an improved algorithm (GWO_FA) that targets the exploration behavior of the gray wolf in the gray wolf algorithm when searching for prey by introducing the attraction parameter β from the firefly algorithm and improving the calculation strategy of the distance variable. The triangle curve is an improved algorithm (GWO_FA_FWA_DA) that performs Gaussian mutation on individuals, introduces the dragonfly algorithm, considers the influence of group factors on individuals, and iterates the individual positions again.

[0192] The GWO algorithm found a good solution early on, but subsequent iterations showed no significant improvement. The GWO_FA algorithm's curve decreased more rapidly in the early stages than the GWO algorithm and stabilized after approximately 10 iterations, indicating that the improved algorithm was able to find a more optimal solution more quickly. The GWO_FA_FWA_DA algorithm's curve converged the fastest of all the aforementioned algorithms, reaching its minimum fitness value within the first three iterations and maintaining it throughout the iterations. This experimental result demonstrates its strong optimization capabilities.

[0193] It can be seen that after 30 iterations, the GWO_FA_FWA_DA algorithm has a better average fitness value than other algorithms, is closer to the global optimal value, and has better performance than other algorithms before improvement.

[0194] exist Figure 18 In the , GWO_FA_FWA_DA algorithm converges faster, indicating that the algorithm has achieved a better balance between global search and local search. Figure 19 Although the convergence speed of GWO_FA_FWA_DA algorithm is still fast, Figure 18 Compared with , more iterations are required to reach a stable state. This shows that the algorithm has the ability to handle nonlinear and non-convex characteristics.

[0195] exist Figure 20 、 Figure 21 In the experiment, the GWO_FA_FWA_DA algorithm still shows a fast convergence characteristic. The convergence speed of GWO and GWO_FA is relatively slow, and there are slight fluctuations during the iteration process, but they also tend to a lower fitness value in the end.

[0196] Figure 22 There are slight fluctuations in the three algorithms during the iteration process, but in the end the GWO_FA_FWA_DA algorithm and the GWO algorithm converge to relatively low fitness values, and the GWO_FA algorithm converges to a higher fitness value. Figure 23 In the convergence process, the GWO_FA_FWA_DA algorithm maintains the relatively lowest fitness value, which proves that the GWO_FA_FWA_DA algorithm performs better when dealing with complex multi-modal problems.

[0197] exist Figure 24 and Figure 25 In the figure, the GWO and GWO_FA algorithms initially stabilized at relatively high fitness values ​​for a short period of time, eventually escaping from the local optimal solution during iteration. However, the GWO_FA_FWA_DA algorithm converged to the minimum fitness value early on, indicating that the GWO_FA_FWA_DA algorithm has good convergence characteristics and is not prone to falling into local optimal solutions.

[0198] The Grey Wolf Algorithm (GWO), the improved Grey Wolf Algorithm (GWO_FA) during the experiment, the Particle Swarm Optimization (PSO), and the Fruit Fly Algorithm (FOA) were selected for comparison with the final algorithm (GWO_FA_FWA_DA) obtained in this paper. The simulation process used random functions to generate data and simulated the convergence of the above algorithms on different functions. The results are as follows Figure 4 As shown: the dotted line is the particle swarm optimization algorithm (PSO); the square curve is the fruit fly algorithm (FOA); and combined Figures 26 to 33 The simulation results show that the circular curve is the unimproved Grey Wolf Algorithm (GWO); the asterisk curve is the algorithm (GWO_FA) that improves the calculation strategy of the distance variable by introducing the attraction parameter β from the Firefly Algorithm based on the exploration behavior of the Grey Wolf in searching for prey; the triangle curve is the improved algorithm (GWO_FA_FWA_DA) that performs Gaussian mutation on individuals, introduces the Dragonfly Algorithm, considers the influence of group factors on individuals, and iterates the individual positions again.

[0199] As can be seen from the figure, the GWO_FA_FWA_DA algorithm typically reaches a low fitness value within a relatively short number of iterations. In many cases, it approaches the final fitness value in early iterations, only to experience further declines in later iterations. Furthermore, the final algorithm in this paper achieves the lowest fitness value across various test functions, demonstrating excellent adaptability. However, GWO_FA has already exhibited significant instability in previous simulations, and this issue also recurs in this simulation, further validating the aforementioned findings.

[0200] From this, we can conclude that compared with the original GWO algorithm and other algorithms, the final GWO_FA_FWA_DA algorithm converges faster and is more stable in the process of finding the optimal solution. Whether it is a single-peak or multi-peak function, it can effectively find the optimal solution by continuously optimizing the quality of the solution.

[0201] exist Figure 26 、 Figure 27 In the figure, the GWO_FA_FWA_DA algorithm's fitness drops rapidly in the first few iterations. The PSO algorithm has a higher fitness value in both graphs, but its optimization performance is inferior to that of the other algorithms. The GWO and GWO_FA algorithms perform similarly in both graphs, ultimately converging to lower fitness values. The FOA algorithm maintains the lowest fitness value throughout the iterations, demonstrating the best optimization results.

[0202] Figure 28 The overall trend is similar to the above image. Figure 29 In the test, FOA converges slowly and finally stabilizes at around -4, performing the worst in the function test. This shows that multimodal problems can easily cause FOA to get lost in the local optimal solution.

[0203] exist Figure 30 The FOA algorithm has the slowest convergence speed, but it still maintains good optimization ability in the later stage of exploration and finally reaches the global optimal solution. The GWO_FA_FWA_DA algorithm proposed in this paper has a better overall convergence effect, but like other algorithms, it falls into a local optimal solution close to the global optimal solution. Figure 31 Among them, the GWO_FA_FWA_DA algorithm has the best overall performance, with a rapid decline in the initial stage, and finally reaching the minimum fitness value and remaining stable. It is followed by GWO_FA and GWO. The overall convergence speed of GWO is slightly lower than that of GWO_FA_FWA_DA algorithm, but they all converge to the global optimal solution in the end.

[0204] exist Figure 32 In the end, the GWO_FA algorithm converges to a higher fitness value, which further proves the conclusion that the algorithm has low stability. Figure 33The results show that the FOA algorithm has a strong global search capability in the early stage and can quickly find areas close to the optimal solution, which are around -0.8 and 101 respectively. However, the optimization accuracy is insufficient in the later stage and it lacks efficient local search capability, and ultimately fails to converge to the optimal solution.

[0205] This paper applies the GWO_FA_FWA_DA-BiLSTM-based lithology identification method to a publicly available well logging dataset specifically designed for machine learning applications in subsurface reservoir lithology identification and classification. The dataset contains seven attributes: _CAL (caliper); _GR (gamma ray); _SP (spontaneous potential); _LLD (laterolog deep); _LLS (laterolog shallow); _AC (acoustic); _DEN (density); _PEF (photoelectric absorption factor); and _Lith_Section (lithology).

[0206] This experiment uses a total of 2,385 well logging data points, including 1,179 for mudstone, 383 for siltstone, 444 for silty mudstone, 344 for muddy siltstone, and 35 for oil shale. Lithology prediction is performed using a training set:test set split ratio of 7:3. The experimental environment is Windows 11 and MATLAB R2023b.

[0207] For detailed method flow, see Figure 34 .

[0208] The experiment was run five times. Due to the initially wide search range, the algorithm could easily get lost in the high-dimensional search. Based on the initial experimental results, the range was gradually narrowed, and the final parameter settings were as follows: 20 gray wolves, 15 dragonflies, and 40 and 50 maximum iterations, respectively. The optimization algorithm used the optimal ranges for the number of hidden layer neurons, maximum number of training cycles, and learning rate for the BiLSTM model to be [50, 150], [30, 80], and [0.005, 0.01], respectively. The intensity of the Gaussian noise was also adjusted during the experiment: sigma (the standard deviation of the Gaussian noise) was reduced to 0.2 to avoid large fluctuations. Furthermore, a penalty function was added to penalize solutions that deviate significantly from the current optimal solution, improving convergence stability.

[0209] In the experiment, silty mudstone, mudstone, oil shale, siltstone, and muddy siltstone in the sample data are defined as 1, 2, 3, 4, and 5, respectively. CAL, GR, SP, LLD, LLS, DEN, PEF, and Lith_Section are selected as input values, and the lithology identification results are used as output values. The hyperparameter values ​​of the BiLSTM model are updated through GWO_FA_FWA_DA, and the lithology identification confusion matrix based on GWO_FA_FWA_DA-BiLSTM is obtained as shown below. Figure 35 As shown. Figure 35 As can be seen from the figure, the model correctly identified 33 data points of siltstone, with a probability of 24.8% for mudstone, 25.7% for siltstone, and 19.3% for argillaceous siltstone. The model correctly identified 302 data points of mudstone and incorrectly identified 49 data points, of which 2.0%, 2.6%, and 9.4% were identified as siltstone, siltstone, and argillaceous siltstone, respectively. The model correctly identified 3 data points of oil shale and incorrectly identified 7 data points. The model correctly identified 91 data points of siltstone and incorrectly identified 21 data points, of which 8.9%, 4.5%, 0.9%, and 4.5% were identified as siltstone, mudstone, oil shale, and argillaceous siltstone, respectively. The model correctly identified 52 data points of argillaceous siltstone and incorrectly identified 79 data points. The results show that the model has a good recognition effect on mudstone and siltstone.

[0210] Figure 36 The upper center line shows the accuracy trend on the training set. The lower solid line shows the loss trend on the validation set. Because the algorithm is in its initial exploratory phase, the target value fluctuates significantly, resulting in a high error. As the number of iterations increases, the accuracy curve stabilizes at around 75%, while the error gradually decreases, and the loss curve also stabilizes, indicating that the optimization objective is gradually converging.

[0211] Specifically, when processing data with a hierarchical structure and a certain degree of natural inspiration, the existing algorithms tend to emphasize local search and ignore global search due to the mismatch between parameter changes and the iterative process, which leads to the emergence of local optimal problems, thereby reducing the robustness of the entire algorithm, directly affecting the parallelism and efficiency of the algorithm execution, and slowing down the convergence of the algorithm. To address the above problems, this solution reasonably controls the exploration ability by setting balance parameters. In the process of controlling the movement mode through historical records, the position vector control coefficient is used. The purpose is to control the iterative process by changing the coefficient and absorb new parameters during the iterative process, thereby balancing the exploration ability of the algorithm. Secondly, by setting the adaptive dimension, a discretized mapping method of the continuous space is realized to control the Gaussian mutation process of the group, accelerate the convergence of the algorithm, shorten the time to find the optimal task scheduling sequence by using fitness as the calculation standard, and reduce the time complexity of the algorithm. Finally, the algorithm accuracy is improved through the robust process set back learning (OBL) algorithm. Finally, through four simulation experiments, the algorithm proposed in this paper is significantly superior to the traditional algorithm in terms of fitness, convergence speed and performance. Based on multiple experiments, a public well logging dataset was used to conduct experiments using this algorithm. The recognition rate was significantly better than the traditional BiLSTM model. This shows that the method designed in this scheme has a good effect compared with the traditional recognition method.

[0212] Aiming at processing high-dimensional data with hierarchical structure and certain natural inspiration, this scheme is based on the characteristics of the gray wolf algorithm. It considers the distance, dimension, behavioral process and execution efficiency of the group, and makes improvements in global search, target change mode and multi-dimensional attributes of individuals. It is compared with traditional algorithms and verified through three types of simulation methods. Finally, it is tested with actual data sets, which shows that the algorithm proposed in this paper is superior to several traditional algorithms in terms of fitness, convergence speed and performance.

[0213] An embodiment of the present invention also provides a model preprocessing system for a lithologic identification model, the system comprising a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, the processor being used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps implemented by the method described above.

[0214] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program implements the steps of the model preprocessing method for the lithologic identification model. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0215] It should be understood by those skilled in the art that the various exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether to implement the system in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried in a carrier wave.

[0216] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0217] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0218] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A model preprocessing method for a lithology identification model, characterized in that: The steps of the method include: Obtaining parameter ranges of multiple hyperparameters in a pre-built lithologic identification model, constructing a first preset number of individuals based on the parameter ranges of the multiple hyperparameters, completing the initialization of the population, and updating the initialization population for a preset number of rounds; Each update round includes global optimization and local optimization; In the global optimization step, the fitness of individuals in the current population is calculated, a second preset number of individuals are selected as target individuals based on the fitness of the individuals, and the positions of the individuals are updated based on the distances between the individuals and the target individuals, to obtain the updated positions of the individuals as the first updated positions; Calculate the Gaussian distribution, calculate the second updated position of the individual in the population based on the first updated position of the individual in the population and the Gaussian distribution, use the second updated position of the individual as the final position of the individual in the global optimization step, and use the following formula to calculate the second updated position: ； in, represents the second updated position, represents the first updated position, Indicates the preset maximum number of iterations, 、 、 and γ are preset constants, D represents the number of dimensions of the position vector, The value of one dimension in the position vector representing the first updated position, Indicates the position vector of the target individual corresponding to The value of the dimension, Represents the preset Gaussian mutation constant, Represents the position vector of the target individual; a position vector representing the first updated position; represents Gaussian distribution; In the local optimization step, the individual influence parameter group of each individual after global optimization is calculated, wherein the individual influence parameter group includes at least one individual influence parameter, and the position of each individual in the global optimization group is optimized again based on the individual influence parameter group to complete the local optimization; After a preset number of rounds of updates, the final individual is screened by fitness, and the lithology identification model uses the hyperparameters of the individual to perform model training, and performs lithology identification after the training is completed.

2. The model preprocessing method of the lithology identification model according to claim 1, characterized in that: The position of the individual is updated based on the distance between the individual and the target individual. In the step of obtaining the updated position of the individual, the step length of each individual relative to a target individual is calculated based on the position of the individual and the position of the target individual, and the updated position of the individual is calculated based on the step length of the individual relative to each target individual.

3. The model preprocessing method of the lithology identification model according to claim 2, characterized in that: In the steps of calculating the step length of each individual relative to a target individual based on the position of the individual and the position of the target individual, and calculating the updated position of the individual based on the step length of the individual relative to each target individual, the step length of the individual relative to a target individual is calculated based on the following formula: ； in, Indicates that individual 1 is relative to the target individual The step length, Indicates the target individual location, represents the coefficient vector of individual 1; Represents individual 1 and target individual distance; The updated position of the individual is calculated based on the individual's step size relative to each target individual based on the following formula: in, represents the updated position of the individual, Represents the superposition value of the step length of the individual relative to each target individual, Indicates the number of target individuals.

4. The model preprocessing method of the lithology identification model according to claim 1, characterized in that: In the step of calculating the second updated position of an individual in a population based on the first updated position of the individual in the population and the Gaussian distribution, and taking the second updated position of the individual as the final position of the individual in the global optimization step, the second updated position of an individual is calculated based on each target individual, and the average value of multiple second updated positions is calculated as the final position of the individual.

5. The model preprocessing method of the lithology identification model according to any one of claims 1 to 4, characterized in that: The individual influence parameters include group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value. In the step of calculating the individual influence parameter group of each individual after global optimization, the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value are calculated according to the following formula: ； ； ； ； ； Among them, S, A, C, F and E represent the group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and hypothetical individual interference value of an individual respectively; X represents the current position of the calculated individual, represents the current position of any individual j outside the individuals calculated in the current population; N represents the number of individuals outside the individuals calculated in the current population; represents the current speed of individual j; represents the attraction of the target individual to the calculated individual; Represents the interference value of the hypothetical individual on the calculated individual.

6. The model preprocessing method of the lithology identification model according to claim 5, characterized in that: In the step of re-optimizing the position of each individual in the global optimization group based on the individual influence parameter group, the movement amount of the individual is calculated based on the individual influence parameter group, and the target position of the individual is calculated based on the movement amount of the individual.

7. The model preprocessing method of the lithology identification model according to claim 5, characterized in that: In the step of calculating the movement amount of the individual's re-moval based on the individual influence parameter group, the movement amount of the individual's re-moval is calculated based on the following formula: ； in, represents the amount of movement of the individual again; s, a, c, f and e represent the calculation weights of the corresponding group attraction value, dynamic inertia weight value, group cohesion value, target attraction value and assumed individual interference value; Indicates the distance between the individual’s current position and the final position of the previous round, Indicates correspondence The calculation weight of .

8. A model preprocessing system for a lithology identification model, characterized in that: The system includes a computer device, the computer device includes a processor and a memory, the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method according to any one of claims 1 to 7.

Images