Search

CN-122020976-A - Online multi-objective optimization method and system for shield performance based on BO-GRU-KAN-MOGWO

CN122020976ACN 122020976 ACN122020976 ACN 122020976ACN-122020976-A

Abstract

The invention discloses a shield performance online multi-objective optimization method and system based on BO-GRU-KAN-MOGWO, wherein the method comprises the steps of selecting important slurry shield tunneling parameters and slurry system parameters as input characteristic variables, selecting key slurry shield construction performance indexes as output variables, and collecting data to construct an original sample set; the method comprises the steps of optimizing super parameters of a GRU-KAN algorithm by Bayesian optimization, constructing a BO-GRU-KAN prediction model, predicting tunneling speed, cutter head abrasion, overturning moment and super-excavation flow to obtain four regression prediction functions, introducing a SHapley addition interpretation method to determine key influence parameters, constructing a multi-objective optimization objective function by taking the four regression prediction functions as fitness functions, carrying out global optimization by adopting an improved multi-objective gray wolf optimization algorithm to determine an optimal slurry shield construction parameter combination, and meeting the tunnel construction requirements of high efficiency, low consumption, safety and accuracy under complex working conditions.

Inventors

  • CHEN HONGYU

Assignees

  • 武汉大学

Dates

Publication Date
20260512
Application Date
20251230

Claims (10)

  1. 1. The shield performance online multi-objective optimization method based on the BO-GRU-KAN-MOGWO is characterized by comprising the following steps: s1, selecting important slurry shield tunneling parameters and slurry system parameters as input characteristic variables, selecting key slurry shield construction performance indexes as output variables, collecting actual measurement data of the input characteristic variables and the output variables in the slurry shield construction process, and constructing an original sample set; S2, optimizing super parameters of a GRU-KAN algorithm by adopting Bayesian optimization, determining an optimal parameter combination by combining 5-fold cross validation, constructing a BO-GRU-KAN prediction model, predicting tunneling speed, cutter head abrasion, overturning moment and super-excavation flow, and obtaining four regression prediction functions; And S3, constructing a multi-objective optimization objective function by taking four regression prediction functions output by the BO-GRU-KAN model as fitness functions, performing global optimization by adopting an improved multi-objective gray wolf optimization algorithm, determining a unique optimal slurry shield construction parameter combination from the obtained Pareto optimal solution set by adopting a TOPSIS method, and simultaneously replacing the current time step optimization result with the previous time step performance index actual value to realize online dynamic optimization.
  2. 2. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO is characterized in that in step S1, the important slurry large shield tunneling parameters and slurry system parameters comprise 22 factors including total thrust, cutter head rotating speed, cutter head torque, cutter head extrusion force, air cushion bin liquid level, air cushion bin pressure, slurry bin pressure-left, slurry bin pressure-right and slurry bin pressure-up, slurry bin pressure-down, propulsion group A thrust, propulsion group B thrust, propulsion group C thrust, propulsion group D thrust, propulsion group E thrust, propulsion group F thrust, slurry inlet density, slurry outlet density, main slurry inlet flow, main slurry outlet flow, main drive backwater flow and external circulation water inlet flow, wherein the 22 factors are taken as input characteristic variables, and the output variables comprise tunneling speed, cutter head abrasion, overturning moment and super-digging flow four key slurry large shield construction performance indexes.
  3. 3. The method for online multi-objective optimization of shield performance based on BO-GRU-KAN-MOGWO of claim 2, wherein step S2 comprises the steps of: normalizing the input characteristic variables and the output variables in the original sample set, and mapping all variables to a [ -1,1] interval; Optimizing super parameters of the GRU-KAN model by adopting a Bayesian optimization algorithm, wherein the super parameters comprise five super parameters including time window size, learning rate, network layer number, hidden layer node number and dropout proportion, and optimizing, and carrying out model accuracy verification by combining a 5-fold cross verification method, so as to determine that the parameter combination with highest accuracy is the optimal parameter of the GRU-KAN prediction model; Dividing the preprocessed sample set into a training set and a testing set according to the ratio of 4:1, and constructing and training a BO-GRU-KAN prediction model; the accuracy of the BO-GRU-KAN prediction model is evaluated by adopting a determination coefficient, a root mean square error and an average absolute error; Introducing SHapley an addition interpretation method, calculating a Shapley value of each input characteristic variable on an output variable prediction result, quantifying the contribution degree and the contribution positive and negative of each input characteristic variable on tunneling speed, cutter head abrasion, overturning moment and overexcavation flow, and determining key influence parameters.
  4. 4. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO according to claim 3, wherein the interpretation function of the SHAP method is expressed as: (4) In the formula, The interpretation function is represented as a function of the interpretation, Representing the number of input features that are to be entered, Indicating whether each feature can be observed or not, Shapely values representing the virtual model without features, Represent the first Shapely values for the individual features; The Shapely value for each feature contribution is calculated by: (5) In the formula, The feature vector is represented by a vector of features, Representation of not including features Is provided for all of the feature subsets of (a), Is that The number of non-zero elements in the matrix, Is the set of all features in the model.
  5. 5. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO according to any one of claims 1 to 4, wherein step S3 comprises: Constructing a multi-objective optimization objective function by taking a tunneling speed, cutter head abrasion, overturning moment and overexcavation flow prediction function output by the BO-GRU-KAN prediction model as an adaptability function; combining engineering actual demands, industry standards and original sample set parameter distribution, defining a lower value limit for key influence parameters And upper limit of ; Adopting a population initialization strategy based on opposite learning, combining an optical lens imaging principle to optimize the initial population quality of a wolf optimization algorithm, and for each wolf individual Generating lens imaging reversal individuals thereof The method comprises the steps of adopting imaging inverse solutions of opposite learning computing optical systems to expand the search range of candidate solutions; Introduction of Dynamic adjustment of convergence factor for chaotic sequence Acquiring a Pareto optimal solution set; Selecting a unique optimal solution from the Pareto optimal solution set by adopting a TOPSIS method, calculating Euclidean distance from each Pareto solution to a positive ideal solution and a negative ideal solution, and selecting a solution closest to the positive ideal solution and furthest from the negative ideal solution as an optimal slurry shield construction parameter combination; And replacing the actual values of the tunneling speed, the cutter head abrasion, the overturning moment and the super-excavation flow performance indexes corresponding to the previous time step with the optimization results of the tunneling speed, the cutter head abrasion, the overturning moment and the super-excavation flow corresponding to the current time step, and taking the actual values as the initial state of the next time step optimization to realize the on-line dynamic optimization of the construction performance.
  6. 6. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO according to claim 5, wherein the multi-objective optimization objective function expression in step S3 is: (6) (7) Wherein, the Is a key influencing parameter; 、 、 、 the tunneling speed, the cutter head abrasion, the overturning moment and the overexcavation flow are respectively corresponding to each other; Represent the first Key influencing parameters; Is a historical sample set.
  7. 7. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO of claim 6, wherein the introducing Dynamic adjustment of convergence factor for chaotic sequence The method specifically comprises the following steps: By initialising parameters Generating Chaos sequence, based on the variable value of the current iteration, the variable value of the next iteration is generated, and the realization is realized Dynamic updating of the chaotic sequence; by normalizing the function to the original The chaotic sequence is subjected to standardized pretreatment to obtain a standardized function And based on standardized functions For original chaos sequence Adjusting to obtain standardized product Chaotic sequence value ; By standardizing the treated Chaotic sequence Embedding convergence factors In the calculation of (a), an improved convergence factor is obtained Values of (2) 。
  8. 8. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO of claim 7, wherein, The generalized equation for the distribution is as follows: (9) Wherein, the Probability density function representing beta distribution, output as variable Probability density values under the distribution; The probability density is a variable to be calculated (corresponding to the slurry shield construction parameter or a variable in an optimization algorithm in the invention); the shape of the distribution is determined for the shape parameter of the beta distribution, ; Is the upper and lower limits of the interval of the beta distribution, ; Is the center point of the beta distribution, ; The dynamic update of the chaotic sequence is represented by the formula (10): (10) Wherein, the Is a scale factor, which controls The amplitude of the chaotic sequence, Is the current number of iterations and, Is the first Updating the position vector for the second iteration; Is the first Primitive at multiple iterations A chaotic sequence; Is the first At the time of iteration, the variable Probability density values at the beta distribution; after normalization treatment Chaotic sequence value Represented by formula (12): (12) Wherein, the Is the first A normalization function at the time of iteration; Improving convergence factor Values of (2) Calculated by formula (13): (13) Wherein, the Is a conventional convergence factor.
  9. 9. The online multi-objective optimization method for shield performance based on BO-GRU-KAN-MOGWO according to any one of claims 6 to 8, wherein the optimal solution is selected from Pareto solutions set by TOPSIS method, and the core is scoring according to the distance between each solution and the ideal solution and the negative ideal solution, and the score of each solution is calculated by formula (14): (14) In the formula, Is the first The score corresponding to the pareto solution, Is the first The distance from the pareto solution to the negative ideal solution, Is the first The distance from pareto to the ideal; And The calculation formulas of (2) are shown as (15) and (16): (15) (16) In the formula, In order to optimize the number of targets, To at the first Under the object of The number of Pareto solutions is chosen, Is the first The average value at the individual target is, And To at the first Maximum and minimum values under the individual targets; And Calculated by formulas such as (17) and (18): (17) (18)。
  10. 10. the online multi-objective shield performance optimization system based on BO-GRU-KAN-MOGWO, which is used for realizing the online multi-objective shield performance optimization method based on BO-GRU-KAN-MOGWO as set forth in any one of claims 1 to 9, comprising: the first main module is used for selecting important slurry shield tunneling parameters and slurry system parameters as input characteristic variables, selecting slurry shield construction performance indexes as output variables, collecting actual measurement data of the input characteristic variables and the output variables in the slurry shield construction process, and constructing an original sample set; The second main module is used for optimizing the super parameters of the GRU-KAN algorithm by adopting Bayesian optimization, determining the optimal parameter combination by combining 5-fold cross validation, constructing a BO-GRU-KAN prediction model, predicting tunneling speed, cutter head abrasion, overturning moment and super-excavation flow, and obtaining four regression prediction functions; The third main module is used for constructing a multi-objective optimization objective function by taking four regression prediction functions output by the BO-GRU-KAN model as fitness functions, performing global optimization by adopting an improved multi-objective gray wolf optimization algorithm, determining a unique optimal slurry shield construction parameter combination from the obtained Pareto optimal solution set by adopting a TOPSIS method, and simultaneously replacing the current time step optimization result with the previous time step performance index actual value to realize online dynamic optimization.

Description

Online multi-objective optimization method and system for shield performance based on BO-GRU-KAN-MOGWO Technical Field The invention belongs to the technical field of shield tunnel construction performance optimization, and particularly relates to a shield performance online multi-objective optimization method and system based on BO-GRU-KAN-MOGWO. Background Along with the acceleration of the urban process and the rapid increase of the development requirements of underground space, the large-diameter slurry shield machine has become core construction equipment of major projects such as urban subway tunnels, river crossing and sea crossing tunnels and the like due to the advantages of large excavation section, uniform slurry pressure transmission, high excavation surface balance control precision, low cutter disc torque load and the like. However, slurry shield construction faces the double challenges of coupling complex geological conditions (such as uneven stratum with hardness and stratum with high water pressure) with multi-performance targets, reasonable configuration of slurry shield construction parameters directly determines engineering efficiency, safety and cost, and the optimization problem of the slurry shield construction is a key bottleneck for restricting intelligent development of slurry shield construction. Currently, the optimization of slurry shield construction parameters mainly has the following technical defects: The single-target optimization limitation is remarkable, the traditional method focuses on single performance indexes (such as only pursuing the improvement of the tunneling speed), the coupling restriction relation among multiple targets is ignored, for example, blind improvement of the tunneling speed easily causes the risks of cutter head abrasion aggravation, overrun of overturning moment or out-of-control of the over-tunneling flow, frequent cutter replacement, tunnel axis deviation, collapse even of the excavation surface and the like, and the comprehensive construction requirements of high efficiency, low consumption, safety, accuracy cannot be met. The accuracy and the interpretability of the prediction model are insufficient, the existing shield performance prediction multi-dependence empirical formula or a simple machine learning model (such as BP neural network and traditional GRU) is difficult to capture strong nonlinear and time-sequence correlation between slurry large shield construction parameters (such as total thrust, slurry bin pressure and propulsion grouping thrust) and performance indexes (such as tunneling speed, cutter head abrasion and the like), and meanwhile, the model lacks the interpretability, the contribution degree of each parameter to a performance target cannot be quantized, and the optimization result is difficult to guide on-site regulation. The optimization algorithm has the performance defects that the traditional multi-objective optimization algorithm (such as NSGA-II and basic gray wolf optimization algorithm) has the problems of low initial population quality, convergence factor rigidity change, imbalance of global search and local development and the like, is easy to fall into a local optimal solution, has poor uniformity and coverage of a Pareto optimal solution set, and is difficult to provide a stable and reliable parameter decision scheme. The existing optimization method is mostly based on offline data static calculation, the time sequence relevance of slurry shield construction is not considered, the construction state of the previous time step directly influences the tunnel stability of the next time step, and the static optimization result cannot adapt to the dynamic geological change in the construction process, so that parameter regulation and control lag and performance fluctuation are large in practical application. Therefore, aiming at the multi-objective collaborative optimization requirement of slurry large shield construction, development of a slurry large shield construction parameter optimization method with high-precision prediction, strong interpretability, high-efficiency optimizing capability and on-line dynamic adaptability is needed to realize collaborative control of tunneling speed, cutter head abrasion, overturning moment and overexcitation flow, and provide technical support for intelligent and accurate regulation and control of slurry large shield construction. Disclosure of Invention Aiming at the defects or improvement demands of the prior art, the invention provides a shield performance online multi-objective optimization method and system based on BO-GRU-KAN-MOGWO, which not only can realize high-precision prediction of construction performance indexes by means of a GRU-KAN model optimized by BO through a core innovation architecture of a BO-GRU-KAN prediction model, improved MOGWO optimization and online dynamic mechanism, but also can solve the problem that the traditional model is easy to sink locally