Deep-sea mining equipment design system based on improved genetic algorithm

Abstract

The invention relates to the technical field of deep-sea mining equipment, and provides a deep-sea mining equipment design system based on an improved genetic algorithm, which can improve the design efficiency and accuracy of deep-sea mining equipment, simplify the coupling strength between subjects, and improve the optimization efficiency while ensuring the accuracy. Comprising a data input layer, a subject decoupling layer, an optimization analysis layer, an intelligent solving layer and an operation interaction layer.

Description

Technical Field

[0001] This invention relates to the field of deep-sea mining equipment technology, specifically to a deep-sea mining equipment design system based on an improved genetic algorithm. Background Technology

[0002] The design of deep-sea mining equipment involves a complex interdisciplinary process involving structure, power, energy consumption, and environmental adaptability. Decoupling these disciplines is difficult, and a systematic theoretical framework and methodology for solving these problems are currently lacking. Existing designs, prioritizing safety, typically adhere to a conservative "functionality first" principle, resulting in a general lack of overall design considerations in deep-sea mining equipment, particularly in the design of deep-sea mining vehicles. This leads to high redundancy and defects in the deployed mining equipment, significantly increasing manufacturing and maintenance costs, impacting the economic viability of mining activities, and severely hindering the development of deep-sea mining.

[0003] Specifically, the following key issues exist: 1. The complex operating environment of the deep sea necessitates the design of mining equipment to consider numerous external factors. Furthermore, the complex structure and materials of mining equipment ultimately lead to the need to input a vast number of constant parameters to construct the underlying environmental baseline in multidisciplinary design. On the one hand, the deep-sea environment, especially the seabed, has complex sediment and seawater characteristics, making it difficult to obtain constant parameters. On the other hand, the variety of constant parameters, the lack of standardized templates and automated verification mechanisms, and the inconsistencies in cross-disciplinary data semantics can easily cause model conflicts, hindering collaborative transmission between various design software and severely reducing design efficiency.

[0004] 2. The design of deep-sea mining equipment involves complex multidisciplinary coupling. Existing tools rely on fixed decoupling strategies and lack quantitative analysis of the dynamic coupling strength between disciplines, making it difficult to adjust decoupling priorities. The complex coupling relationships directly lead to long solution times and increased computational costs for the established models.

[0005] 3. In the design of deep-sea mining equipment, different systems and disciplines have different requirements for the accuracy of the established models. If low-precision processing is uniformly adopted, the design quality will be affected. If high-precision simulation is adopted for all, the workload of solving the problem will be significantly increased. At present, there is a lack of optimization strategies that combine high-precision simulation with low-precision proxy models. Existing technologies cannot verify the convergence and accuracy of the model from a global perspective, which affects the credibility of the optimization results.

[0006] 4. Existing optimization algorithms are mainly NSGA-II and NCGA. In deep-sea equipment design, NSGA-II, in high-dimensional multi-objective (4 or more) optimization, may suffer from a lack of global diversity due to the local density of adjacent solutions. As the number of objectives increases, the computational complexity grows exponentially, leading to uneven distribution of the solution set and a tendency to get trapped in local optima. NCGA requires preset niche radius or density thresholds, and parameter selection has a significant impact on the results. Furthermore, it is difficult to effectively delineate niche regions in high-dimensional problems. Both of these algorithms have limited search capabilities in high-dimensional problems, requiring multiple iterations to converge, resulting in high computational costs. In addition, their dynamic constraint adjustment capabilities are insufficient, making it difficult to balance convergence speed and diversity, especially in scenarios with strong multidisciplinary coupling, which can easily lead to iterative oscillations or non-convergence. Summary of the Invention

[0007] One of the objectives of this invention is to propose a deep-sea mining equipment design system based on an improved genetic algorithm, which can improve the design efficiency and accuracy of deep-sea mining equipment, simplify the coupling between disciplines, and improve optimization efficiency while ensuring accuracy.

[0008] The technical solution of the present invention is as follows: A deep-sea mining equipment design system based on an improved genetic algorithm includes: Data Input Layer: Used to configure templated parameter ports, allowing users to define parameterized input templates and set input constants, input variables, constraints, and optimization objectives; Subject decoupling layer: includes subject establishment module and coupling analysis module, responsible for the integration and coupling of various subject models; Optimization and analysis layer: This includes the experimental design module and the data modeling module, which are responsible for executing experimental design and optimization algorithms and supporting user modeling decisions; Intelligent solution layer: Adopts a multi-stage optimization principle, optimizes the objectives in stages, first optimizes a single objective and then expands to multiple objectives; designs optimization strategies for the surrogate model, can automatically perform experimental design sampling and update the surrogate model, and automatically call the quasi-Newton method (L-BFGS), SPEA2, and improved NSGA-III optimization algorithms. Operational Interaction Layer: Utilizing PySide interface design technology to provide UI controls and functions, ensuring efficient operation through multidisciplinary design optimization. It efficiently manages the thread pool using QThreadPool and QRunnable technologies, introduces a cloud data storage module for parallel processing to avoid frequent thread creation and destruction, and supports collaborative editing and conflict detection by experts in mechanical, control, and environmental fields.

[0009] Furthermore, the data input layer sets the main input constant parameter and the secondary input constant parameter for the input constants; The input variables are divided into structural design variables, power and energy design variables, and support system variables. The structural design variables are divided into the length, width, height, external features, topology, local structural degrees of freedom, and strength redundancy of the connecting parts of the mining equipment. The power and energy design variables are divided into propulsion device power and layout, battery capacity, and energy distribution strategy. The supporting system variables are divided into umbilical cable armor thickness and maximum load of deployment and recovery device. The constraints are categorized into behavioral constraints, intensity constraints, and discipline consistency constraints. The optimization objectives include the weight and distribution of deep-sea mining equipment, travel-recovery resistance, and operational energy consumption.

[0010] Furthermore, the data input layer includes the following operational steps: S110: The main input constant parameters are entered by the designer. S120: Establish a deep-sea environment database, input secondary input constant parameters, and based on the deep-sea environment function of the polymetallic nodule mining area, use the main input constant parameters as independent variables to complete the automatic prediction and input of multi-dimensional characteristic parameters of seawater, sediment, and nodules at different depths in the collection area. S130: Combine machine learning with embedded automated verification rules to perform sub-constant range checks and logical consistency verification.

[0011] Furthermore, step S130 includes: S131: After the user inputs the main constant parameters, the pre-trained random forest model is called to predict the validity probability p of the combination of input constant parameters; if the validity probability p ≥ 0.9 threshold, the test is passed. If the legality probability p < 0.9 threshold, a range check rule base comparison is triggered, indicating that the limit is exceeded or the parameters do not match, and a recalculation or manual review process is triggered. S132: After the environment function automatically predicts the input constant, the outlier score q is calculated using the isolated forest model; if the outlier score q ≤ 0.6, the parameters are considered reasonable. If the anomaly score q>0.6, initiate logical consistency verification, check parameter correlation, generate a visual report for complex anomalies, highlight contradictory parameters and recommend correction suggestions, and realize parameter visualization manual verification and error correction; S133: The correlation characteristics are as follows: hydrostatic pressure has a positive correlation with seawater depth, with an approved pressure-to-depth ratio of 0.01 MPa / m; seabed temperature is adjusted to 5℃-25℃ from the surface to 2000m, based on the marine environment and air temperature of the mining area; 2℃-4℃ from 2000-3500m; 1℃-2℃ from 3500-4000m; and slightly higher from 4000-6000m due to the adiabatic self-pressure effect of seawater, adjusted according to the statistical conditions of seabed hydrothermal vents in the mining area. The salinity of seawater tends to be uniform after exceeding the halocline, at approximately 34‰-35‰; Seawater density increases with depth, reaching 1025-1030 kg / m³ at depths of 4000-6000 m. Seawater viscosity increases with decreasing temperature and slightly with increasing salinity, reaching 1.37-1.69 × 10⁻³ Pa·s at depths of 4000-6000 m. The ocean current velocity at depths of 4000-6000m is affected by density differences and geostrophic currents, ranging from 3cm / s to 30cm / s.

[0012] Furthermore, the subject decoupling layer includes the following operational steps: S210: The main body of deep-sea mining equipment is divided into structural discipline, configuration discipline, propulsion discipline, and energy consumption discipline through the discipline establishment module; S220: By analyzing the characteristics of each component, the subject-based module determines the data flow of constants and input variables that it transmits to each subject and system layer. S230: Introduces the Sobol index and correlation coefficient matrix into the coupling analysis module to quantify the coupling strength between disciplines. A Sobol index < 0.2 is defined as a weakly coupled discipline, a Sobol index of 0.2 to 0.4 is defined as a moderately coupled discipline, and a Sobol index > 0.4 is defined as a strongly coupled discipline. A correlation coefficient |r| > 0.5 between input variables is considered a strong correlation, while |r| < 0.5 is considered a weak correlation. Prioritize disciplines with high Sobol indices or strong correlations, and establish a proxy model of discipline coupling structure based on sensitivity parameters to guide decoupling priorities; Jacobi iterative parallel execution is used for weakly coupled disciplines, while serial coordination is used for strongly coupled disciplines.

[0013] Furthermore, step S220 includes: S221: In the design, the structural discipline transmits the equipment geometry and topological constraints to the configuration discipline with a transmission factor of 0.65-0.85; transmits the structural weight and distribution to the propulsion discipline with a transmission factor of 0.85-0.95; and transmits the weight and distribution to the energy consumption discipline with a transmission factor of 0.85-0.95. S222: Configuration discipline transmits layout feasibility feedback to structural discipline, with a transmission factor set at 0.55-0.65; it transmits fluid dynamics shape parameters and propulsion benefits to propulsion discipline, with a transmission factor set at 0.45-0.55; and it transmits equipment spatial layout and local energy distribution strategies to energy consumption discipline, with a transmission factor set at 0.75-0.90. S223: The propulsion discipline transmits the structural strength requirements of the propulsion installation location to the structural discipline, with a transmission factor set to 0.68-0.78; transmits the space occupancy constraints to the configuration discipline, with a transmission factor set to 0.85-0.95; and transmits the real-time power demand curve and power consumption characteristics to the energy consumption discipline, with a transmission factor set to 0.8-0.9. S224: The energy consumption discipline transmits the sensitivity of energy consumption to weight distribution to the structural discipline, with a transmission factor set at 0.65-0.75; transmits the layout of cable power supply equipment and energy consumption equipment to the configuration discipline, with a transmission factor set at 0.35-0.65; and transmits the maximum allowable power threshold under energy consumption constraints to the propulsion discipline, with a transmission factor set at 0.85-0.95.

[0014] Furthermore, the optimization analysis layer supports the integration of process models built by Isight, ModelCenter, Tosca, and Optistruct; It also includes a text parser, a system command executor, a calculator, and dedicated interface components for finite element software. The experimental design module mainly conducts structural mechanics tests, corrosion resistance assessment tests, fatigue tests, hydrodynamic tests, and flow field simulation and analysis tests for deep-sea equipment. Different experimental design strategies are adopted based on the number of optimization objectives and their coupling relationships. When the number of optimization objectives n < 3, and the objectives are not in the same discipline or the mutual influence coefficient c < 0.3, a full factorial design test is used. When the number of optimization objectives n < 3, and the objectives are in the same discipline or the mutual influence coefficient c > 0.3, an orthogonal low-workload design test is used. When the number of optimization objectives n > 3, a Latin hypercube or central composite design test is used. The data modeling module is equipped with a parametric integration interface for commercial software or self-developed programs for multidisciplinary design simulation, including SolidWorks, ANSYS, LongRuan4D-GIS, OpenFOAM, digital twin interactive control platform, and Simulink; enabling parametric integration and automatic calling of self-developed programs and commercial CAD / CAE programs. Furthermore, the data modeling module provides a multinomial response surface model algorithm and also has a surrogate model algorithm interface. Based on the interface specification, external surrogate model algorithms can be embedded into this platform. A multi-fidelity optimization strategy is adopted, which mixes high-precision simulation and low-precision surrogate models. High-precision simulation prediction accounts for 65%-75% of the global target weight, while low-precision surrogate model prediction accounts for 25%-35% of the global target weight. When the deviation d between the low-precision surrogate model and the high-precision simulation result exceeds 12%, the weight of the low-precision surrogate model is reduced to 0.1. Based on the sensitivity of each system unit to the discipline, high-precision optimization is performed only on highly sensitive parameters. After every 5 local optimizations, a high-precision full-system simulation is performed. The overall convergence criteria are that the rate of change of the objective function is less than 1% for 3 consecutive iterations and the cross-disciplinary constraint violation is less than 5%.

[0015] Furthermore, the multidisciplinary design in the data modeling module includes: In structural engineering, high-precision simulations should be prioritized for detailed stress, strain, and fatigue analysis in three-dimensional geometry, as well as for material plasticity, creep, anisotropy, fluid loads, and structural responses. Low-precision surrogate models should be prioritized for rapid static analysis of rod and shell elements. The propulsion discipline conducts high-precision simulation modeling of propeller-assisted propulsion devices, and a low-precision surrogate model is used for the preliminary relationship between propulsion power and propulsion speed; The dynamic energy consumption simulation during the start-up / impact load change process of the energy consumption discipline adopts high-precision simulation, while the preliminary prediction of energy consumption per unit mining volume adopts a low-precision surrogate model. In the initial optimization stages of overall dimensions, layout, and center of gravity, the configuration discipline uses a low-precision surrogate model, while high-precision simulation is used for configuration driving-recovery resistance analysis.

[0016] Furthermore, the improved NSGA-III optimization algorithm in the intelligent solution layer includes the following steps: S310: Initialize the population: Based on historical data and the set of best points, generate an initial near-optimal diversity population using chaotic logistic mapping; S320: Assess fitness: Assess each individual and calculate its fitness value under multiple optimization objectives; S330: Fast Non-Dominated Sort: Store the sets of dominant and dominated individuals, sort all individuals in the set of dominated individuals, and if an individual has zero dominant individuals, it is considered a non-dominated individual, and its Pareto level is set to the current highest level plus one. S340: Crowding Calculation: In order to maintain population diversity, the crowding of each individual is calculated for subsequent elite selection; S350: Generation of offspring: Parameters are dynamically adjusted according to population diversity, and the next generation of population is generated through genetic operations. At the same time, 5% to 10% of elite individuals are retained to ensure population diversity and convergence. The generated offspring individuals are subjected to local perturbation and acceptance criteria judgment in simulated annealing. In the selection phase, the acceptance probability of simulated annealing is used to replace roulette wheel selection. The initial temperature of simulated annealing is set to 100 to 300, and the temperature decay coefficient is set to 0.90 to 0.99. S360: Merge Populations: Merge the parent and offspring populations to form a new population; S370: Elite Selection: Divide the merged and mixed population into 3 to 6 subgroups, select individuals with high fitness and low crowding as elite individuals for the next generation population, prioritize the retention of feasible solutions, and gradually relax the constraint tolerance using a dynamic constraint method to complete the parallel distributed optimization and then merge them. S380: Repeated Iteration: Repeat the above steps until the set stopping condition is met; S390: Obtaining Pareto optimal solutions: The individuals in the final population are the Pareto optimal solution set, where each solution cannot be completely dominated by other solutions under all objectives.

[0017] Furthermore, the data input layer adopts HDFS and MinIO as distributed storage strategies and forms, and utilizes Memcached in-memory database and efficient indexing strategies to establish a Spark computing framework; The data input layer dynamically generates data structures, adopts JSON general format, text format, and ontology to define data semantics and formulate interdisciplinary data standards; it completes data classification by recording data source, type, and unit to form a data ownership list and establishes a metadata framework. The data input layer sets multi-weight and gradient constraints in the constraints, and adopts a multi-level constraint restriction strategy with priority safety > environment > cost > energy consumption optimization. The coupling analysis module of the subject decoupling layer is equipped with a random forest classification model, which is trained with historical optimization data to identify coupling parameter combinations that are prone to oscillation or non-convergence in advance. It also provides a real-time monitoring panel to display coupling residuals, iteration counts, subject model call status, and built-in benchmark cases of typical coupling problems for verifying decoupling algorithms and performance tuning. The model of the data modeling module in the optimization analysis layer is set according to the complexity and nonlinearity of the experimental data; The intelligent solution layer uses a quasi-Newton algorithm to solve single-objective optimization, and the SPEA2 algorithm to optimize multi-objective optimization models with 2 to 3 objectives. When the number of objectives exceeds 3, the improved NSGA-III algorithm is introduced. The interactive interface is divided into three interfaces: design, monitoring, and output. The design interface includes a categorized tree-structured component library, a process structure tree, a global / local variable management area, a parameter mapping relationship management area, a priority setting area, and a resource quota selection area. The monitoring interface includes a task timeline view area and a progress monitoring bar. The output interface includes a multi-objective optimization result chart area, a data sandbox area, and a report generation area for key parameter impact and optimal solution recommendation.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention establishes an environmental database at the data input layer and performs automatic prediction of secondary constant parameters and data classification to ultimately improve the design efficiency and accuracy of deep-sea mining equipment. At the discipline decoupling layer, a coupling proxy model based on discipline sensitivity is set up to simplify the coupling strength between disciplines and avoid the design process from becoming bogged down in complex decoupling processes. At the optimization analysis layer, multi-simulation optimization is implemented, using a combination of high-precision simulation and low-precision proxy models to improve optimization efficiency while ensuring accuracy. At the intelligent solution layer, different algorithms are used to solve problems of varying complexity. Furthermore, the NSGA-III genetic algorithm is improved for the multidisciplinary model of deep-sea mining equipment to avoid getting trapped in local optima when solving complex models. Attached Figure Description

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0021] Figure 1 This is a general framework diagram of the present invention; Figure 2 This is a diagram of the data input layer structure of the present invention; Figure 3 This is a structural diagram of the subject decoupling layer of the present invention; Figure 4 This is a structural diagram of the optimization analysis layer of the present invention; Figure 5 This is a structural diagram of the intelligent solution layer of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1

[0024] like Figure 1 As shown, this embodiment provides a deep-sea mining equipment design system based on an improved genetic algorithm. It is mainly used for designing deep-sea mining vehicles and their supporting operational systems. The system is written in Python, with front-end application development using frameworks such as Vue, and the interface and business logic built using Javascript. It includes five major frameworks: a data input layer, a subject decoupling layer, an optimization analysis layer, an intelligent solution layer, and an operation interaction layer.

[0025] Specifically, the data input layer is used to configure templated parameter ports, allowing users to define parameterized input templates and set input constants, input variables, constraints, and optimization objectives. The data input layer adopts HDFS and MinIO as distributed storage strategies and forms, utilizes Memcached in-memory database and efficient indexing strategies to accelerate high-frequency data access, establishes a Spark computing framework, and improves large-scale data processing capabilities. The data input layer dynamically generates data structures to adapt to different optimization scenarios. It adopts JSON universal format, text format, and ontology to define data semantics and formulate interdisciplinary data standards to ensure interoperability. It completes data classification by recording data source, type, and unit to form a data ownership list, establishes a metadata framework, and enhances traceability and consistency. The data input layer sets multi-weight and gradient constraints in the constraints, and adopts a multi-level constraint restriction strategy with priority safety > environment > cost > energy consumption optimization to ensure that multi-disciplinary joint simulation obtains the target optimization results with trade-offs.

[0026] The aforementioned data input layer sets primary and secondary input constant parameters; it categorizes input variables into structural design variables, power and energy design variables, and support system variables; structural design variables are further divided into the length, width, height, shape characteristics, topology, local structural degrees of freedom, and connector strength redundancy of mining equipment (deep-sea mining vehicles, umbilical cables, ROVs, AUVs); power and energy design variables are categorized into propulsion device power and layout, battery capacity, and energy distribution strategies (operation mode and standby energy consumption ratio); support system variables are categorized into umbilical cable armor thickness and maximum load of deployment and recovery devices; constraints are categorized into behavioral constraints (travel-recovery resistance, thrust, operational energy consumption (electricity consumption), weight and weight distribution, reserve buoyancy, envelope volume), strength constraints (strength and stability, stress), and discipline consistency constraints; optimization objectives include the weight and distribution of deep-sea mining equipment, travel-recovery resistance, operational energy consumption (electricity consumption), and corresponding economic optimizations (high-strength material cost, manufacturing complexity).

[0027] The data input layer includes the following steps: S110: The main input constant parameters are entered by the designer, including seawater depth, sediment cross shear strength, structural material type, corrosion resistance requirements, and redundancy design standards. S120: Establish a deep-sea environment database, input secondary constant parameters, and based on the deep-sea environment function of the polymetallic nodule mining area, use the main input constant parameters as independent variables to complete the automatic prediction and input of multi-dimensional characteristic parameters of seawater, sediment, and nodules at different depths in the collection area. Among them, the automatically predicted secondary constants include environmental constants (hydrostatic pressure, seabed temperature, seawater salinity, seawater density and viscosity, ocean current velocity) and sediment characteristic constants (shear strength, cohesion, internal friction angle). S130: Combine machine learning with embedded automated verification rules to perform sub-constant range checks and logical consistency verification.

[0028] Specifically, step S130 includes: S131: After the user inputs the main constant parameters, the pre-trained random forest model is called to predict the validity probability p of the combination of input constant parameters; if the validity probability p ≥ 0.9 threshold, the test is passed. If the legality probability p < 0.9 threshold, a range check rule base comparison is triggered, indicating that the limit is exceeded or the parameters do not match, and a recalculation or manual review process is triggered. S132: After the environment function automatically predicts the secondary input constant, the anomaly score q is calculated using the isolated forest model. q is based on the average path length of the sample across all isolated trees. ; in, :sample The path length (number of splits) in an isolated tree; :sample Average path length across all trees; Sample size is The expected average path length at that time (which can be regarded as a normalized constant); The calculation formula is: ; for The harmonic mean, (Euler-Mascheroni constant); Since it is necessary to cover 95% of historical anomalies, the parameters are considered reasonable if the anomaly score q ≤ 0.6. If the anomaly score q>0.6, initiate logical consistency verification, check parameter correlation, generate a visual report for complex anomalies, highlight contradictory parameters and recommend correction suggestions, and realize parameter visualization manual verification and error correction; S133: The correlation characteristics are as follows: hydrostatic pressure has a positive correlation with seawater depth, with an approved pressure-to-depth ratio of 0.01 MPa / m; seabed temperature is adjusted to 5℃-25℃ from the surface to 2000m, based on the marine environment and air temperature of the mining area; 2℃-4℃ from 2000-3500m; 1℃-2℃ from 3500-4000m; and slightly higher from 4000-6000m due to the adiabatic self-pressure effect of seawater, adjusted according to the statistical conditions of seabed hydrothermal vents in the mining area. Seawater salinity tends to be uniform beyond the halocline (usually 100-300m), at approximately 34‰-35‰. Seawater density increases with depth, reaching approximately 1025-1030 kg / m³ at depths of 4000-6000 m. Seawater viscosity increases with decreasing temperature and slightly with increasing salinity, reaching approximately 1.37-1.69 × 10⁻³ Pa·s at depths of 4000-6000 m. The velocity of ocean currents at depths of 4000-6000m is affected by density differences and geostrophic currents, and is generally between a few centimeters per second and tens of centimeters per second, that is, between 3cm / s and 30cm / s.

[0029] Subject decoupling layer: includes subject establishment module and coupling analysis module, responsible for the integration and coupling of various subject models.

[0030] The coupling analysis module includes a random forest classification model, which is trained using historical optimization data to identify coupling parameter combinations that are prone to oscillation or non-convergence in advance. It also provides a real-time monitoring panel to display coupling residuals, iteration counts, and subject model call status. It includes a benchmark case of a typical coupling problem (fluid dynamics-structure coupling) to verify the decoupling algorithm and performance tuning. The aforementioned subject decoupling layer includes the following operational steps: S210: The main body of deep-sea mining equipment is divided into structural discipline, configuration discipline, propulsion discipline, and energy consumption discipline through the discipline establishment module; S220: By analyzing the characteristics of each component, the subject-based module determines the data flow of constants and input variables that it transmits to each subject and system layer. Step S220 specifically includes: S221: In the design, the structural discipline transmits the equipment geometry and topological constraints (maximum envelope volume) to the configuration discipline, with a transmission factor set to 0.65-0.85 (0.65 in this embodiment); transmits the structural weight and distribution to the propulsion discipline, with a transmission factor set to 0.85-0.95 (0.9 in this embodiment); and transmits the weight and distribution to the energy consumption discipline, with a transmission factor set to 0.85-0.95 (0.85 in this embodiment). S222: The configuration discipline transmits layout feasibility feedback to the structure discipline, with a transmission factor set to 0.55-0.65 (0.45 in this embodiment); it transmits fluid dynamics shape parameters and propulsion benefits to the propulsion discipline, with a transmission factor set to 0.45-0.55 (0.5 in this embodiment); and it transmits equipment spatial layout and local energy distribution strategies to the energy consumption discipline, with a transmission factor set to 0.75-0.90 (0.9 in this embodiment). S223: The propulsion discipline transmits the structural strength requirements of the propeller installation location to the structural discipline, with a transmission factor set to 0.68-0.78 (0.7 in this embodiment); it transmits the space occupancy constraints to the configuration discipline, with a transmission factor set to 0.85-0.95 (0.85 in this embodiment); and it transmits the real-time power demand curve and power consumption characteristics to the energy consumption discipline, with a transmission factor set to 0.8-0.9 (0.9 in this embodiment). S224: The energy consumption discipline transmits the sensitivity of energy consumption to weight distribution to the structural discipline, with a transmission factor set to 0.65-0.75 (0.7 in this embodiment). It transmits the layout of cable power supply equipment and energy consumption equipment to the configuration discipline, with a transmission factor set to 0.35-0.65 (0.6 in this embodiment). It transmits the maximum allowable power threshold under energy consumption constraints to the propulsion discipline, with a transmission factor set to 0.85-0.95 (0.9 in this embodiment).

[0031] S230: Introduces the Sobol index and correlation coefficient matrix into the coupling analysis module to quantify the coupling strength between disciplines. A Sobol index < 0.2 is defined as a weakly coupled discipline, a Sobol index of 0.2 to 0.4 is defined as a moderately coupled discipline, and a Sobol index > 0.4 is defined as a strongly coupled discipline. A correlation coefficient |r| > 0.5 between input variables is considered a strong correlation, while |r| < 0.5 is considered a weak correlation. Prioritize disciplines with high Sobol indices (>0.3) or strong correlations (|r|>0.5), and establish a surrogate model of discipline coupling structure based on sensitivity parameters to guide decoupling priorities; Jacobi iterative parallel execution is used for weakly coupled disciplines, while serial coordination is used for strongly coupled disciplines.

[0032] Optimization and Analysis Layer: This layer includes the experimental design module and the data modeling module, which are responsible for executing experimental design and optimization algorithms and supporting user modeling decisions.

[0033] The aforementioned optimization analysis layer supports the integration of process models built with Isight, ModelCenter, Tosca, and Optistruct, helping customers migrate models from legacy multidisciplinary optimization platforms. It also includes a text parser, a system command executor, a calculator, and dedicated interface components for finite element software. The aforementioned experimental design modules mainly conduct structural mechanics tests, corrosion resistance assessment tests, fatigue tests, hydrodynamic tests, and flow field simulation and analysis tests for deep-sea equipment. Different experimental design strategies are adopted according to different numbers of optimization objectives and coupling relationships. When the number of optimization objectives n < 3, and the optimization objectives are not located in the same discipline or the mutual influence coefficient c < 0.3, a full factorial design test is adopted. When the number of optimization objectives n < 3, and the optimization objectives are located in the same discipline or the mutual influence coefficient c > 0.3, an orthogonal low-workload design test is adopted. When the number of optimization objectives n > 3, a Latin hypercube or central composite design test is adopted.

[0034] The aforementioned data modeling module is equipped with parametric integration interfaces for commercial software or self-developed programs for multidisciplinary design simulation, including SolidWorks, ANSYS (for strength and stability simulation of mining equipment shells and frames, umbilical cable connection components), LongRuan4D-GIS (for dynamically integrating seawater parameters and seabed sediment characteristics to establish an environment-equipment interactive model), OpenFOAM (for simulating fluid resistance and propulsion efficiency of deep-sea mining vehicles, ROVs / AUVs in complex ocean currents to optimize shape design and thruster layout), a digital twin interactive control platform (for real-time simulation of deep-sea mining equipment operation status, supporting multi-equipment collaborative operation and emergency obstacle avoidance), and Simulink (for optimizing battery capacity and energy distribution strategies); enabling parametric integration and automatic invocation of self-developed programs and commercial CAD / CAE programs; Furthermore, the data modeling module provides a multinomial response surface model algorithm and also has a surrogate model algorithm interface. Based on the interface specification, external surrogate model algorithms can be embedded into this platform. A multi-fidelity optimization strategy is adopted, which mixes high-precision simulation (error <5%) and low-precision surrogate model (error <20%). High-precision simulation prediction accounts for 65%-75% of the global target weight (75% in this embodiment), and low-precision surrogate model prediction accounts for 25%-35% of the global target weight (20% in this embodiment). When the deviation d between the low-precision surrogate model and the high-precision simulation result exceeds 12%, the weight of the low-precision surrogate model is reduced to 0.1.

[0035] Based on the sensitivity of each system unit to the discipline, high-precision optimization is only performed on highly sensitive parameters (>0.5). After every 5 local optimizations, a high-precision full system simulation is performed. The overall convergence criteria are that the rate of change of the objective function is <1% for 3 consecutive iterations and the cross-disciplinary constraint violation is <5%.

[0036] The data modeling module sets the model according to the complexity and nonlinearity of the experimental data; linear polynomial response surface model is applicable when the variable dimension is ≤5 or the Sobol interaction index is <0.05 or the linear misfit test is ≥0.1; quadratic polynomial response surface model is applicable when 5 < variable dimension is ≤15 or the Sobol interaction index is ≥0.1 or 0.05 ≤ linear misfit test is <0.3; higher-order polynomial response surface model is applicable when the variable dimension is >15 or the Sobol interaction index is ≥0.3 or the linear misfit test is <0.05.

[0037] Furthermore, the multidisciplinary design in the aforementioned data modeling module includes: In structural engineering, high-precision simulations should be prioritized for detailed stress, strain, and fatigue analysis in three-dimensional geometry, as well as for material plasticity, creep, anisotropy, fluid loads, and structural responses. Low-precision surrogate models should be prioritized for rapid static analysis of rod and shell elements. The propulsion discipline conducts high-precision simulation modeling of propeller-assisted propulsion devices, and a low-precision surrogate model is used for the preliminary relationship between propulsion power and propulsion speed; The dynamic energy consumption simulation during the start-up / impact load change process of the energy consumption discipline adopts high-precision simulation, while the preliminary prediction of energy consumption per unit mining volume adopts a low-precision surrogate model. In the initial optimization stages of overall dimensions, layout, and center of gravity, the configuration discipline uses a low-precision surrogate model, while high-precision simulation is used for configuration driving-recovery resistance analysis.

[0038] Intelligent solution layer: Adopts a multi-stage optimization principle, optimizes the objectives in stages, first optimizes a single objective and then expands to multiple objectives; designs optimization strategies for the surrogate model, can automatically perform experimental design sampling and update the surrogate model, and automatically call the quasi-Newton method (L-BFGS), SPEA2 (Intensity Pareto Evolutionary Algorithm 2), and improved NSGA-III optimization algorithm.

[0039] Specifically, the aforementioned quasi-Newton method (L-BFGS) algorithm is used to solve single-objective optimization problems, the SPEA2 algorithm is set to optimize multi-objective optimization models with 2 to 3 objectives, and when the number of objectives exceeds 3, the improved NSGA-III algorithm is introduced.

[0040] Specifically, the improved NSGA-III optimization algorithm in the aforementioned intelligent solution layer includes the following steps: S310: Initialize the population: Based on historical data and the set of best points, generate an initial near-optimal diversity population using a chaotic Logistic mapping (the chaos factor of the Logistic mapping is designed to be 3.57~4.0, and is designed to be 4.0 in this embodiment), including 80~120 individuals (120 individuals in this embodiment). Each individual is composed of input variables; in addition, it can also be the cross-sectional area, thickness, length, width, height, support structure form and frame size of deep-sea equipment components in the structural science, the installation position, connection method and relative positional relationship of functional components on deep-sea equipment in the configuration science, the installation position and direction of propulsion devices on deep-sea equipment in the propulsion science, and the transmission ratio, efficiency and energy distribution method in the energy consumption science; use a neural network to predict the quality of individuals, and remove individuals with abnormal quality from the set of best points; S320: Fitness Evaluation: Evaluate each individual and calculate its fitness value under multiple optimization objectives. The fitness value can be calculated using the objective function; repeatedly called objective functions are cached to avoid redundant calculations. S330: Fast Non-Dominated Sort: Store the sets of dominant and dominated individuals, sort all individuals in the set of dominated individuals, and if an individual has zero dominant individuals, it is considered a non-dominated individual, and its Pareto level is set to the current highest level plus one; based on the above fast non-dominated sorting logic, non-dominated sorting is performed on the individuals in the population, dividing the individuals into different Pareto front levels, ensuring that each individual has the opportunity to become the Pareto optimal solution; S340: Crowding Calculation: In order to maintain population diversity, the crowding of each individual is calculated for subsequent elite selection; S350: Generation of offspring: Based on the dynamic adjustment of parameters according to population diversity, the next generation of the population is generated through genetic operations (crossover probability is set to 0.85~0.95, 0.9 in this embodiment; mutation probability is set to 1 / dimension~3 / dimension, 2 / dimension in this embodiment; high dimensions can be uniformly set to 0.02~0.05, 0.02 in this embodiment). At the same time, 5%~10% (10% in this embodiment) of elite individuals are retained to ensure population diversity and convergence. The generated offspring individuals are subjected to simulated annealing local perturbation and acceptance criteria judgment. In the selection stage, the acceptance probability of simulated annealing is used instead of roulette wheel selection. The initial temperature of simulated annealing is set to 100~300, 100 in this embodiment, and the temperature decay coefficient is set to 0.90~0.99, preferably 0.90. Alternatively, the population can be divided into a genetic subpopulation and a particle swarm (PSO). The genetic subpopulation undergoes crossover and mutation, while the PSO subpopulation is updated according to velocity. The inertia weight of the particle swarm is set to 0.4~0.9, and in this embodiment, it is set to 0.5-0.8, with dynamic adjustment. When the variable dimension is less than 80, it is set to 0.4~0.6; in this example, it is set to 0.5. When the variable dimension is 80~100, it is set to 0.6~0.8; in this embodiment, it is set to 0.7. When the variable dimension is 100~120, it is set to 0.8~0.9. In this example, it is set to 0.8, and the two are periodically exchanged for excellent individuals; Alternatively, differential evolution strategies such as "DE / rand / 1" or "DE / best / 1" can be used to generate mutated individuals, with the differential mutation coefficient set to 0.4~0.8 (preferably 0.5) to enhance local development capabilities; S360: Merge Population: Merge the parent and child populations to form a new population. At this time, only the non-dominance relationship is updated for newly added individuals, rather than a global reordering, reducing computational redundancy. S370: Elite Selection: Divide the merged and mixed population into 3 to 6 subgroups, and select individuals with high fitness and low crowding as elite individuals for the next generation population. Prioritize retaining feasible solutions, and gradually relax the constraint tolerance using a dynamic constraint method. The initial value of the dynamic constraint tolerance is 0.01, and it is increased by 0.005 to 0.01 each time. In this embodiment, it is increased by 0.005. After completing the parallel distributed optimization, merge the subgroups. S380: Repeated Iteration: Repeat the above steps until the set stopping condition is reached (such as reaching the maximum number of iterations, which is set to 100~300, and in this embodiment to 300 (set to 100~150 when the variable dimension is less than 80, set to 150~200 when the variable dimension is 80~100, and set to 200~300 when the variable dimension is 100~120), with convergence no improvement algebra ≥10). S390: Obtain Pareto optimal solutions (also known as Pareto efficiency): The individuals in the final population are the Pareto optimal solution set, where each solution cannot be completely dominated by other solutions under all objectives.

[0041] Through the above steps, the improved NSGA-III can effectively search for Pareto optimal solutions to multi-objective optimization problems, providing designers with diverse and balanced solutions and helping them make better decisions during the design process.

[0042] Operational Interaction Layer: Utilizing PySide interface design technology to provide UI controls and functions, ensuring efficient operation through multidisciplinary design optimization. It efficiently manages the thread pool using QThreadPool and QRunnable technologies, introduces a cloud data storage module for parallel processing to avoid frequent thread creation and destruction, and supports collaborative editing and conflict detection by experts in multiple fields such as mechanics, control, and environment.

[0043] Meanwhile, the operation interaction layer's working interface is divided into three interfaces: design, monitoring, and output. The design interface includes a categorized tree structure component library, a process structure tree, a global / local variable management area, a parameter mapping relationship management area, a priority setting area, and a resource quota selection area. The monitoring interface includes a task timeline view area and a progress monitoring bar. The output interface includes a multi-objective optimization result chart area, a data sandbox area, and a report generation area for key parameter impact and optimal solution recommendation.

[0044] Table 1 shows the technical specifications of the final deep-sea mining vehicle design based on Example 1 of this invention:

[0045] Example 2: This embodiment is specifically designed for deep-sea mining vehicles, and differs from Embodiment 1 in that: 1. In the steps of the discipline coupling layer, S221: the transfer factor is set to 0.85 to transfer the structural weight and distribution to the propulsion discipline, the transfer factor is set to 0.95 to transfer the weight and distribution to the energy consumption discipline, and the transfer factor is set to 0.95. In S222, the transmission factor is set to 0.55 to transmit fluid dynamics shape parameters and propulsion benefits to the propulsion discipline, the transmission factor is set to 0.5 to transmit equipment spatial layout and local energy distribution strategies to the energy consumption discipline, and the transmission factor is set to 0.8. In S223, the transfer factor is set to 0.68 to transfer space occupancy constraints to the configuration discipline, the transfer factor is set to 0.95 to transfer real-time power demand curves and power consumption characteristics to the energy consumption discipline, and the transfer factor is set to 0.8. In S224, the transmission factor is set to 0.65 to transmit the layout of cable power supply equipment and energy consumption equipment to the configuration discipline, the transmission factor is set to 0.4 to transmit the maximum allowable power threshold under energy consumption constraints to the propulsion discipline, and the transmission factor is set to 0.85.

[0046] 2. The multinomial response surface model algorithm provided by the data modeling module in the optimization analysis layer has a high-precision simulation prediction accounting for 65% of the global target weight, and a low-precision surrogate model prediction accounting for 35% of the global target weight. When the deviation d between the low-precision surrogate model and the high-precision simulation result exceeds 12%, the weight of the low-precision surrogate model is reduced to 0.1.

[0047] 3. In step S310 of the intelligent solution layer, the chaos factor of the Logistic mapping is designed to be 3.58, including 80 individuals; In S350, the crossover probability through genetic operations is set to 0.8; the mutation probability is set to 1 / dimension, and for high dimensions, it can be uniformly set to 0.02. At the same time, 5% of elite individuals are retained to ensure population diversity and convergence; The initial temperature for simulated annealing was set to 300°C, and the temperature decay coefficient was set to 0.95. The particle swarm inertia weight is set to 0.4, and is also dynamically adjusted. It is set to 0.4 when the variable dimension is less than 80, 0.6 when the variable dimension is 80~100, and 0.8~0.9 when the variable dimension is 100~120. The two exchange superior individuals periodically; or the differential evolution strategy of "DE / rand / 1" or "DE / best / 1" is used to generate mutated individuals, with the differential mutation coefficient set to 0.48 to enhance local development capabilities.

[0048] Table 2 shows the technical specifications of the final deep-sea mining vehicle design in this embodiment:

[0049] Example 3: This embodiment is used to design auxiliary AUVs / ROVs for mining operations, and differs from Embodiment 1 in that: 1. In step S221 of the discipline decoupling layer: the transfer factor is set to 0.65 to transfer the structural weight and distribution to the advancement discipline, the transfer factor is set to 0.85 to transfer the weight and distribution to the energy consumption discipline, and the transfer factor is set to 0.85. In S222, the transmission factor is set to 0.55 to transmit fluid dynamics shape parameters and propulsion benefits to the propulsion discipline, the transmission factor is set to 0.45 to transmit equipment spatial layout and local energy distribution strategies to the energy consumption discipline, and the transmission factor is set to 0.90. In S223, the transfer factor is set to 0.78 to transfer space occupancy constraints to the configuration discipline, the transfer factor is set to 0.95 to transfer real-time power demand curves and power consumption characteristics to the energy consumption discipline, and the transfer factor is set to 0.8. In S224, the transmission factor is set to 0.7 to transmit the layout of cable power supply equipment and power consumption equipment to the configuration discipline, the transmission factor is set to 0.6 to transmit the maximum allowable power threshold under energy consumption constraints to the propulsion discipline, and the transmission factor is set to 0.95.

[0050] 2. The multinomial response surface model algorithm provided by the data modeling module in the optimization analysis layer has a high-precision simulation prediction accounting for 70% of the global target weight, and a low-precision surrogate model prediction accounting for 30% of the global target weight. When the deviation d between the low-precision surrogate model and the high-precision simulation result exceeds 12%, the weight of the low-precision surrogate model is reduced to 0.1. The model of the data modeling module is set according to the complexity and nonlinearity of the experimental data.

[0051] 3. In step S310 of the intelligent solution layer, the chaos factor of the Logistic mapping is designed to be 3.8, including 100 individuals; In S350, the crossover probability through genetic operations is set to 0.95; the mutation probability is set to 3 / dimension, and for higher dimensions, it can be uniformly set to 0.05. At the same time, 10% of elite individuals are retained to ensure population diversity and convergence; The initial temperature for simulated annealing was set to 200°C, and the temperature decay coefficient was set to 0.99. The particle swarm inertia weight is set to 0.9 and dynamically adjusted. When the variable dimension is less than 80, it is set to 0.6; when the variable dimension is 80~100, it is set to 0.8; and when the variable dimension is 100~120, it is set to 0.9. The two groups periodically exchange superior individuals. Alternatively, the differential evolution strategy of "DE / rand / 1" or "DE / best / 1" can be used to generate mutated individuals. The differential mutation coefficient is set to 0.65 to enhance local development capabilities.

[0052] Table 3 shows the technical specifications of the final deep-sea mining vehicle design in this embodiment:

[0053] Comparative Example 1: A deep-sea mining vehicle was designed using a general-purpose multidisciplinary design platform, the X-ship performance comprehensive optimization platform. Comparative Example 1 is specifically customized for the surface environment of ships; however, the models do not consider seabed sediments or submerged seawater environments, requiring manual reconstruction of the CAE models, resulting in limited openness. Its core capabilities focus on the collaboration of fluid mechanics and static structures, but lack extended support for key disciplines such as dynamic response, control, and manufacturing processes. The platform exhibits uneven performance in optimizing the fidelity of deep-sea operational equipment models; surrogate models show high accuracy, but the fidelity for complex, strongly coupled problems (such as fluid-structure resonance) is significantly insufficient. Computational efficiency is at a weekly optimization cycle, which can be accelerated through distributed parallelism, but algorithm convergence requires numerous iterations, presenting a certain efficiency bottleneck. Its automation level is limited, reducing manual intervention by only 40% through a mature CAD / CAE toolchain. It does not support robust optimization (lacking failure probability control capabilities), lacks sensitivity analysis functions, and multi-objective optimization is limited to two objectives, failing to meet the requirements of large-scale system and component collaborative optimization for deep-sea equipment. Its openness is weak, only supporting the import of typical CAE models. This platform is suitable for deterministic design scenarios that are dominated by a single discipline and are weakly coupled, such as ship fluid shape optimization, but it cannot cope with the requirements of high reliability or complex deep-sea systems.

[0054] Table 4 shows the technical specifications of the final deep-sea mining vehicle design in Comparative Example 1:

[0055] Comparative Example 2: A deep-sea mining vehicle is designed using a general-purpose multidisciplinary design platform and a complex industrial product multidisciplinary optimization platform. It supports collaboration among structural mechanics, aerodynamics, and electromagnetics, but lacks support for key deep-sea mining operations such as control and propulsion. Its advantage lies in the high accuracy of the surrogate model; however, it suffers from insufficient fidelity in strongly coupled disciplines, poor simplification and computational model capabilities, and optimization cycles lasting up to months. This platform employs distributed parallel processing logic and acceleration strategies, enabling good linking between models and reducing manual intervention by 80%. It achieves good full-process automation and integrates a CAD / CAE toolchain for automatic task scheduling. This platform possesses single-system multi-objective optimization capabilities (supporting ≥3 objectives), but lacks an intelligent solution filtering mechanism; it provides basic robust optimization (failure probability ≤1e-3), failing to meet the high reliability requirements of deep-sea operating environments; it only supports local unit sensitivity analysis and cannot handle system-level coupling sensitivity. The automation advantages are offset by inefficient computation and weakly coupled processing capabilities, making it only suitable for optimizing non-strongly coupled, low-dynamic-response vehicle components (such as traditional chassis structure design).

[0056] Table 5 shows the technical specifications of the final deep-sea mining vehicle design in Comparative Example 2:

[0057] The results show that the deep-sea mining vehicle designed in the embodiments achieves a high level of lightweighting, with a lightweighting coefficient of 4.26-4.33. The ground pressure of the deep-sea mining vehicle designed in embodiments 1-3 is reduced by 20%, and the ground pressure unevenness is significantly reduced to 0.05-0.12, fully ensuring the operational stability of the walking mechanism. In addition, the thrust / drag ratio is 1.21-1.35, avoiding the shortcomings of insufficient thrust reserve and redundant thrust settings that were significant in comparative embodiments 1 and 2. Regarding energy consumption, the embodiments, through multidisciplinary optimization of energy consumption, achieve an energy consumption of 4.4-5.6 kW*h / t, achieving low-energy operation of less than 6 kW*h / t.

[0058] The table below shows the comparison results of the design platform's advancements in the examples and comparative examples:

[0059] The comparison shows that Examples 1-3 achieve a comprehensive breakthrough in the advancement of the multidisciplinary optimization design platform, making them suitable for the deep-sea equipment design environment. In contrast, the comparative examples lack the capability to adapt to deep-sea mining equipment design in many aspects, specifically: In terms of breadth and depth, it covers up to five disciplines, including static-dynamic structures, fluid dynamics, control, propulsion, and manufacturing processes, significantly surpassing the limitations of Comparative Example 1 (two disciplines) and Comparative Example 2 (three disciplines). By automatically switching between surrogate models and high-precision CAE simulation, it solves the shortcomings of insufficient fidelity in the complex models of Comparative Example 1 and poor fidelity in strongly coupled disciplines of Comparative Example 2. Furthermore, it utilizes multi-gradient techniques to efficiently handle nonlinear strongly coupled problems. In terms of efficiency and performance, the embodiment achieves a weekly optimization cycle comparable to Comparative Example 1 and far superior to the monthly time consumption of Comparative Example 2. Simultaneously, the CPU / GPU hybrid parallel acceleration ratio reaches 80 times (Comparative Example 1 is 60 times, Comparative Example 2 is 50 times), and the algorithm convergence speed (50-200 iterations) is significantly improved compared to Comparative Example 1 (100-500 iterations) and Comparative Example 2 (300-400 iterations). In terms of methodology and architecture, the embodiment compresses the robustly optimized failure probability to... (Reduced by two orders of magnitude compared to deterministic optimization), Comparison ratio 1 ( ) and Comparative Example 2 ( It is more reliable and supports optimization with more than three objectives and system-to-subsystem level collaboration. Its system-wide sensitivity analysis capability also fills the gap in Comparative Example 1. In addition, the embodiment is a CAE toolchain customized for complex deep-sea environments, supporting the integration of multiple legacy models, which is more technically challenging than the conventional scenario adaptation of Comparative Example 1 / 2; the multi-process interactive interface further optimizes the user experience and enables efficient collaborative design for complex engineering problems.

[0060] In summary, the above embodiments have the following technical effects: 1. The data input layer allows users to customize parameterized input templates through a templated parameter port, categorizing constants into primary and secondary types: primary constant parameters (such as seawater depth and sediment cross-shear strength) are directly input by designers; secondary constant parameters (such as hydrostatic pressure and sediment cohesion) are automatically predicted based on environmental functions of polymetallic nodule mining areas, reducing the burden of manual input. The deep-sea environment database integrates historical data and machine learning verification rules to perform range checks and logical consistency verification on parameters, triggering recalculation or manual review for abnormal data. A common format is used to define data semantics, establishing unified data standards to resolve semantic inconsistencies; a metadata framework records data sources, types, and units, enhancing traceability; distributed storage and the Spark computing framework support large-scale data collaborative processing, ensuring cross-platform interoperability. Ultimately, this solves the problem of complex and unstandardized input of deep-sea environmental parameters.

[0061] 2. The subject decoupling layer introduces the Sobol index and correlation coefficient matrix to classify the coupling strength between subjects into three categories: weak, medium, and strong. Weakly coupled subjects are optimized using Jacobi iterative parallel execution; strongly coupled subjects are optimized using serial coordinated optimization. The random forest classification model is trained using historical data to identify coupling parameter combinations prone to oscillations in advance, reducing the risk of non-convergence. Built-in benchmark cases for typical problems such as fluid dynamics-structure coupling are used to verify the performance of the decoupling algorithm; a real-time monitoring panel displays key indicators such as coupling residuals and iteration counts, supporting dynamic tuning. Dynamic coupling strength quantification and decoupling priority adjustment are implemented, solving the problems of complex multi-disciplinary coupling and rigid decoupling strategies.

[0062] 3. A multi-fidelity optimization strategy was developed, which combines high-precision simulation and low-precision surrogate models in the optimization analysis layer. Weights are dynamically allocated based on sensitivity, with high-precision simulation being prioritized for highly sensitive parameters and surrogate models used for low-sensitivity parameters. A bias correction method was designed. The experimental design module selects strategies (full factorial, Latin hypercube, etc.) based on the number of targets to reduce redundant experiments. A scheme for local high-precision optimization and global simulation verification was designed, achieving periodic full-system verification and convergence control, thus solving the problem of balancing model accuracy and computational efficiency.

[0063] 4. A diverse initial population is generated using chaotic logistic mapping to avoid local optima; simulated annealing, PSO, and differential evolution (DE) are embedded to enhance local exploitation capabilities; the constraint tolerance is gradually relaxed, prioritizing the retention of feasible solutions. A multi-stage optimization and parallel computation approach is designed, starting with a single objective and then expanding to multiple objectives, using staged optimization to reduce complexity; the population is divided into 3-6 subgroups for parallel optimization, and efficiency is improved through elite selection and subgroup partitioning. Ultimately, the improved NSGA-III algorithm effectively overcomes the limitations of the NSGA-II and NCGA algorithms in deep-sea mining equipment design.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep-sea mining equipment design system based on an improved genetic algorithm, characterized in that, include: Data Input Layer: Used to configure templated parameter ports, allowing users to define parameterized input templates and set input constants, input variables, constraints, and optimization objectives; Subject decoupling layer: includes subject establishment module and coupling analysis module, responsible for the integration and coupling of various subject models; Optimization and analysis layer: This includes the experimental design module and the data modeling module, which are responsible for executing experimental design and optimization algorithms and supporting user modeling decisions; Intelligent solution layer: Adopts a multi-stage optimization principle, optimizes the objective in stages, first optimizes a single objective and then expands to multiple objectives; designs optimization strategies for the surrogate model, can automatically perform experimental design sampling and update the surrogate model, and automatically call quasi-Newton method, SPEA2, and improved NSGA-III optimization algorithms; Operational Interaction Layer: Utilizing PySide interface design technology to provide UI controls and functions, ensuring efficient operation through multidisciplinary design optimization. It efficiently manages the thread pool using QThreadPool and QRunnable technologies, introduces a cloud data storage module for parallel processing to avoid frequent thread creation and destruction, and supports collaborative editing and conflict detection by experts in mechanical, control, and environmental fields.

2. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 1, characterized in that, The data input layer sets the main input constant parameter and the secondary input constant parameter for the input constant; The input variables are divided into structural design variables, power and energy design variables, and support system variables. The structural design variables are divided into the length, width, height, external features, topology, local structural degrees of freedom, and strength redundancy of the connecting parts of the mining equipment. The power and energy design variables are divided into propulsion device power and layout, battery capacity, and energy distribution strategy. The supporting system variables are divided into umbilical cable armor thickness and maximum load of deployment and recovery device. The constraints are categorized into behavioral constraints, intensity constraints, and discipline consistency constraints. The optimization objectives include the weight and distribution of deep-sea mining equipment, travel-recovery resistance, and operational energy consumption.

3. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 2, characterized in that, The data input layer includes the following operational steps: S110: The main input constant parameters are entered by the designer. S120: Establish a deep-sea environment database, input secondary input constant parameters, and based on the deep-sea environment function of the polymetallic nodule mining area, use the main input constant parameters as independent variables to complete the automatic prediction and input of multi-dimensional characteristic parameters of seawater, sediment, and nodules at different depths in the collection area. S130: Combine machine learning with embedded automated verification rules to perform sub-constant range checks and logical consistency verification.

4. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 3, characterized in that, Step S130 includes: S131: After the user inputs the main constant parameters, the pre-trained random forest model is called to predict the validity probability p of the combination of input constant parameters; if the validity probability p ≥ 0.9 threshold, the test is passed. If the legality probability p < 0.9 threshold, a range check rule base comparison is triggered, indicating that the limit is exceeded or the parameters do not match, and a recalculation or manual review process is triggered. S132: After the environment function automatically predicts the input constant, the outlier score q is calculated using the isolated forest model; if the outlier score q ≤ 0.6, the parameters are considered reasonable. If the anomaly score q>0.6, initiate logical consistency verification, check parameter correlation, generate a visual report for complex anomalies, highlight contradictory parameters and recommend correction suggestions, and realize parameter visualization manual verification and error correction; S133: The correlation characteristics are as follows: hydrostatic pressure has a positive correlation with seawater depth, with an approved pressure-to-depth ratio of 0.01 MPa / m; seabed temperature is adjusted to 5℃-25℃ from the surface to 2000m, based on the marine environment and air temperature of the mining area; 2℃-4℃ from 2000-3500m; 1℃-2℃ from 3500-4000m; and slightly higher from 4000-6000m due to the adiabatic self-pressure effect of seawater, adjusted according to the statistical conditions of seabed hydrothermal vents in the mining area. The salinity of seawater tends to be uniform after exceeding the halocline, ranging from 34‰ to 35‰. Seawater density increases with depth, reaching 1025-1030 kg / m³ at depths of 4000-6000 m. Seawater viscosity increases with decreasing temperature and slightly with increasing salinity, reaching 1.37-1.69 × 10⁻³ Pa·s at depths of 4000-6000 m. The ocean current velocity at depths of 4000-6000m is affected by density differences and geostrophic currents, ranging from 3cm / s to 30cm / s.

5. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 4, characterized in that, The subject decoupling layer includes the following operational steps: S210: The main body of deep-sea mining equipment is divided into structural discipline, configuration discipline, propulsion discipline, and energy consumption discipline through the discipline establishment module; S220: By analyzing the characteristics of each component, the subject-based module determines the data flow of constants and input variables that it transmits to each subject and system layer. S230: Introduces the Sobol index and correlation coefficient matrix into the coupling analysis module to quantify the coupling strength between disciplines. A Sobol index < 0.2 is defined as a weakly coupled discipline, a Sobol index of 0.2 to 0.4 is defined as a moderately coupled discipline, and a Sobol index > 0.4 is defined as a strongly coupled discipline. A correlation coefficient |r| > 0.5 between input variables is considered a strong correlation, while |r| < 0.5 is considered a weak correlation. Prioritize disciplines with high Sobol indices or strong correlations, and establish a proxy model of discipline coupling structure based on sensitivity parameters to guide decoupling priorities; Jacobi iterative parallel execution is used for weakly coupled disciplines, while serial coordination is used for strongly coupled disciplines.

6. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 5, characterized in that, Step S220 includes: S221: In the design, the structural discipline transmits the equipment geometry and topological constraints to the configuration discipline with a transmission factor of 0.65-0.85; transmits the structural weight and distribution to the propulsion discipline with a transmission factor of 0.85-0.95; and transmits the weight and distribution to the energy consumption discipline with a transmission factor of 0.85-0.

95. S222: Configuration discipline transmits layout feasibility feedback to structural discipline, with a transmission factor set at 0.55-0.65; it transmits fluid dynamics shape parameters and propulsion benefits to propulsion discipline, with a transmission factor set at 0.45-0.55; and it transmits equipment spatial layout and local energy distribution strategies to energy consumption discipline, with a transmission factor set at 0.75-0.

90. S223: The propulsion discipline transmits the structural strength requirements of the propulsion installation location to the structural discipline, with a transmission factor set to 0.68-0.78; transmits the space occupancy constraints to the configuration discipline, with a transmission factor set to 0.85-0.95; and transmits the real-time power demand curve and power consumption characteristics to the energy consumption discipline, with a transmission factor set to 0.8-0.

9. S224: The energy consumption discipline transmits the sensitivity of energy consumption to weight distribution to the structural discipline, with a transmission factor set at 0.65-0.75; transmits the layout of cable power supply equipment and energy consumption equipment to the configuration discipline, with a transmission factor set at 0.35-0.65; and transmits the maximum allowable power threshold under energy consumption constraints to the propulsion discipline, with a transmission factor set at 0.85-0.

95.

7. A deep-sea mining equipment design system based on an improved genetic algorithm according to claim 6, characterized in that, The optimization analysis layer supports the integration of process models built by Isight, ModelCenter, Tosca, and Optistruct; It also includes a text parser, a system command executor, a calculator, and dedicated interface components for finite element software. The experimental design module mainly conducts structural mechanics tests, corrosion resistance assessment tests, fatigue tests, hydrodynamic tests, and flow field simulation and analysis tests for deep-sea equipment. Different experimental design strategies are adopted based on the number of optimization objectives and their coupling relationships. When the number of optimization objectives n < 3, and the objectives are not in the same discipline or the mutual influence coefficient c < 0.3, a full factorial design test is used. When the number of optimization objectives n < 3, and the objectives are in the same discipline or the mutual influence coefficient c > 0.3, an orthogonal low-workload design test is used. When the number of optimization objectives n > 3, a Latin hypercube or central composite design test is used. The data modeling module is equipped with a parametric integration interface for commercial software or self-developed programs for multidisciplinary design simulation, including SolidWorks, ANSYS, LongRuan4D-GIS, OpenFOAM, digital twin interactive control platform, and Simulink; enabling parametric integration and automatic calling of self-developed programs and commercial CAD / CAE programs. Furthermore, the data modeling module provides a multinomial response surface model algorithm and also has a surrogate model algorithm interface. Based on the interface specification, external surrogate model algorithms can be embedded into this platform. A multi-fidelity optimization strategy is adopted, which mixes high-precision simulation and low-precision surrogate models. High-precision simulation prediction accounts for 65%-75% of the global target weight, while low-precision surrogate model prediction accounts for 25%-35% of the global target weight. When the deviation d between the low-precision surrogate model and the high-precision simulation result exceeds 12%, the weight of the low-precision surrogate model is reduced to 0.

1. Based on the sensitivity of each system unit to the discipline, high-precision optimization is performed only on highly sensitive parameters. After every 5 local optimizations, a high-precision full-system simulation is performed. The overall convergence criteria are that the rate of change of the objective function is less than 1% for 3 consecutive iterations and the cross-disciplinary constraint violation is less than 5%.

8. The deep-sea mining equipment design system based on an improved genetic algorithm according to claim 7, characterized in that, The multidisciplinary design in the data modeling module includes: In structural engineering, high-precision simulations should be prioritized for detailed stress, strain, and fatigue analysis in three-dimensional geometry, as well as for material plasticity, creep, anisotropy, fluid loads, and structural responses. Low-precision surrogate models should be prioritized for rapid static analysis of rod and shell elements. The propulsion discipline conducts high-precision simulation modeling of propeller-assisted propulsion devices, and a low-precision surrogate model is used for the preliminary relationship between propulsion power and propulsion speed; The dynamic energy consumption simulation during the start-up / impact load change process of the energy consumption discipline adopts high-precision simulation, while the preliminary prediction of energy consumption per unit mining volume adopts a low-precision surrogate model. In the initial optimization stages of overall dimensions, layout, and center of gravity, the configuration discipline uses a low-precision surrogate model, while high-precision simulation is used for configuration driving-recovery resistance analysis.

9. A deep-sea mining equipment design system based on an improved genetic algorithm according to claim 8, characterized in that, The improved NSGA-III optimization algorithm in the intelligent solution layer includes the following steps: S310: Initialize the population: Based on historical data and the set of best points, generate an initial near-optimal diversity population using chaotic logistic mapping; S320: Assess fitness: Assess each individual and calculate its fitness value under multiple optimization objectives; S330: Fast Non-Dominated Sort: Store the sets of dominant and dominated individuals, sort all individuals in the set of dominated individuals, and if an individual has zero dominant individuals, it is considered a non-dominated individual, and its Pareto level is set to the current highest level plus one. S340: Crowding Calculation: In order to maintain population diversity, the crowding of each individual is calculated for subsequent elite selection; S350: Generation of offspring: Parameters are dynamically adjusted according to population diversity, and the next generation of population is generated through genetic operations. At the same time, 5% to 10% of elite individuals are retained to ensure population diversity and convergence. The generated offspring individuals are subjected to local perturbation and acceptance criteria judgment in simulated annealing. In the selection phase, the acceptance probability of simulated annealing is used to replace roulette wheel selection. The initial temperature of simulated annealing is set to 100 to 300, and the temperature decay coefficient is set to 0.90 to 0.

99. S360: Merge Populations: Merge the parent and offspring populations to form a new population; S370: Elite Selection: Divide the merged and mixed population into 3 to 6 subgroups, select individuals with high fitness and low crowding as elite individuals for the next generation population, prioritize the retention of feasible solutions, and gradually relax the constraint tolerance using a dynamic constraint method to complete the parallel distributed optimization and then merge them. S380: Repeated Iteration: Repeat the above steps until the set stopping condition is met; S390: Obtaining Pareto optimal solutions: The individuals in the final population are the Pareto optimal solution set, where each solution cannot be completely dominated by other solutions under all objectives.

10. A deep-sea mining equipment design system based on an improved genetic algorithm according to claim 1, characterized in that, The data input layer adopts HDFS and MinIO as distributed storage strategies and forms, and uses Memcached in-memory database and efficient indexing strategies to establish the Spark computing framework. The data input layer dynamically generates data structures, adopts JSON general format, text format, and ontology to define data semantics and formulate interdisciplinary data standards; it completes data classification by recording data source, type, and unit to form a data ownership list and establishes a metadata framework. The data input layer sets multi-weight and gradient constraints in the constraints, and adopts a multi-level constraint restriction strategy with priority safety > environment > cost > energy consumption optimization. The coupling analysis module of the subject decoupling layer is equipped with a random forest classification model, which is trained with historical optimization data to identify coupling parameter combinations that are prone to oscillation or non-convergence in advance. It also provides a real-time monitoring panel to display coupling residuals, iteration counts, subject model call status, and built-in benchmark cases of typical coupling problems for verifying decoupling algorithms and performance tuning. The model of the data modeling module in the optimization analysis layer is set according to the complexity and nonlinearity of the experimental data; The intelligent solution layer uses a quasi-Newton algorithm to solve single-objective optimization, and the SPEA2 algorithm to optimize multi-objective optimization models with 2 to 3 objectives. When the number of objectives exceeds 3, the improved NSGA-III algorithm is introduced. The operation interaction layer's working interface is divided into three interfaces: design, monitoring, and output. The design interface includes a categorized tree structure component library, a process structure tree, a global / local variable management area, a parameter mapping relationship management area, a priority setting area, and a resource quota selection area. The monitoring interface includes a task timeline view area and a progress monitoring bar; the output interface includes a multi-objective optimization result chart area, a data sandbox area, and a report generation area for key parameter impact and optimal solution recommendation.

Images