Method for optimizing anti-wave-impact structure of near-shore high-piled wharf of port

Abstract

The invention discloses a wave impact resistance structure optimization method for a near-shore high-piled wharf of a port. The method comprises the following steps: firstly, based on historical observation and reanalysis data, carrying out conditional sampling according to a significant wave height, a peak period, an incidence direction, storm surge, a long-period energy ratio and a water depth, and obtaining a sea condition sample set covering a design quantile; secondly, building a high-pile wharf parameterized geometry and working condition model; and finally, performing high-fidelity full-sample regression verification on the candidate solution to form a design result document including design variables, sea condition coverage domains, robustness indexes and key response statistics. According to the method, high reliability and high efficiency are unified, and the method is suitable for wave impact resistance design and engineering implementation of the near-shore long-piled wharf.

Description

Technical Field

[0001] This invention relates to the field of port and coastal engineering, and in particular to the optimization design method and fluid-structure-soil coupling and intelligent optimization technology for nearshore high-pile wharf wave-resistant structures. Background Technology

[0002] This invention belongs to the field of port and coastal engineering technology, addressing the design and evaluation of wave-impact resistant structures for nearshore high-pile wharves under storm surges and strong waves. Nearshore high-pile wharves are characterized by exposed pile foundations, lightweight superstructures, multiple deck openings, and diverse auxiliary components. They are subject to the coupled effects of multiple mechanisms, including wave jumps, wave impacts, heave, scour, and pile-soil interactions. Under extreme sea conditions, they are prone to localized deck overpressure, sudden increases in heave force, amplified peak internal forces at the pile tops, and fatigue problems. Existing engineering designs are mostly based on statistical quantiles or sea state parameters recommended by standards, supplemented by conventional hydroelastic or Morison-type calculations and empirical corrections. While feasible, these designs fail to adequately characterize highly nonlinear free surface processes, impact pressures caused by air entrainment, the redistribution effect of deck openings and protective components on the flow field, and time-varying pile-soil stiffness caused by scour. In numerical simulation, methods such as RANS-VOF, SPH, and LBM have been used for wave impact analysis. However, in large-scale 3D scenarios, they suffer from high computational costs, cumbersome parameter tuning, and significant sensitivity to mesh and time steps, making it difficult to support systematic assessments of hundreds or thousands of sea states. Regarding surrogate modeling, existing data-driven alternative models often rely on regression or meta-models with limited samples, resulting in weak physical consistency and boundary feasibility constraints, and high extrapolation risks. In optimization design, common practices target a single scenario or a few representative sea states, using heuristics or local searches to determine geometric and layout parameters. This often ignores the propagation of uncertainties and tail risk control caused by sea state, material, soil, and construction deviations, making it difficult to establish verifiable trade-offs between performance, cost, and constructability. In coupling analysis, pile-soil interactions typically use static or simplified p-y curves, which fail to reflect dynamic stiffness and damping changes under scour evolution, and the bidirectional coupling between superstructure dynamics and fluid loads is insufficient. Regarding verification and delivery, existing processes mostly focus on comparisons and checks of individual operating conditions, lacking regression verification covering a wider sea state range and systematic archiving of design results, which affects the traceability and review efficiency of the solutions. In summary, in the wave impact resistance design of nearshore high-pile wharves, there is still a lack of an integrated technical path oriented towards engineering applications that can unify "sea state generation—multi-fidelity calculation—physical embedding proxy—fluid—structure—soil coupling—multi-blob bar optimization—system verification," in order to reduce assessment costs, control tail risks, and output feasible design formulations while ensuring physical consistency. Summary of the Invention

[0003] One objective of this invention is to propose an optimization method for the wave-impact resistance structure of nearshore high-pile quays in ports. This invention employs techniques such as physically constrained conditional sea state generation, high-fidelity numerical calculation of two-phase free surfaces and prediction using physically embedded neural operators, fluid-structure-soil dynamic coupling, multi-objective optimization using pluripotentiary bars, and multi-fidelity Bayesian optimization. Based on historical observation and reanalysis data, a conditional vector is constructed according to significant wave height, peak period, incident direction, storm surge, long-period energy proportion, and water depth. Diffusion-based generation sampling is then implemented to obtain a sea state sample set covering the design quantile. This establishes a data set including pile diameter, pile spacing, inclined pile ratio, deck opening ratio and opening arrangement, flange height, and skirt length. The geometry and layout of energy dissipation components, as well as the parameterized geometric and working condition models of boundary, incident, and flow field conditions, and multi-stage scour and soil parameters, are used to perform unsteady calculations on sea state samples in differentiable smooth particle hydrodynamics or phase-field lattice Boltzmann solvers to output free surface evolution, local pressure time history, deck heave force, pile top internal force, and displacement response. A mid-fidelity wave-structure response predictor is obtained through physical embedding training with momentum, continuity equation residuals, and boundary conditions as regularization terms. A computational chain is constructed that couples the superstructure dynamic model with the pile-soil interaction model and updates pile-soil parameters with scour using dynamic Py or equivalent substructure methods to obtain the full structural time history response. The geometric and structural parameters are defined as follows: The design variables consist of a set of parameters, an objective function consisting of peak wave force, uplift force, peak displacement, critical section bending moment and shear force, fatigue cumulative damage, and material and construction costs, and a constraint set consisting of bearing capacity, stability, displacement limit, fatigue life, structural boundary, and construction feasibility. Under uncertainties related to sea state, materials, soil, and construction deviations, tail risk constraints are applied to the objective or constraints to form a multi-Brussels bar optimization model. A multi-fidelity Bayesian optimization model is employed, primarily using medium-fidelity prediction for evaluation and strategically incorporating high-fidelity calculations for calibration and convergence determination. This model outputs the Pareto front and candidate solutions, and high-fidelity full-sample regression is performed on the candidate solutions to generate a set of parameters including design variables. The design result document, which includes measured values, sea state coverage area, robustness indicators, key response statistics, and construction parameters, enables comprehensive evaluation, constrained optimization, and verifiable delivery of wave impact resistance design for nearshore high-pile wharves under multi-source uncertainty conditions. It has the technical advantages of covering the design quantile with sea state samples while maintaining physical consistency, reducing the dependence on full high-fidelity calculations while satisfying conservation and boundary conditions, providing coupled solutions for bearing capacity, displacement, and fatigue consistency verification under scour and pile-soil dynamic influences, achieving tail-risk-constrained design optimization with conditional risk value, obtaining verifiable Pareto solutions through multi-fidelity calibration, and supporting engineering verification and implementation with standardized result documents.

[0004] Beneficial effects

[0005] This invention employs historical observation and reanalysis data, with c = (Hs ,T p ,θ0,Δη,E lp d) represents the physical constraint diffusion-based generation and sampling of sea state for condition vectors, which realizes the construction of sea state sample sets covering the design quantile level of the target sea area. This effectively improves the coverage and representativeness of sea state samples for extreme combinations (from representative working conditions to sample domain evaluation), and enhances the consistency and comparability of load inputs (uniform caliber generation within the same condition domain).

[0006] This invention employs high-fidelity numerical computation on two-phase free surfaces and training of a fidelity predictor within a physically embedded neural operator to achieve the calculation of η(t) and p. loc (t), F deck The reusable mapping of key responses such as (t), N*pile(t), and U*pile(t) is used in the optimization iteration. The evaluation is mainly based on medium fidelity and high fidelity calibration is interspersed according to the strategy. This effectively increases the number of sea state-design combinations that can be evaluated per unit time, reduces the dependence on full high fidelity calls, and enhances the consistency of the evaluation process with the constraints of conservation and boundary conditions.

[0007] This invention employs a fluid-structure-soil coupled calculation chain, coupling the superstructure dynamic model with the pile-soil interaction model, and using a dynamic p-y or equivalent substructure method to update the pile-soil stiffness and damping under scour scenarios. This achieves displacement, internal force, and uplift force calculations consistent with wave load time histories, effectively improving the accuracy of peak displacement, critical section bending moment / shear force, and fatigue cumulative damage assessment in response to working condition evolution, and enhancing adaptability under scour evolution and pile-soil interaction conditions.

[0008] This invention employs multi-objective constraint modeling and uses CVaR α The method enables convergence determination of tail risk measurement through multi-fidelity Bayesian optimization, high-fidelity calibration, and full-sample regression verification and result documentation for candidate solutions. It achieves constrained optimization and Pareto solution set output for the tail of the objective and constraints at a given confidence level, effectively improving the stability and constraint satisfaction rate of candidate solutions in the verification of extended sea state sample sets, and enhancing the traceability of design results and the availability of engineering delivery. Attached Figure Description

[0009] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0010] Figure 1 This is a schematic diagram of the process of "an optimization method for wave impact resistant structure of a high-pile wharf near the port" proposed in this invention;

[0011] Figure 2This is a schematic diagram of the parametric geometry and working condition model elements of the "Optimization Method for Wave Impact Resistance Structure of a High-Pile Wharf Near the Port" proposed in this invention.

[0012] Figure 3 This is a schematic diagram of the multi-fidelity calculation and optimization relationship of "an optimization method for the wave impact resistant structure of a nearshore high-pile wharf" proposed in this invention. Implementation

[0013] The specific implementation of this invention addresses the wave impact optimization design of nearshore high-pile wharves in ports under combined sea conditions such as strong oblique incident waves, storm surge, and increased long-period energy ratio. It provides a reproducible parameter range, numerical implementation process, and training / optimization configuration, enabling those skilled in the art to complete deployment, calculation, and delivery without creative labor.

[0014] Step 1 includes the following:

[0015] S1.1 Collect historical observation data set D obs With reanalysis dataset D rea And merge to generate the basic sea state dataset D0;

[0016] S1.2 performs quality control, missing data interpolation, and resolution unification on D0 to obtain the normalized dataset D1;

[0017] S1.3 Calculation of significant wave height H based on D1 s Peak period T p Incident direction θ0, storm surge Δη, long-period energy percentage E lp Construct a condition vector with water depth d.

[0018] c=(H s ,,T p ,,θ0,,Δη,,E lp ,,d);

[0019] S1.4 is generated during the forward diffusion process.

[0020]

[0021] S1.5 utilizes a noise cancellation network F during the reverse generation process. θ Constructing conditional probabilities with energy conservation constraints

[0022] p θ (x*0∣c);

[0023] S1.6 sets the confidence quantile p*d, and for p θ (x*0|c) samples N times to obtain the sample set x*0 (i) *i=1 N ;

[0024] S1.7 outputs the sea state sample set S*sea for use in subsequent steps.

[0025] Step S2 includes the following:

[0026] S2.1 Set the pile foundation parameters and determine the pile diameter D sequentially. pile Pile spacing S pile Inclined pile ratio R incl ;

[0027] S2.2 Set the parameters of the superstructure and determine the pile cap length L in sequence. cap Pile cap width B cap Liang Gao

[0028] h beam Deck opening ratio P deck A*deck is the aperture arrangement matrix.

[0029] S2.3 Set the parameters of the protective components, and determine the edge plate height h*edge and skirt plate length l in sequence. skirt The geometric parameter vector g*diss of the energy dissipation component and its arrangement matrix

[0030] S2.4 Set the overall layout parameters and determine the deck elevation z in sequence. deck berth direction θ berth and the C*layout matrix for vertical and horizontal layout;

[0031] S2.5 generates a parametric geometric model M*geo and performs mesh generation, outputting a geometry and mesh dataset G. mesh ;

[0032] S2.6 in G mesh Wave incidence conditions, flow field conditions, and soil boundary conditions are applied to form a parametric chemical condition model M. load ;

[0033] S2.7 Based on the scour depth sequence d scour The pile-soil interaction parameters are updated along with the soil layer parameter matrix D*soil to generate a comprehensive parametric geometric and working condition model M*struct.

[0034] Step S3 includes the following:

[0035] S3.1 Select either a differentiable smooth particle hydrodynamics solver or a phase-field lattice Boltzmann solver to establish a high-fidelity two-phase free surface solver G. hi ;

[0036] S3.2 Set the water density ρ w air density ρ aDynamic viscosity of water (μ) w Aerodynamic viscosity μ a Surface tension coefficient σ and gravitational acceleration vector g;

[0037] S3.3 Import M struct With G mesh Apply boundary conditions to the solution domain;

[0038] S3.4 Press S sea =x*0 (i) *i=1 N Load the incident field conditions, set the time step Δt and the total simulation duration T. sim ;

[0039] S3.5 Calculate the velocity field u in each time step n. n Pressure field p n relative to the free liquid surface position η n ,satisfy

[0040]

[0041] where ρ∈ρ w ,ρ a τ is the viscous stress tensor;

[0042] S3.6 Extract deck lift force Pile top internal force vector N*pile (i) (t), pile top displacement vector U*pile (i) (t) and local pressure time history

[0043] S3.7 Output η (i) (t k ), N*pile (i) (t), U*pile (i) (t).

[0044] Step S4 includes:

[0045] S4.1 Concatenate the sea state parameter vector c with the structural geometry parameter vector g*struct to form the input vector x*in;

[0046] S4.2 Constructing a Physically Embedded Neural Operator Model N θ And define physical residuals

[0047]

[0048] and boundary condition residual R b The physical loss L is obtained. phys =R p +Rb ;

[0049] S4.3 The supervised dataset output from step S3 Training N θ Constructing data loss

[0050]

[0051] S4.4 Connect L*data with L phys Linear weighting yields

[0052] L total =w d L data +w p L phys ,

[0053] Where w d With w p Let L be the weighting coefficient, and minimize it through backpropagation. total Update N θ parameter;

[0054] S4.5 output mid-fidelity waveform—structural response predictor P mid For subsequent use.

[0055] Step S5 includes:

[0056] S5.1 Establishing the dynamic model M of the superstructure sup Its mass, stiffness, and damping matrices are M*sup, K*sup, and C*sup, respectively;

[0057] S5.2 Establish a pile-soil interaction model M*ps, and use a dynamic py relationship for the horizontal stiffness curve k. py (z) and at the scouring depth d scour Next update;

[0058] S5.3 will M sup With M ps Couple substructures to form an overall equation.

[0059]

[0060] S5.4 Based on the external load in step S3 and P in step S4 mid The results are solved using time-history iteration, outputting U*pile(t), N*pile(t), and U... deck (t).

[0061] Step S6 includes:

[0062] S6.1 Define the design variable vector

[0063] v=(D pile ,,S pile ,,R incl ,,L cap ,,B cap ,,h beam ,,P deck ,,h edge ,,l skirt ,,z deck ),

[0064] And set the upper and lower bounds v*min, v*max;

[0065] S6.2 Define the objective function vector

[0066] J = (F peak ,,F deck,peak ,,U deck,peak ,,U pile,peak ,,M sec ,,V sec ,,D fat C cost );

[0067] S6.3 Define the constraint set C = F ult ≥F peak ,,U lim ≥U deck,peak ,,D fat ≤D lim Geometric construction limits and construction feasibility limits;

[0068] S6.4 Constructing a Multi-Objective Constrained Optimization Problem Make v∈[v*min,v*max] and satisfy C.

[0069] Step S7 includes:

[0070] S7.1 Establishing the joint uncertainty distribution P ξ ξ covers sea state, material parameters, soil parameters, and construction deviations;

[0071] S7.2 For each objective function J k Calculate Conditional Value at Risk

[0072]

[0073] Where q α For the corresponding quantiles;

[0074] S7.3 Construct a partial Bruker optimization model with CVaR*α(J*k) as constraints, and generate a robust evaluation sample set S rob .

[0075] Step S8 includes:

[0076] S8.1 P in step S4 mid Establish a Gaussian process agent GP and initialize the sample library D*0;

[0077] S8.2 Select the desired improvement criterion EI(v) as the acquisition function and calculate...

[0078] EI(v)=(μ(v)-J * ),Φ(z)+σ(v),φ(z),

[0079] J * Let μ and σ be the current optimal values, respectively, and let Φ and φ be the predicted mean and variance of GP, respectively.

[0080] S8.3 Under the condition of satisfying the robustness requirement of step S7, for the design variable v with the largest EI. * Call G*hi from step S3 to perform calibration and update the sample library D;

[0081] S8.4 When the convergence criterion is satisfied, output the Pareto front set P and the candidate solution set V. * ;

[0082] S8.5 for each v * ∈V * Perform high-fidelity full-sample regression validation to generate a design results document that includes design variable values, sea state coverage area, robustness indicators, and key response statistics. Example 1

[0083] To verify the applicability and verifiability of this invention under engineering conditions, a nearshore high-pile wharf expansion project on the northeast coast of Jintang Island in Zhoushan City was selected as the application object. The offshore area of ​​this site is affected by the southeast monsoon and typhoon remnants. Traditional verification based on a small number of representative working conditions may result in excessive deck lifting force and pile top horizontal displacement approaching the limit when there is a high proportion of strong oblique incident and long-period energy. Existing schemes do not adequately consider the degradation of pile-soil dynamic stiffness during the scour evolution stage, leading to deviations in fatigue life assessment. This invention constructs a sea state sample domain based on historical observation and reanalysis data in this scenario, establishes a parametric geometric and working condition model, outputs key time history responses through a two-phase free surface high-fidelity solver, trains a physically embedded neural operator to form a mid-fidelity predictor, and couples the superstructure and pile-soil model to form a fluid-structure-soil dynamic calculation chain. Under the uncertainties of sea state, materials, soil, and construction deviations, the tail performance is constrained by conditional value of risk, and multi-fidelity Bayesian optimization is used to search for design variables. Finally, regression verification is completed on an extended sample, and the design formulation is output. The application process is executed in a unified software and hardware environment to ensure data traceability and reproducibility.

[0084] The scenario conditions were compiled from operational data provided by the wharf owner and data from local marine observation stations. The target structure is a high-pile wharf with a five-span continuous beam superstructure, where longitudinal load-bearing is provided by a pile group. The deck has multiple sets of openings to meet drainage and pipeline crossing requirements. Protective components include flanges and skirts, and the berth direction has a fixed angle with the dominant wave direction. During a severe storm in 2024, the existing design recorded localized deck bottom pressure exceeding the initial design reference value. Subsequent finite-condition numerical analysis showed that the horizontal displacement at the pile top was close to the specification limit. The owner required that, without significantly increasing material costs, the design and layout be optimized to control heave, internal forces, and fatigue indicators, while simultaneously covering sea state combinations up to the 0.99 quantile. This invention completed data construction, optimization, and verification from March to June 2025. The site, timeline, and original requirements are shown in the table below.

[0085] Table 1 Project Site and Time Information

[0086] project content Site Expansion section of the nearshore high-pile wharf on the northeast shore of Jintang Island, Zhoushan City, Zhejiang Province coordinate 30.01°N, 121.97°E Implementation time March 2025 to June 2025 Existing problems Under conditions of strong oblique incidence and long-period energy, the heave is excessive and the pile top displacement is close to the limit. Target Controlling heave, displacement, and fatigue parameters while constraining costs under sea state coverage at 0.99 quantile.

[0087] The sea state sample domain was constructed according to the S1 process of this invention, with a time window covering nearly twenty years of reanalysis data and site tide records. The statistical range of significant wave heights is 1.2–5.8 m, peak periods are 5–14 s, the incident direction relative to the berth normal is distributed between -45° and 45°, and the proportion of long-period energy is between 0.05–0.22. To verify the superimposed effects of storm surge, a sample set was formed by combining local station tide records with wave state data. In this scenario, this invention uses 2000 samples as input for design and optimization, and controls tail coverage with a confidence quantile of 0.99. Representative sea state statistics are shown in the table below.

[0088] Table 2. Statistics of sea state samples (site: northeast coast of Jintang Island, Zhoushan; time window: 2005–2024)

[0089] index mean median Quantile 0.95 Quantile 0.99 Remark Significant wave height Hs(m) 2.7 2.5 4.6 5.6 Verification by reanalysis and observation Peak period Tp(s) 9.1 8.7 12.8 13.9 JONSWAP spectral fitting Incident direction θ0 (°) 3 2 34 42 relative berth normal Storm surge Δη(m) 0.38 0.33 0.86 1.07 Including astronomical tides Long-period energy percentage Elp 0.12 0.11 0.20 0.22 Energy share below 0.5–2 Hz Water depth d (m) 11.4 11.2 14.6 15.8 Average during engineering survey period

[0090] The parametric geometry and working condition model were constructed according to the S2 specification. The initial geometry referenced the existing design, with a pile diameter of 1.2m, pile spacing of 5D, a slanted pile ratio of 0.2, pile cap dimensions of 4.0m × 4.0m, beam height of 1.2m, deck opening ratio of 0.12 arranged in an interlaced matrix, edge plate height of 0.35m, skirt length of 1.0m, deck elevation of 4.2m, and a berth orientation deflected 15° relative to the shoreline normal. The working condition model used JONSWAP spectrum incident flow and a constant background velocity of 0.6m / s. The scour sequence was set in a multi-level scene of 0–2.0D, and soil parameters were derived from survey reports and laboratory tests. The geometry, working condition, and soil parameters of this invention in this scene are summarized below.

[0091] Table 3. Parametric Geometry and Working Condition Model Elements (Initial Combination and Variable Range)

[0092] elements initial value Variable range Remark Pile diameter Dpile (m) 1.20 0.80–2.00 steel pipe piles Spile (m) 6.0(5D) 4D–7D Consistent in all directions Inclined pile ratio Rincl 0.20 0–0.50 staggered arrangement Pile cap L×B (m) 4.0×4.0 3.0–6.0 Regular grid Beam height (hbeam(m)) 1.20 0.8–1.8 Equal height section Pdeck opening ratio 0.12 0.08–0.30 staggered array Hedge height (m) 0.35 0.20–0.70 Straight Skirt length (lskirt(m)) 1.00 0.5–1.8 Vertical skirt board Deck elevation zdeck (m) 4.20 3.8–5.2 relative to the reference plane Berth direction θberth (°) 15 0–45 relative to shoreline normal Background flow velocity (m / s) 0.6 0.2–1.2 constant profile flushing sequence dscour 0, 0.5D, 1.0D, 1.5D, 2.0D Cascade verification

[0093] In the high-fidelity calculation of the two-phase free surface, a differentiable SPH solver was used with a particle spacing of 0.1 m, a CFL number of 0.2, and a single sample simulation duration of 180 s, with the first 60 s as a transition period and the last 120 s for data collection and statistics. For each combination, the free surface evolution and local pressure time histories were output, and the peak values ​​and statistics of deck lifting force, pile top internal forces, and displacements were extracted. Using these "teacher data," a physical embedded neural operator was trained to form a mid-fidelity predictor with a network width of 96 and a depth of 5. The loss function consisted of data terms and physical residual terms with a weight ratio of 0.7:0.3. Subsequently, a coupled calculation chain between the superstructure and the pile-soil model was established. Dynamic p-y curves were used to update the soil spring stiffness and damping under different scour levels. Time history integration was performed using the Newmark method, and mid-fidelity output was used as the main load, with high-fidelity calibration interspersed according to a strategy.

[0094] In the optimization phase, pile diameter, pile spacing, proportion of inclined piles, pile cap size, beam height, deck opening ratio and arrangement, flange height, skirt length, deck elevation, and berth orientation were used as design variables. The objective function included peak wave impact force, uplift force, peak deck and pile top displacement, critical section bending moment and shear force, fatigue cumulative damage, and material and construction costs. The constraint set corresponded to bearing capacity, stability, displacement, fatigue life, and structural boundaries. Under the uncertainty of sea state, materials, soil, and construction deviations, a confidence level of 0.95 was set, and tail performance was limited by conditional value of risk. Multifidelity Bayesian optimization was used for the search, with midfidelity as the master evaluator and highfidelity calibration inserted every ten rounds. After optimization convergence, a Pareto front was formed, and full-sample highfidelity regression validation was performed on 500 sea state samples, outputting the design results document.

[0095] To demonstrate the effectiveness of this invention, a comparison was made between the traditional representative operating condition method and the method of this invention under the same budget and hardware conditions, spanning from March 2025 to June 2025. The hardware environment consisted of dual-socket servers, each equipped with a 64-core CPU and two A100 GPUs, with a unified software stack. The comparison of optimization efficiency, computational resources, and model accuracy is as follows.

[0096] Table 4 Comparison of Computational Resources and Optimization Efficiency (Zhoushan Scenario, March 2025 to June 2025)

[0097] index Traditional representative working condition method Method of the present invention Remark Number of representative working conditions 12 articles 2000 sample domains (medium-fidelity main evaluation) Different sea state coverage methods High-fidelity call count 120 times 14 times Fixed calibration frequency strategy Medium-fidelity call count not applicable 286 times Neural operator reasoning Single-round evaluation throughput 1–2 plans / day 12–18 plans / day Same budget, same hardware Total number of iterations 28 62 More thorough early exploration Optimize total time 67 days 39 days Including calibration and verification Convergence criterion achieved Unstable Achieved (EI < 1% in the last 10 rounds) Unified Judgment Criteria

[0098] The structural response and robustness indices are compared across representative sea state groups and extended samples as follows. The method of this invention, while satisfying displacement and load constraints, reduces uplift and peak local pressure to below existing schemes, and maintains indices within limits under scour levels of 1.0D–2.0D. Fatigue damage is assessed using rainflow counts and mining criteria at 10^6 cycles.

[0099] Table 5 Comparison of key responses and robustness indicators (partial statistics, α = 0.95)

[0100]

[0101]

[0102] The impact of changes in structural parameters on the response is reflected in the combined adjustment of the deck opening ratio, flange height, and skirt length. The method of this invention adjusts the opening ratio from 0.12 to 0.18 and employs a staggered array, resulting in smoother gas-liquid exchange and reduced lift. The flange height is adjusted from 0.35m to 0.45m, and the skirt length is increased from 1.0m to 1.4m, leading to a more balanced redistribution of the flow field during oblique incidence. The deck elevation is adjusted from 4.2m to 4.6m, increasing the free plate height. The pile spacing is slightly adjusted from 5D to 5.5D, and the proportion of inclined piles is adjusted from 0.2 to 0.3 to improve the overall stiffness of the pile group. A comparison of the structure and cost before and after optimization is as follows.

[0103] Table 6 Comparison of Construction Parameters and Engineering Quantity Estimates

[0104] project Initial Plan Candidate solutions of this invention range of change Pdeck opening ratio 0.12 0.18 +50％ Hedge height (m) 0.35 0.45 +0.10 Skirt length (lskirt(m)) 1.0 1.4 +0.4 Deck elevation zdeck (m) 4.2 4.6 +0.4 Spile (m) 6.0 6.6 +0.6 Inclined pile ratio Rincl 0.20 0.30 +0.10 Steel volume (t) 100% baseline 96.1% baseline -3.9％ <![CDATA[Volume of the structural member (m 3 )]]> 100% baseline 102.4% baseline +2.4％ Materials + Construction Costs 100% baseline 95.2% baseline -4.8％

[0105] The project delivery requirements include full-sample high-fidelity regression validation of candidate solutions. This invention performs SPH simulations on 500 samples, extracts key statistics, and compares them with medium-fidelity predictions. The root mean square relative error remains within a specified threshold. The validation results are shown in the table below.

[0106] Table 7. High-fidelity regression validation statistics (500 samples)

[0107]

[0108]

[0109] The application process was not described in bullet points; the workflow progressed along a unified timeline. During the data preparation phase, nearly twenty years of sea state data were time-aligned and quality-controlled. The incident field was generated using a diffusion-based model to form the sample domain, avoiding biases caused by relying on only a small number of representative cases. Geometric and operational condition models were built within a parametric framework, considering variations in pile-soil stiffness under different scour levels. High-fidelity solving output teacher data with a differentiable SPH, while a medium-fidelity neural operator served as the master evaluator, providing timely response mappings. Coupled computational chains completed the dynamic interaction between the superstructure and the pile-soil system at the same time step. In the optimization phase, conditional value of risk (VHR) was used to limit tail risk, ensuring controllable performance of indicators at a confidence level of 0.95. Multi-fidelity Bayesian optimization achieved error convergence through limited high-fidelity calibration. The converged candidate solutions underwent regression validation on 500 samples, resulting in a design outcome document containing design variables, sea state coverage, robustness indicators, key response statistics, and construction parameters. This document was delivered to the owner for subsequent preliminary design and construction drawing refinement.

[0110] This embodiment demonstrates the engineering application path of the present invention under complex sea conditions and scour evolution, solving the problems of insufficient sea condition coverage, high computational cost, and inadequate coupling analysis in traditional methods, forming a closed loop from the sample domain to optimization and verification. Data shows that, under the same budget and hardware conditions, the synergy between mid-fidelity master evaluation and high-fidelity calibration can significantly improve optimization throughput and reduce reliance on full-scale high-fidelity simulation. In the sample domain with a quantile of 0.99, key indicators such as lift, local pressure, and displacement decrease, constraint satisfaction rate increases to 98.6%, fatigue damage decreases to 0.62, and the overall material and construction cost decreases to 95.2% of the baseline level. These results provide a verifiable, traceable, and engineering-deliverable formula for the wave-impact-resistant design of nearshore high-pile wharves in ports.

Claims

1. A method for optimizing the wave-impact resistance structure of a nearshore high-pile wharf in a port, characterized in that, Includes the following steps: S1. Based on historical observation data and reanalysis data, conditional sampling is performed according to the significant wave height, peak period, incident direction, storm surge, long-period energy ratio and water depth conditions of the target sea area to obtain a sea state sample set covering the design quantile level; the above step outputs the sea state sample set. S2. Establish a parametric geometric and working condition model for the nearshore high-pile wharf, determine the pile diameter, pile spacing, inclined pile ratio, pile cap size, beam height, deck opening ratio and opening arrangement, flange height, skirt length, geometric parameters and arrangement of energy dissipation components, deck elevation, berth direction and longitudinal and transverse arrangement, set boundary conditions and wave incidence and flow field conditions, and give multi-level scour scenarios and soil layer parameters; the above steps output a comprehensive parametric geometric and working condition model. S3. Establish a two-phase free surface high-fidelity solver, select either the smooth particle hydrodynamics solver or the phase field lattice Boltzmann solver, perform unsteady calculations on the sea state sample set and the integrated parameterized geometry and working condition model, and output the free surface evolution, local pressure time history, deck lifting force, pile top internal force and displacement response; the step outputs a high-fidelity response dataset. S4. Train the physical embedded neural operator surrogate model, taking sea state parameters and geometric parameters as inputs, and the high-fidelity response dataset as a supervision signal. Combine the physical residuals and boundary condition constraints to complete the training, and obtain the wave-structure response predictor for medium-fidelity evaluation calls; the output of the step is the medium-fidelity predictor. S5. Establish a fluid-structure-soil coupled calculation chain, couple the superstructure dynamic model with the pile-soil interaction model, use dynamic py characteristics or equivalent substructure method to represent pile-soil stiffness and damping, update pile-soil parameters under scour scenario, and iteratively solve the time history response of the whole structure based on the high-fidelity response dataset and the medium-fidelity predictor. The steps described above output a coupled time-history response dataset. S6. Define the set of optimization design variables as a bounded vector of the parameterized geometric and structural parameters. Define the set of objective functions including peak wave impact force, uplift force, peak displacement of deck and pile top, bending moment and shear force of key sections, fatigue cumulative damage and material and construction costs. Define the set of constraints including bearing capacity, stability, displacement limit, fatigue life, structural boundary and construction feasibility limit. The output of this step is the definition of the optimization problem. S7. Under the uncertainty of sea state, material parameters, soil parameters and construction deviation, tail risk limits are imposed on the objective function or the constraints based on conditional value of risk, confidence levels are set and robust evaluation sample set is generated; the steps output a robust optimization model and a robust evaluation sample set. S8. Multi-fidelity Bayesian optimization is used to iteratively search the design variables. The medium-fidelity predictor is used as the main evaluator, and the high-fidelity solution is called intermittently according to a preset strategy for calibration and convergence determination. Under the constraints and robustness requirements, the Pareto front and candidate solution set are obtained, and high-fidelity full-sample regression is performed on the candidate solutions for verification. The design result document is output, which includes the design variable values, sea state coverage area, robustness index, key response statistics and construction parameters. The step outputs the design result document.

2. The method as described in claim 1, characterized in that, Step S1 includes: S1.1 Collect historical observation data set D obs With reanalysis dataset D rea And merge to generate the basic sea state dataset D0; S1.2 Perform quality control, missing data interpolation, and resolution unification on D0 to obtain a normalized dataset D1; S1.3 Calculate the significant wave height H based on D1. s Peak period T p Incident direction θ0, storm surge Δη, long-period energy percentage E lp Construct a condition vector with water depth d. c=(H s ,,T p ,,θ0,,δη,,E lp ,,d); S1.4 is generated during the forward diffusion process. S1.5 utilizes a noise cancellation network F during the reverse generation process. θ Constructing conditional probabilities with energy conservation constraints p θ (x*0∣c); S1.6 sets the confidence quantile p*d, and for p θ (x*0|c) samples N times to obtain the sample set x*0 (i) *i=1 N ; S1.7 outputs the sea state sample set S*sea for use in subsequent steps.

3. The method as described in claim 1, characterized in that, Step S2 includes: S2.1 Set the pile foundation parameters and determine the pile diameter D sequentially. pile Pile spacing S pile Inclined pile ratio R incl ; S2.2 Set the parameters of the superstructure and determine the pile cap length L in sequence. cap Pile cap width B cap , Liang Gao h beam Deck opening ratio P deck A*deck is the aperture arrangement matrix. S2.3 Set the parameters of the protective components, and determine the edge plate height h*edge and skirt plate length l in sequence. skirt The geometric parameter vector g*diss of the energy dissipation component and its arrangement matrix S2.4 Set the overall layout parameters and determine the deck elevation z in sequence. deck berth direction θ berth and the C*layout matrix for vertical and horizontal layout; S2.5 generates a parametric geometric model M*geo and performs mesh generation, outputting a geometry and mesh dataset G. mesh ; S2.6 in G mesh Wave incidence conditions, flow field conditions, and soil boundary conditions are applied to form a parametric chemical condition model M. load ; S2.7 Based on the scour depth sequence d scour The pile-soil interaction parameters are updated along with the soil layer parameter matrix D*soil to generate a comprehensive parametric geometric and working condition model M*struct.

4. The method as described in claim 1, characterized in that, Step S3 includes: S3.1 Select either a differentiable smooth particle hydrodynamics solver or a phase-field lattice Boltzmann solver to establish a high-fidelity two-phase free surface solver G. hi ; S3.2 Set the water density ρ w air density ρ a Dynamic viscosity of water (μ) w Aerodynamic viscosity μ a Surface tension coefficient σ and gravitational acceleration vector g; S3.3 Import M struct With G mesh Apply boundary conditions to the solution domain; S3.4 Press S sea =x*0 (i) *i=1 N Load the incident field conditions, set the time step Δt and the total simulation duration T. sim ; S3.5 Calculate the velocity field u in each time step n. n Pressure field p n relative to the free liquid surface position η n ,satisfy where ρ∈ρ w ,ρ a τ is the viscous stress tensor; S3.6 Extract deck lift force Pile top internal force vector N*pile (i) (t), pile top displacement vector U*pile (i) (t) and local pressure time history S3.7 Output η (i) (t k )、 N*pile (i) (t), U*pile (i) (t).

5. The method as described in claim 1, characterized in that, Step S4 includes: S4.1 Concatenate the sea state parameter vector c with the structural geometry parameter vector g*struct to form the input vector x*in; S4.2 Constructing a Physically Embedded Neural Operator Model N θ And define physical residuals and boundary condition residual R b The physical loss L is obtained. phys =R p +R b ; S4.3 The supervised dataset output from step S3 Training N θ Constructing data loss S4.4 Connect L*data with L phys Linear weighting yields L total =w d L data +w p L phys , Where w d With w p Let L be the weighting coefficient, and minimize it through backpropagation. total Update N θ parameter; S4.5 output mid-fidelity waveform—structural response predictor P mid For subsequent use.

6. The method as described in claim 1, characterized in that, Step S5 includes: S5.1 Establishing the dynamic model M of the superstructure sup Its mass, stiffness, and damping matrices are M*sup, K*sup, and C*sup, respectively; S5.2 Establish a pile-soil interaction model M*ps, and use a dynamic py relationship for the horizontal stiffness curve k. py (z) and at the scouring depth d scour Next update; S5.3 will M sup With M ps Couple substructures to form an overall equation. S5.4 Based on the external load in step S3 and P in step S4 mid The results are solved using time-history iteration, outputting U*pile(t), N*pile(t), and U... deck (t).

7. The method as described in claim 1, characterized in that, Step S6 includes: S6.1 Define the design variable vector v=(D pile ,,S pile ,,R incl ,,L cap ,,B cap ,,h beam ,,P deck ,,h edge ,,l skirt ,,z deck ), And set the upper and lower bounds v*min, v*max; S6.2 Define the objective function vector J=(F peak ,,F deck,peak ,,U deck,peak ,,U pile,peak ,,M sec ,,V sec ,,D fat ,,C cost ); S6.3 Define the constraint set C = F ult ≥F peak ,,U lim ≥U deck,peak ,,D fat ≤D lim Geometric construction limits and construction feasibility limits; S6.4 Constructing a Multi-Objective Constrained Optimization Problem Make v∈[v*min,v*max] and satisfy C.

8. The method as described in claim 1, characterized in that, Step S7 includes: S7.1 Establishing a joint uncertainty distribution P ξ ξ covers sea state, material parameters, soil parameters, and construction deviations; S7.2 For each objective function J k Calculate Conditional Value at Risk Where q α For the corresponding quantiles; S7.3 Construct a partial Bruker optimization model with CVaR*α(J*k) as constraints, and generate a robust evaluation sample set S rob .

9. The method as described in claim 1, characterized in that, Step S8 includes: S8.1 P in step S4 mid Establish a Gaussian process agent GP and initialize the sample library D*0; S8.2 Select the desired improvement criterion EI(v) as the acquisition function and calculate... J * Let μ and σ be the current optimal values, respectively, and let Φ and φ be the predicted mean and variance of GP, respectively. S8.3 Under the condition of satisfying the robustness requirement of step S7, for the design variable v with the largest EI. * Call G*hi from step S3 to perform calibration and update the sample library D; S8.4 When the convergence criterion is satisfied, output the Pareto front set P and the candidate solution set V. * ; S8.5 for each v * ∈V * Perform high-fidelity full-sample regression validation to generate a design results document that includes design variable values, sea state coverage area, robustness indicators, and key response statistics.

Images