Ship route multi-objective optimization method and device for composite sea state scenario and electronic equipment

Abstract

The present application belongs to the field of ship traffic engineering, and relates to a ship route multi-objective optimization method and device for composite sea state scenarios, and an electronic device. The method comprises the following steps: identifying a composite sea state type and generating a high-resolution scenario field through multi-source meteorological data, establishing a multi-objective function containing safety risks and economic benefits and differentiated constraints, adaptively adjusting weights by using an improved evolutionary algorithm to search for a Pareto optimal solution set, and screening reliable schemes through diversity maintenance, physical simulation verification, historical mode verification and emergency margin testing; at the same time, relying on an online learning and incremental learning mechanism, continuously optimizing motion model parameters, generating network weights, predicting model parameters and local calibration coefficients. The present application has high-fidelity scenario description, adaptive dynamic weighting, multi-level safety protection and full-life-cycle evolution capability, and significantly improves the practicability and engineering reliability of the solution set.

Description

Technical Field

[0001] This invention relates to the field of marine traffic engineering technology, and in particular to a method, apparatus and electronic equipment for multi-objective optimization of ship routes under complex sea conditions. Background Technology

[0002] In existing technologies, intelligent planning of ship routes under complex sea conditions (multiple disaster factors superimposed, such as strong winds, giant waves, strong currents, and low visibility occurring simultaneously or alternately) mostly focuses on a single meteorological factor (such as considering only wind and waves) or a static environmental field, simplifying route optimization into a deterministic problem. This makes it difficult to cope with the complex wind and rain scenarios in real marine environments where multiple disasters are coupled and spatiotemporal changes are drastic.

[0003] The core shortcomings of existing technologies are reflected in the following aspects: Insufficient scenario representation capability: Traditional methods rely on coarse-resolution reanalysis data or single forecast products, which cannot accurately identify the evolution patterns of complex sea states (such as the superposition of typhoon leading-edge swells and local strong convection), resulting in distortion of the environmental input field and unreliable optimization basis.

[0004] The objectives and constraints are too simplistic: most optimization models only balance fuel consumption and sailing time, ignoring the differentiated risks such as hull rolling, wave impact, and propeller outburst under complex sea conditions. Moreover, the safety constraints are mostly fixed thresholds, lacking adaptability to different types of disasters (such as transverse wave resonance vs. top wave bottoming).

[0005] The weighting is subjectively fixed: the trade-off coefficient between risk and economic goals is usually preset by experience and cannot be dynamically adjusted according to the real-time situation and risk level. Under extreme conditions, it is easy to give an unsafe solution that is highly economical but also highly risky.

[0006] The models lack closed-loop updates: ship motion models, environmental generation models and forecast models are mostly calibrated offline and statically, and cannot be corrected online using measured data during navigation. This leads to the accumulation of errors over time, insufficient verification in historical extreme cases, lack of physical verification and emergency margin testing, and difficulty in ensuring reliability. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a multi-objective optimization method for ship routes under complex sea state scenarios, employing the following technical solution, including the following steps: Based on multi-source meteorological observation data, we identify and classify complex sea state event types and generate high-resolution meteorological scene fields. Establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and construct differentiated safety constraints under complex sea states; An improved multi-objective evolutionary algorithm is used to adaptively adjust the weights of risk and economic objectives based on the real-time identified scenario types, and to search for the Pareto optimal route solution set. The Pareto solution set is subjected to diversity preservation and sparsification processing to generate a set of optimal route options for decision-makers to choose from. Independent physical simulation verification, historical flight pattern verification, and emergency safety margin testing for extreme scenarios are conducted on candidate route schemes to eliminate unsafe schemes. The route plan is updated on a rolling basis based on the latest observation data, and the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model are continuously optimized through online learning and full life cycle incremental learning mechanisms.

[0008] To address the aforementioned technical problems, this invention also provides a multi-objective optimization device for ship routes under complex sea state scenarios, employing the following technical solution, including: The identification module is used to identify and classify complex sea state event types based on multi-source meteorological observation data, and generate high-resolution meteorological scene fields; The module is used to establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and to construct differentiated safety constraints under complex sea states; The adjustment module is used to adaptively adjust the weights of risk objectives and economic objectives based on the real-time identified scenario types using an improved multi-objective evolutionary algorithm, and to search for the Pareto optimal route solution set. The generation module is used to maintain diversity and sparsify the Pareto solution set to generate a set of optimal route options for decision-makers to choose from. The testing module is used to perform independent physical simulation verification, historical flight pattern verification, and emergency safety margin testing for candidate route schemes, and to eliminate unsafe schemes. The optimization module is used to continuously update the route plan based on the latest observation data, and continuously optimize the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model through online learning and full life cycle incremental learning mechanisms.

[0009] To address the aforementioned technical problems, the present invention also provides an electronic device that employs the technical solution described below, comprising a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the aforementioned multi-objective optimization method for ship routes under complex sea conditions.

[0010] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium, which employs the technical solution described below. The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the aforementioned multi-objective optimization method for ship routes under complex sea conditions.

[0011] Compared with the prior art, the present invention has the following main advantages: (1) It has the function of high-fidelity wind and rain complex scenario characterization: through multi-source observation fusion and conditional generation adversarial network, it realizes the type identification of complex sea state events and high-resolution scenario field reconstruction, which significantly improves the spatial details of environmental input and the accuracy of disaster coupling characterization.

[0012] (2) The adaptive dynamic weights have been optimized: the improved multi-objective evolutionary algorithm can dynamically adjust the risk and economic weights based on the disaster type identified in real time, so that the route solution set can achieve a scenario-adaptive balance between safety redundancy and economy, avoiding one-size-fits-all decision-making.

[0013] (3) It has multi-level safety assurance: It introduces a triple screening mechanism of independent physical simulation verification, historical navigation mode comparison and emergency margin test in extreme scenarios, and eliminates unsafe schemes from three dimensions of dynamics, statistics and extreme working conditions, which greatly improves the engineering credibility of the solution set.

[0014] (4) It has the ability to continuously evolve throughout the entire life cycle: It integrates online learning and incremental learning mechanisms to drive the ship motion model parameters, generator network weights, random forest forecast parameters and empirical relationship local coefficients to be updated with new observation data, so that the system has the adaptive ability to optimize as it is used, and its long-term timeliness is significantly better than that of static models.

[0015] (5) The solution set is highly practical: Through diversity maintenance and sparsification, it outputs Pareto front schemes with moderate size and obvious differences, which makes it easy for decision-makers to quickly select the best according to the cruise preference, taking into account both theoretical optimality and practical ease of operation. Attached Figure Description

[0016] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart of an embodiment of the multi-objective optimization method for ship routes under complex sea state scenarios of the present invention; Figure 2 This is a schematic diagram of a structure of an embodiment of the multi-objective optimization device for ship routes in complex sea state scenarios of the present invention. Figure 3 This is a schematic diagram of the structure of an embodiment of the electronic device of the present invention. Detailed Implementation

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings are used to distinguish different objects and not to describe a particular order.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] It should be noted that the multi-objective optimization method for ship routes under complex sea state scenarios provided in the embodiments of the present invention is generally executed by a server / terminal device, and correspondingly, the multi-objective optimization device for ship routes under complex sea state scenarios is generally set in the server / terminal device.

[0022] In this embodiment: MMG, Mathematical Modeling Group, is a method for modeling the hydrodynamics of ships in a separate manner.

[0023] GNSS, Global Navigation Satellite System.

[0024] PWV, Precipitable Water Vapor, is a type of vapor that can precipitate water vapor.

[0025] ERA5-Land, the fifth-generation European regional land surface reanalysis dataset.

[0026] LOF stands for Local Outlier Factor.

[0027] CFDP, Clustering by Fast Search and Find of Density Peaks, is a fast clustering algorithm based on density peaks.

[0028] cGAN stands for Conditional Generative Adversarial Network.

[0029] WGAN-GP, a Wasserstein generative adversarial network with gradient penalty.

[0030] PatchGAN is a block discriminator generative adversarial network structure.

[0031] NSGA-III, Nondominated Sorting Genetic Algorithm III, is the third generation of nondominated sorting genetic algorithm.

[0032] SBX, Simulated Binary Crossover.

[0033] PM stands for Polynomial Mutation.

[0034] TOPSIS, Technique for Order Preference by Similarity to IdealSolution, is a sorting method that approximates the ideal solution.

[0035] Cesium is an open-source 3D geospatial visualization engine.

[0036] Leaflet is a lightweight open-source 2D WebGIS map library.

[0037] DTW, Dynamic Time Warping.

[0038] MPC stands for Model Predictive Control.

[0039] RLS, Recursive Least Squares.

[0040] EWC, Elastic Weight Consolidation.

[0041] ASIC, Application Specific Integrated Circuit.

[0042] FPGA stands for Field-Programmable Gate Array.

[0043] DSP stands for Digital Signal Processor.

[0044] SRAM stands for Static Random Access Memory.

[0045] EEPROM, Electrically Erasable Programmable Read-Only Memory.

[0046] PROM stands for Programmable Read-Only Memory.

[0047] SMC stands for Smart Media Card.

[0048] SD stands for Secure Digital.

[0049] It should be understood that the number of terminal devices, networks, and servers is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used.

[0050] Example 1 Please refer to Figure 1 The diagram illustrates a flowchart of an embodiment of the multi-objective optimization method for ship routes under complex sea state scenarios according to the present invention. The multi-objective optimization method for ship routes under complex sea state scenarios includes the following steps: Step S1: Based on multi-source meteorological observation data, identify and classify composite sea state event types, and generate a high-resolution meteorological scene field.

[0051] In this embodiment, the electronic equipment (e.g., server / terminal device) on which the multi-objective optimization method for ship routes under complex sea state scenarios runs can receive ship route multi-objective optimization requests under complex sea state scenarios via wired or wireless connections. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods.

[0052] In this embodiment, step S1 may specifically include the following steps: S11 performs spatiotemporal matching, outlier detection and removal, missing value imputation, and adaptive filtering and smoothing on data from ground automatic weather stations, weather radars, and GNSS water vapor inversion or reanalysis to obtain a standardized meteorological dataset.

[0053] A two-stage data assimilation method combining spatiotemporal Kriging interpolation and random forest missing value imputation was employed. First, multi-source data were collected from the study area, including automatic weather stations (minute-by-minute wind speed, wind direction, gusts, temperature, humidity, and precipitation intensity), next-generation weather radar (combined reflectivity and radial velocity), and GNSS-PWV precipitable water vapor data (or ERA5-Land reanalysis data). Timestamps were unified (accurate to the minute level) and spatial coordinates were matched (unified to the WGS-84 projected coordinate system) for all data types. To address missing data, a random forest model was used to impute missing values ​​based on data from surrounding stations and historical data from the same period. For outliers, the Local Outlier Factor (LOF) algorithm was used for detection and removal. After quality control, the data was smoothed using adaptive Kalman filtering to reduce instrument noise and random fluctuations.

[0054] The purpose of step S11 is to ensure that all meteorological observation data entering subsequent models are consistent, continuous, and reliable, avoiding scenario misjudgments due to data quality issues. In ocean-going applications, the river sections are characterized by steep terrain, sparse meteorological station deployment, and limited spatial representativeness; therefore, multi-source data fusion is a necessary prerequisite for overcoming the limitations of single-source data.

[0055] S12, based on wind speed and precipitation intensity thresholds to trigger event detection, extracts multi-dimensional feature vectors including gust wind speed, average wind speed, rainfall intensity, wind direction standard deviation and wind speed rise rate, and uses an unsupervised clustering algorithm to classify events into pure strong wind type, pure heavy rainfall type and wind and rain composite type scenarios.

[0056] A Fast Clustering Based on Density Peak (CFDP) algorithm is constructed to perform unsupervised classification of quality-controlled meteorological time series. The basic unit of a complex sea state event is defined as a duration, with wind speed ≥10.8 m / s (threshold for force 6 wind) or hourly precipitation ≥20 mm as triggering conditions, and the simultaneous or sequential occurrence of both as the criteria for determining the complex event.

[0057] The clustering feature vector includes: maximum 3-second gust wind speed (Ugust), 10-minute average wind speed ( ), 10-minute rainfall intensity (R10), wind direction standard deviation (characterizing the degree of drastic change in wind direction), and wind speed rise rate (dU / dt, characterizing the sudden change in gusts).

[0058] The CFDP algorithm automatically identifies event boundaries, divides consecutive time intervals into independent events, and categorizes events into three scenario types based on feature similarity: C1 - pure strong wind type (wind speed exceeds limits, precipitation is weak), C2 - pure heavy rainfall type (precipitation exceeds limits, wind speed does not reach level 6), and C3 - wind and rain composite type (both exceed limits or occur sequentially). For type C3, it is further subdivided into wind speed-dominated subtype, precipitation-dominated subtype, and synchronous peak subtype.

[0059] The impact mechanisms of different scenario types on ship navigation safety are drastically different—C1 has wind heel moment as the main risk source, C2 has visibility reduction as the main risk source, and C3 requires a dual risk coupling assessment. Automatic classification ensures that the subsequent optimization model can adaptively adjust the objective function weights for different sea state types.

[0060] The rate of wind speed increase is defined as: .

[0061] Where: U(t) — instantaneous wind speed at time t, in m / s; Δt — Time step, taken as 1 minute (i.e. 60 s); dU / dt — The derivative of wind speed with respect to time, reflecting the severity of sudden changes in wind speed, measured in m / s. 2 It characterizes the severity of short-term sudden changes in wind speed, and its dimensions match the definition of acceleration. It is a key indicator for identifying sudden strong winds such as downbursts.

[0062] Standard deviation of wind direction Defined as: The output unit is °, and it is dimensionless before being used for clustering calculations.

[0063] in: — The wind direction at the i-th observation time, in degrees (°); — Average wind direction during the period, in degrees (°); N— Total number of observations within the time period, dimensionless; — Wind direction standard deviation, in degrees (°). The larger the value, the more turbulent the wind direction changes, the higher the uncertainty of the wind direction of the ship, and the greater the difficulty of maneuvering.

[0064] The purpose of step S12 is to discretize the continuous meteorological time series into scenario events with clear physical boundaries, providing a scenario-driven input pattern for route optimization.

[0065] S13 uses the classified event type labels as conditions to perform super-resolution reconstruction of the low-resolution meteorological forecast field, and outputs high spatiotemporal resolution wind speed, gust, wind direction and precipitation intensity forecast fields for key flight segment grid points in the study area.

[0066] Using a conditional generative adversarial network (cGAN), with the event type labels identified in step S12 as conditional variables, super-resolution reconstruction of the low-resolution (1 km / 1 h) meteorological forecast field is performed, and the objective fields of wind speed, gusts, wind direction and precipitation intensity of grid points (spatial resolution 250 m, temporal resolution 10 minutes) of key navigation sections of the research river are output.

[0067] The generator uses a U-Net structure, with inputs of an ERA5-Land coarse-grid forecast field and topographic factors (elevation, slope, and narrowing ratio), and outputs a high-resolution field. The discriminator uses a PatchGAN structure to determine whether the generated high-resolution field is consistent with the actual observation in terms of statistical distribution.

[0068] The training data consists of historical case sets from 2021 to 2025, and the WGAN-GP loss function is used to improve training stability and sample diversity. At the model output, a physical consistency constraint layer is added to force the output wind speed field to satisfy the mass conservation principle (acceleration on the windward slope and deceleration on the leeward slope) and the precipitation field to satisfy the vertical integral liquid water content constraint.

[0069] By learning the mapping relationship between historical high-resolution observations and low-resolution forecasts, cGAN can generate high-resolution scenario fields in real time with only coarse grid input, providing sufficient spatial granularity for flight path optimization.

[0070] The objective function of cGAN is: .

[0071] Where: G — Generator, which maps low-resolution input to high-resolution output; D — Discriminator, which determines whether the input image is a true high-resolution field; x — Input conditional variables, including low-resolution weather forecast field and topographic factors; y — Real high-resolution weather field (training labels); z — a random noise vector used to increase the diversity of the generated results; — The L1 loss weight coefficient, set to 100 (dimensionless), is used to balance the contributions of adversarial loss and L1 regularization term; The regularization term, in the units of meteorological field dimensions (e.g., wind speed in m / s), forces the generated results to be consistent with the actual values ​​at each grid point. — The mathematical expectation on the joint distribution (x, y) represents the average over all possible combinations of input conditions and the true high-resolution field; — The mathematical expectation on the joint distribution (x, z) represents the average over all possible combinations of input conditions and random noise; — The natural logarithm function is used to construct adversarial loss in the form of cross-entropy.

[0072] First item Encourage the discriminator to correctly identify real high-resolution fields; the second item Encourage the generator to deceive the discriminator; third item L1 constraints ensure the accuracy of the numerical values ​​at each grid point. The combination of these three factors ensures that the generated scenario is both statistically realistic and numerically accurate.

[0073] The purpose of step S13 is that the river section and canyon terrain in the ocean application scenario are complex, and the coarse grid forecast of the meteorological model cannot distinguish the local funneling effect and the orographic lifting effect.

[0074] The purpose of step S1 is to transform the chaotic multi-source meteorological monitoring data into a set of composite sea state scenarios with a clear structure and physical interpretability. Meteorological disasters in river sections used in the open ocean are often not caused by a single factor—sudden strong winds brought by downbursts are often accompanied by short-term heavy rainfall, forming wind-rain composite sea states; the funnel effect of canyon topography further exacerbates the spatiotemporal heterogeneity of the wind field.

[0075] Step S2: Establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and construct differentiated safety constraints under complex sea states.

[0076] In this embodiment, step S2 may specifically include the following steps: S21 defines the route safety risk as the cumulative integral of risk exposure over time periods along the entire route. The wind tilting moment ratio is calculated based on the forecast gust speed, the ship's windward area, the wind pressure coefficient, the wind pressure center height, and the wind direction angle. The visibility deviation is determined based on the ratio of the forecast precipitation intensity to the minimum safe operating visibility of the route segment. The comprehensive risk is the higher of the two.

[0077] The safety risk of a flight path is defined as the cumulative integral of the risk exposure of all segments along the entire path. For any candidate path... Parameterized as a function of time t Total risk objective function Defined as: .

[0078] in: — Route The total risk target value, in h (hours), represents the cumulative time equivalent of the weighted risk exposure; — Candidate tracks, two-dimensional spatial curves with time t as a parameter; t0 — Departure time, in hours; T — Estimated flight time, in hours; — A strong wind risk index (wind tilt moment ratio) updated hourly along the flight path, dimensionless, with a theoretical range of values. The safety threshold reference value is 0.95; — Visibility deviation updated hourly along the flight path, dimensionless, with a theoretical range of values ​​as follows: The safety threshold reference value is 0.35 (lower limit); — Spatial weighting factor, dimensionless, >1.0 for meandering river sections, bridge areas, and narrow navigation sections, and 1.0 for open and straight navigation sections; — Choose the larger of the two to reflect the weakest link effect, that is, the safety level of a flight route is determined by the weakest risk factor; t — Time integration variable, in hours, integrated along the route from t0 to t0 + T.

[0079] Strong wind risk index The calculation followed the definition given in the project background. However, it is extended to a spatiotemporal dynamic form: .

[0080] Where: s — spatial arc length coordinates along the flight path, in meters; t — Time coordinate, in seconds; f(s, t) — the ratio of wind tilting moments at the spacetime point (s, t), dimensionless; — Air density, taken as 1.225 kg / m³ 3 (Under standard atmospheric pressure and 15°C conditions); — The square of the predicted gust speed. The unit is m / s, therefore its square unit is m. 2 / s 2 ; — Projected area of ​​the ship above the waterline, in m² 2 ; — Wind pressure coefficient, dimensionless, depends on the shape of the ship's superstructure; — Vertical distance from the center of wind pressure to the point of hydrodynamic action, in meters; — The real-time angle between the forecast wind direction and the longitudinal profile of the ship, in degrees (°). It is dimensionless and takes values ​​[0, 1]. — Minimum capsizing moment of a ship under given loading conditions, expressed in N·m (Newton-meter).

[0081] Visibility Deviation The dynamic calculation is as follows: .

[0082] Where: d(s, t) — the visibility deviation at the spatiotemporal point (s, t), dimensionless; — Estimated atmospheric visibility due to precipitation, in meters; — Minimum visibility required for safe operation of this segment, in meters; R(s, t) — 10-minute average precipitation intensity, in mm / h; a, b — empirical coefficients defined locally, with the dimension of a being m·(mm / h). b b is a dimensionless index obtained by fitting historical precipitation-visibility synchronous observation data; — The negative b power of precipitation intensity, expressed in (mm / h)−b.

[0083] The purpose of step S21 is to directly embed the ship stability physical mechanism into the route optimization objective. Traditional optimization methods often simplify meteorological conditions into exogenous fixed parameters, while this formula achieves real-time coupling between the meteorological field and ship response through the dynamic calculation of f and d. The same wind field will produce completely different f values ​​for Class 5 and Class 6 restricted navigation ship types, thereby driving the optimization algorithm to generate differentiated routes for different ship types.

[0084] S22, establish a weighted economic objective function that includes the expected total flight time and the expected total fuel consumption. The expected total flight time is obtained by integrating the ratio of the range infinitesimal element to the ground speed, and the expected total fuel consumption is obtained by calculating the relationship between propulsion power and speed and wind-induced additional drag correction.

[0085] Define the economic objective function Weighted combination of flight time and fuel consumption: .

[0086] in: — Route The economic target value is a weighted composite indicator of time and fuel consumption, and its dimensions depend on... The processing method (here used as a decision indicator, which can be compared after dimensionless processing); — Weighting coefficients, dimensionless. The default value is 0.6, which reflects the priority of timeliness, and users can adjust it; — Estimated total sailing time, in hours; — Estimated total fuel consumption, in tons.

[0087] The travel time is calculated by relating the range element ds to the ground speed. The integral of the ratio is obtained as follows: .

[0088] Where: ds — range element, in meters; — The speed of the ship relative to the ground at position s, in m / s; — Total sailing time, in seconds (can be converted to hours).

[0089] Ground speed Water flow velocity and ship speed in still water Common impacts: .

[0090] in: — Ship speed over land, in m / s; — The speed of a ship in still water (determined by the engine speed), measured in m / s; — Water flow velocity (positive value in the downstream direction), unit: m / s; — The angle between the heading and the direction of the water flow, expressed in degrees (°). Dimensionless, downstream Against the current .

[0091] Fuel consumption The relationship between propulsion power and speed is expressed as a cube: .

[0092] in: — Total fuel consumption, in kg (can be converted to t); P(s) — Ship propulsion power, in W (i.e., kg·m² / s³). — Propulsion efficiency, dimensionless, typically taken as 0.55–0.65; Q — Calorific value of fuel oil, in J / kg (i.e., m³ / kg fuel oil). 2 / s 2 Marine diesel fuel is typically taken as 42,700 kJ / kg = 4.27 × 10⁻⁶ kJ / kg. 7 J / kg; k — The drag coefficient related to the ship type, in the dimension of kg / m, which needs to be calibrated through ship model tests or CFD calculations; — The cube of the speed in still water, in meters (m) 3 / s 3 .

[0093] In strong winds, ships need to increase rudder angle compensation to maintain course, which increases drag, thus introducing a wind resistance correction term: .

[0094] in: — Total propulsion power after considering wind resistance, in watts (W); — Static water resistance power, in watts (W); — Wind pressure (force), dimensionless, kg·m / s 2 = N, where The average wind speed at a height of 10 m is expressed in m / s. — Speed ​​component, in m / s.

[0095] The purpose of step S22 is to establish a quantitative measurement standard for route efficiency. (Trinomial) The economic model allows dispatchers to flexibly adjust according to operational strategies (rushing shifts or saving costs). The wind resistance correction term ensures that, under strong wind conditions, the optimization algorithm will not choose a dangerous route that braves strong winds in pursuit of the shortest time, because high wind speeds will cause a surge in fuel consumption, which will be automatically penalized in the objective function.

[0096] S23, establish upper and lower limits of hard constraints for the wind tilt moment ratio and the visibility deviation, establish hard constraints for the safe distance of the waterway boundary, and establish coupled constraints for the combined wind and rain scenario, requiring that the joint index of the wind tilt moment ratio and the visibility deviation does not exceed a predetermined threshold.

[0097] Constraints are divided into hard constraints (which must be met, otherwise the route is infeasible) and soft constraints (which allow for a certain degree of violation, but are penalized). Hard constraints include: ; ; Channel boundary constraints: , .

[0098] Where: f(s, t) — wind tilting moment ratio, dimensionless, with an upper limit of 0.95 corresponding to a safety margin close to the overturning criticality; d(s, t) — Visibility deviation, dimensionless, lower limit 0.35, corresponding to the minimum visibility threshold required for safe operation; — Indicates the track All spatial positions on the surface are valid; — This indicates that the condition holds true for the entire time interval from departure to arrival; — Channel boundary line (two-dimensional curve); — The shortest distance from track point s to the channel boundary line, in meters; — Safety distance, taken as 3 times the width of the ship, in meters.

[0099] Soft constraints are: .

[0100] in: — Appeared along the entire route The probability is dimensionless; — Appeared along the entire route The probability is dimensionless; — The upper limit of the permissible probability for high-risk strong winds is set at 0.05, dimensionless; — The upper limit of the permissible probability for high risk in low visibility, taken as 0.05, dimensionless.

[0101] In addition, coupling constraints are added for the combined wind and rain L3 scenario: When the L3 scenario is activated.

[0102] Where: f(s, t) — wind tilting moment ratio, dimensionless, with values ​​ranging from... The typical safe range is [0, 0.95]. d(s, t) — Visibility deviation, dimensionless, values... The typical safe range is [0.35, 1.0]. 1-d— When hour, Dimensionless; The product f · (1-d) — dimensionless, with an upper limit of 0.35, which is the calibrated critical value for composite risk.

[0103] The physical meaning of this constraint is that under the combined effects of wind and rain, even if a single risk does not reach the threshold, the coupling effect may produce a nonlinearly amplified safety threat. Therefore, the combined index of the two factors must not exceed the critical value. The derivation of this constraint is based on the inversion conclusion of the Eastern Star incident—at that time, the f value had exceeded 0.90, while rainfall caused visibility to approach zero. The coupling of the two factors caused the crew to completely lose their emergency maneuverability.

[0104] Step S23 elevates safety from an objective to a hard constraint, ensuring that the optimized flight path will not exceed the safety red line at any point in time and space. In particular, the L3 coupling constraint is a unique contribution of this method—traditional optimization methods often handle wind and rain separately, neglecting the nonlinear composite effect when both act simultaneously. The constraint, in the form of f · (1-d), results in a product close to 1 when f is close to 1 (strong wind) and d is close to 0 (zero visibility), far exceeding the calibration threshold of 0.35, and will be automatically rejected by the optimization algorithm.

[0105] The purpose of step S2 is to transform the engineering problem of route optimization into a rigorous multi-objective mathematical programming problem. Ship route optimization inherently involves multiple conflicting objectives—the shortest sailing time (economic efficiency) and the lowest risk exposure (safety) are often mutually exclusive. This step uses three-layer modeling to define safety risk objectives, efficiency and economic objectives, and differentiated risk constraints under complex sea conditions, providing a clear optimization direction for the solver in step S3.

[0106] Step S3: An improved multi-objective evolutionary algorithm is used to adaptively adjust the weights of risk objectives and economic objectives based on the real-time identified scenario types, and to search for the Pareto optimal route solution set.

[0107] In this embodiment, step S3 may specifically include the following steps: S31 employs the third generation of the non-dominated sorting genetic algorithm as its optimization engine. Through adaptive reference point generation, constraint dominance priority strategy, and population initialization strategy based on scenario switching, it searches for the Pareto optimal route solution set that satisfies the constraints.

[0108] An improved NSGA-III (third generation of non-dominated sorting genetic algorithm) is adopted as the main optimization engine. Compared with the classic NSGA-III, the improvements of this method are reflected in three aspects: Adaptive Reference Point Generation: Traditional NSGA-III uses reference points that are fixedly distributed in the high-dimensional target space. However, when the Pareto front has an irregular shape, fixing the reference points can lead to uneven distribution of the solution set. This method uses reference points calculated in step S2... and Given the current value range, K-means clustering is used to dynamically generate reference points in the target space, so that the distribution of reference points adaptively matches the shape of the front.

[0109] Constraint Domination Priority Strategy: In non-dominated ranking, individuals that violate hard constraints are placed with the lowest priority, even if their objective function value is excellent, and they will not participate in the domination ranking. Specifically, this is implemented by defining the constraint violation degree. Divide individuals into feasible solutions and infeasible solutions There are two categories, and during sorting, all feasible solutions take priority over infeasible solutions.

[0110] Population initialization strategy based on weather scenario switching: The scenario type (S1 / S2 / S3) identified in step S1 serves as prior knowledge to guide the generation of the initial population. For scenario C1, the initial population is biased towards flight segments with lower wind speeds and downwind directions; for scenario C2, the initial population is biased towards areas with weaker precipitation intensity; for scenario C3, the initial population needs to consider both wind speed and precipitation dimensions.

[0111] Chromosome coding: Each chromosome represents a candidate route, encoded as a sequence of critical waypoints. ,in As the starting point, For destination, intermediate waypoint The route is continuously variable within navigable waterways. Cubic spline interpolation is used to smoothly connect discrete waypoints into a continuous curve, ensuring that the radius of curvature of the route is not less than the minimum turning radius of the ship (usually 3-5 times the ship's length).

[0112] Crossover and mutation operators: Simulated binary crossover (SBX) and polynomial mutation (PM) are employed, both of which preserve the structural characteristics of the parent solution. Crossover probability. Probability of mutation (m is the number of waypoints). Population size Maximum number of generations .

[0113] The purpose of step S31 is as follows: The advantage of NSGA-III lies in its ability to maintain the diversity of solution sets in multiple directions simultaneously, avoiding getting trapped in local optima. The improved reference point generation mechanism ensures that the solution set is evenly distributed in the target space, preventing situations where some regions have overly dense solution sets while others have no solutions. The constraint dominance priority strategy guarantees that all output routes are physically safe and feasible—a fundamental principle that cannot be compromised in the life-or-death route optimization problem.

[0114] S32 dynamically adjusts the priority weights of risk objectives and economic objectives in the optimization process based on the real-time identified scenario types, and avoids abrupt weight changes at scenario boundaries through a smooth transition function.

[0115] Design a weight adaptive adjustment module to dynamically adjust the priority of objectives in multi-objective optimization based on the scenario type identified in real time in step S1. The basic logic is as follows: Scenario C1 (Strong Wind Type): , ; Scenario C2 (Pure Heavy Rainfall): , ; Scenario C3 (Combined Wind and Rain): , ; in: — Risk Objectives The weighting coefficients in the optimization are dimensionless and range from [0, 1]. — Economic objectives The weighting coefficients in the optimization are dimensionless and satisfy... .

[0116] Weight adjustment is not a simple fixed switch, but rather introduces a sigmoid smooth transition function to avoid oscillations in optimization results caused by weight jumps at scenario boundaries: .

[0117] in: — Risk weight at time t, dimensionless; — Baseline weight, set to 0.6, dimensionless; — Scenario-driven increments, dimensionless: 0.3 for composite types, 0.15 for pure strong wind types, and 0.1 for pure rainfall types; — Sigmoid function Dimensionless; — The meteorological feature vector at time t, which includes elements such as wind speed and rainfall intensity, with different dimensions (dimensionless after standardization). — Scenario switching threshold vector, and Same dimension, corresponding units; — Euclidean distance, dimensionless after standardization; — The transition band width parameter controls the smoothness of the weight change and is dimensionless after standardization.

[0118] Step S32 serves the purpose of adapting weights, which is the core manifestation of the dynamic attributes of this method. Fixed-weight multi-objective optimization, when facing dynamically changing complex sea conditions, produces inefficient solutions—a route generated in the early stages of strong winds may prioritize economy, but the risk of the same route increases dramatically as wind speeds rise. Through scenario-driven real-time weight adjustment, the optimization algorithm can proactively shift its focus to safety when weather conditions worsen, achieving intelligent decision-making that adapts to changing circumstances and considers weather conditions.

[0119] S33 calculates the similarity between the current scenario and historical scenarios, extracts elite individuals from the Pareto front solution set of similar historical scenarios as prior seeds, mixes them with randomly generated individuals to form the initial population, realizes cross-scenario knowledge transfer and accelerates optimization convergence.

[0120] A transfer learning strategy is employed, using the Pareto front solution set obtained from optimization in similar historical scenarios as the prior seed for the initial population in the current scenario. The specific method is as follows: (1) Context similarity calculation: Define the current context With historical context Similarity: .

[0121] in: — The similarity between the current scenario and the historical scenario, dimensionless, with values ​​(0, 1]. — The feature vector of the current situation includes statistics such as average wind speed, maximum gust, average rainfall intensity, and wind direction dispersion. The dimensions of each dimension are different (dimensionless after standardization). — Feature vectors of historical scenarios, dimensions and The same, dimensionless after standardization; — The squared Euclidean distance between two eigenvectors, which is dimensionless after standardization; — Bandwidth parameter controls the rate at which similarity decays with distance; after standardization, it is dimensionless.

[0122] (2) Knowledge Transfer: Extract the Pareto front solution set from the top K historical optimization tasks with the highest similarity to the current scenario. A select few individuals constitute the migrating population. .

[0123] (3) Mixed population initialization: final initial population ,in The ratio of randomly generated new individuals to the two 30%, It accounts for 70%. This proportion was determined through experimental verification—an excessively high proportion of migrating individuals will inhibit the ability to explore new solution spaces, while an excessively low proportion will result in an insignificant migration effect.

[0124] The purpose of step S33 is that strong winds / heavy rainfall events in river sections used in ocean-going applications exhibit seasonality and weather-related recurrence (such as the annual plum rain season and the midsummer convection period), making historical optimization experience highly valuable for reuse. Transfer learning allows the current optimization task to start from scratch, building upon historical best solutions, thus increasing the convergence speed by approximately 40%–60% (based on preliminary simulation data from this project). This is crucial in real-time navigation management—when weather changes abruptly, the system needs to provide new route suggestions within minutes, rather than waiting for hours of evolutionary computation.

[0125] Step S3 aims to efficiently search for the optimal or near-optimal route solution set based on the mathematical model established in Step S2. The difficulty of multi-objective optimization problems lies in the fact that there is no single optimal solution among the objective functions; instead, there exists a set of Pareto optimal solutions that weigh the differences within the objective space. This step improves search efficiency and solution set quality through a three-layer adaptive mechanism.

[0126] Step S4: Perform diversity maintenance and sparsification processing on the Pareto solution set to generate a set of optimal route options for decision-makers to choose from.

[0127] In this embodiment, step S4 may specifically include the following steps: S41 employs an adaptive grid clustering method to sparsify the final non-dominated solution set, retaining representative solutions in each non-empty cell and interpolating virtual solutions to control the size of the solution set within a preset range while maintaining the continuity of the Pareto front.

[0128] After the NSGA-III evolution in step S3 is completed, the final non-dominated solution set is obtained. (The size is typically 50-100 individuals). However, some individuals are very close in the target space, constituting redundant solutions; others are too sparsely distributed in the target space, resulting in insufficient choices for the decision-maker. This step uses an adaptive grid clustering method to... Perform sparsification: (1) Target space Divided into adaptive meshes, number of meshes Automatically determined based on the solution set distribution density: .

[0129] in: — The number of grid divisions in each target dimension, a dimensionless integer; — The number of individuals in the final non-dominated solution set, a dimensionless integer; — The expected average grid density, i.e. the expected number of solution individuals in each grid, is taken as 5 and is dimensionless; — The round-up function.

[0130] (2) In each non-empty cell, only one representative solution is retained—the individual closest to the geometric center of the cell.

[0131] (3) For empty spaces (adjacent regions without solutions), linear interpolation is used to fill in virtual solutions to ensure the continuity of the Pareto front.

[0132] The size of the solution set after sparsification is controlled within This approach retains the complete form of the Pareto frontier while eliminating redundant information, facilitating verification and display in downstream steps.

[0133] Step S41 serves the purpose of Pareto front sparsification, a crucial bridge connecting optimization algorithms and engineering decisions. Providing decision-makers with more than 30 alternatives is impractical in real-world navigation management—captains and dispatchers cannot review them all within a limited timeframe. Sparsification reduces the solution set to a scale manageable by the human brain, while gridding ensures that the retained solutions are evenly distributed in the target space, avoiding the omission of important trade-offs.

[0134] S42, calculate the distance between each solution in the sparsified Pareto solution set and the positive and negative ideal solutions, determine the relative proximity and sort them in descending order, and output the recommended solution ranking based on user preferences.

[0135] Based on the sparsified Pareto solution set, TOPSIS (Topology for Ideal Solutions) is used to provide decision-makers with recommended rankings. The basic idea of ​​TOPSIS is that the optimal solution should be closest to the positive ideal solution and furthest from the negative ideal solution at the same time.

[0136] The ideal solution is defined as follows: and (That is, both objectives are to be minimized, but in reality, both are usually mutually exclusive). The negative ideal solution is defined as... and .

[0137] For the i-th individual in the Pareto solution set, its Euclidean distance to the positive ideal solution is: .

[0138] Distance from the negative ideal solution Similarly, the relative similarity is: .

[0139] in: — The Euclidean distance from the i-th solution to the ideal solution, dimensionless (after normalization). — The Euclidean distance from the i-th solution to the negative ideal solution, dimensionless; — The risk target value for the i-th option, in h; — The risk target value of the ideal solution (the minimum value among all options), in h; — Risk target value of the negative ideal solution (maximum value among all options), in h; — The maximum and minimum values ​​of the risk objective in the Pareto solution set, in h; — The economic target value of the i-th scheme (dimensions as defined); — The economic objective value of the ideal solution (the minimum value among all options); — The economic objective value of the negative ideal solution (the maximum value among all options); — The relative similarity of the i-th scheme, dimensionless. The larger the value, the better the i-th solution.

[0140] according to Sort all Pareto solutions in descending order, and the one ranked first is the TOPSIS recommended solution.

[0141] Building on this, user preference interaction functionality will be added: decision-makers can adjust settings via a slider. The system updates the TOPSIS ranking in real time based on the timeliness-economic weight in step S2, providing recommended solutions for different preferences such as "I'd rather consume more fuel to arrive earlier" or "I'm not in a hurry but want the safest option."

[0142] Step S42 serves to provide TOPSIS with an objective, transparent, and reproducible ranking basis for solutions, eliminating the subjective bias of decision-makers when faced with multiple Pareto solutions. The integration with user preference interactions allows the method to adapt to the actual needs of different navigation management scenarios—for example, tourist passenger ships have strict schedule restrictions and may prefer time-priority solutions; while dangerous goods transport ships must choose the safest option.

[0143] S43 overlays a meteorological risk layer onto the selected route plan in a 3D geospatial scene for visualization rendering, and automatically extracts recommended heading, recommended speed, estimated arrival time, and maximum heel angle parameters for each segment from the optimal route plan for direct use by the ship's navigator.

[0144] Develop a WebGIS 3D visualization module based on the Cesium or Leaflet map engine to render the selected optimal flight path in a 3D terrain scene. The displayed content includes: (1) Route geometry: superimposed on the high-precision DEM topography of the river section in the ocean use scenario, and the f value of different sections is represented by colored lines (red for high risk, yellow for medium risk, and green for low risk).

[0145] (2) Dynamic wind field arrows: The high-resolution wind field vector output in step S1 is superimposed around the flight path, allowing decision-makers to see intuitively which locations on the flight path face crosswinds (high risk) and which locations face tailwinds (low risk).

[0146] (3) Visibility contour lines: The spatial distribution of visibility reduction caused by precipitation intensity is superimposed, and the visibility level is represented by a transparency layer.

[0147] At the same time, navigation parameters for direct use by the ship's navigator are automatically extracted from the optimal route plan: Recommended course for each leg: (True heading, unit: degrees); Recommended speed for each leg of the journey: (Unit: Section) Estimated arrival times at each waypoint: (Unit: h); Maximum heel angle throughout the entire range: (Unit: degree), the calculation formula is: .

[0148] in: — Ship's heel angle, in degrees (°); — Dimensionless; — Wind tilt moment, in N·m, derived from step S2 Calculated; GM—Meltdown height, in meters, reflects the inherent stability of a ship against heel; — Discharge volume, in tons (t), 1 t = 1000 kg; g is the acceleration due to gravity, and its value is... .

[0149] The parameter of maximum heel angle throughout the journey can help the captain predict the maximum degree of tilt of the ship in the most dangerous section of the route.

[0150] The purpose of step S43 is to transform abstract numerical optimization results into intuitive geospatial understanding through 3D visualization, enabling the captain and dispatchers to clearly see the spatial distribution of route risks at a glance. The automatically extracted navigation parameters directly serve the bridge operations, seamlessly converting optimization results into navigation commands and shortening the time lag between decision-making and execution.

[0151] Step S4 transforms the non-dominated solution set generated by the multi-objective evolutionary algorithm into a set of route options that can be directly used by decision-makers. Unlike single-objective optimization, which outputs a single optimal solution, multi-objective optimization outputs a set of solutions—each of which carries a different trade-off between safety and economy. This step, through three layers of processing, ensures that the generated set of route options possesses diversity, stability, and engineering practicality.

[0152] Step S5 involves conducting independent physical simulation verification, historical flight mode verification, and emergency safety margin testing for candidate route schemes, eliminating unsafe schemes.

[0153] In this embodiment, step S5 may specifically include the following steps: S51 uses a hull-propeller-rudder separation modeling method to perform time-domain simulation of the six degrees of freedom motion of the ship on the candidate route. If the roll angle exceeds the limit, the rudder effect is lost, or the heading deviation exceeds the correction capability during the simulation, the route plan is determined to be unsafe.

[0154] Six-degree-of-freedom motion refers to the full-dimensional motion of a ship, including pitch, sway, heave, roll, pitch, and bow. The MMG (Mathematical Modeling Group) separate modeling method is employed to perform high-precision time-domain simulations of the ship's six-degree-of-freedom motion along candidate routes. The MMG model models the hydrodynamic forces acting on the hull as three independent parts: hull, propeller, and rudder, providing higher physical accuracy and parameter adjustability compared to a monolithic model.

[0155] The equations of motion for a ship in the horizontal plane with three degrees of freedom (pitch, sway, and yaw) are as follows: ; ; .

[0156] Where: m — ship mass, in kg; m x , m y — The additional mass of the ship in the pitching and swaying directions, in kg; I zz — The moment of inertia of a ship about its z-axis (vertical axis), expressed in kg·m. 2 ; J zz — Additional moment of inertia about the z-axis, in kg·m 2 ; u, v — the ship's pitching speed (in the direction of the bow) and swaying speed (in the lateral direction), in m / s; r — Bow roll rate, in rad / s; — Sway acceleration, roll acceleration, and pitch acceleration, in m / s². 2 m / s 2 rad / s 2 ; X H Y H, N H — The hydrodynamic forces and yaw moments of the hull in the longitudinal and transverse directions, respectively, in N, N, and N·m; X P Y P , N P — The propeller force and torque, in units of N, N, and N·m respectively; X R Y R , N R — Rudder force and torque, in units of N, N, and N·m respectively; X wind Y wind , N wind — Wind force and torque, in units of N, N, and N·m respectively.

[0157] Wind force and torque are calculated in real time based on the high-resolution wind field output in step S1: ; ; .

[0158] in: — Wind-induced swell, sway, and pitching moment, in units of N, N, and N·m, respectively; — Air density, taken as 1.225 kg / m³ 3 ; — The resultant velocity of the wind relative to the ship, in m / s; — Resultant velocity squared, in meters 2 / s 2 ; — Projected area of ​​the ship above the waterline, in m² 2 ; — Overall length of the vessel, in meters; — Swell coefficient, roll coefficient, and yaw moment coefficient, dimensionless, relative to wind direction angle. The function is pre-calibrated by wind tunnel testing or CFD calculation; — Relative wind angle (the angle of the wind relative to the direction of the ship's bow), in degrees (°).

[0159] The equation of motion for roll (fourth degree of freedom) is: .

[0160] Where: Ixx — the moment of inertia of the ship about the x-axis (longitudinal axis), in kg·m 2 ; J xx — Additional moment of inertia about the x-axis, in kg·m 2 ; — Roll acceleration, in rad / s 2 ; — Roll damping ratio, dimensionless; GM — Initial stability height, in meters; — Discharge volume, in kg; — Wind-induced roll moment, in N·m; — Wave disturbance moment, in N·m, is calculated from the wave spectrum of the river section.

[0161] Simulation time step Use 0.1 seconds to simulate the ship's motion response during the entire journey along the candidate route. If the following occurs during the simulation: Roll angle (Maximum safe roll angle for passenger ships); Rudder effect loss (rudder blade outflow velocity) ); heading deviation If the correction fails, the route plan is deemed unsafe and eliminated.

[0162] The purpose of step S51 is to provide a physical second opinion, independent of the optimization algorithm, for route safety verification through MMG simulation. The f-index in the optimization algorithm is a static stability assessment, while MMG simulation can capture dynamic instability phenomena—such as parametric roll, loss of rudder effectiveness, and other nonlinear and transient effects. This dual verification mechanism of static assessment and dynamic simulation is a key guarantee for ensuring the absolute safety of the route plan.

[0163] S52 retrieves historical tracks with similar shapes to the candidate route from the historical AIS database, extracts the actual navigation status of similar tracks under corresponding weather conditions, and issues a navigability warning if most historical tracks choose to decelerate or drift, or have records of running aground, or have no records of passing through.

[0164] Using the historical AIS data accumulated in steps S1 and S2, trajectory similarity retrieval is performed on the historical flight records of candidate routes under similar weather conditions. A route similarity metric is defined. .

[0165] in: — Similarity between candidate routes and historical routes, dimensionless, with values ​​(0, 1]. — The candidate flight path trajectory (output of step S4) is a two-dimensional spatial sequence; — Historical AIS trajectories are two-dimensional spatial sequences; DTW — Dynamic Time Warped Distance, used to measure the similarity in shape between two trajectories (allowing for stretching and deformation on the time axis), with the dimension in meters (since the coordinates of trajectory points are in meters, the cumulative distance of DTW is in meters). — Bandwidth parameter, in meters, controls the rate at which similarity decays with DTW distance.

[0166] Retrieve the top M historical tracks (M = 30) with the highest shape similarity to the candidate route from the historical AIS database, and extract the actual speed, rate of change of course, and navigation status (normal navigation / deceleration / drifting / berthing) of these historical tracks under the corresponding weather conditions. If any of the following conditions are met, issue a navigability warning for the candidate route: (1) More than 70% of similar historical trajectories chose to slow down or drift at that time (indicating that the route would be difficult to navigate under similar weather conditions); (2) Any similar historical trajectory records an event of loss of control or grounding (indicating a known safety hazard on the route); (3) Of the historical segments traversed by the candidate route, more than 30% of the segments had no record of any ships passing through under similar weather conditions in the past (indicating that there is a blind spot in the knowledge of the route).

[0167] The purpose of step S52 is to serve as a bridge between physical simulation and actual navigation experience, as AIS behavior pattern verification acts as a means of verifying the behavior patterns of real ships. While physical simulation offers high accuracy, it struggles to encompass all complex real-world factors (such as localized eddies, shore wall effects, and interference from other vessels). Historical AIS data records the actual performance of real ships under the combined influence of these factors, providing invaluable supplementation and verification to physical simulation. The combination of these two elements forms a complete safety verification loop integrating theoretical physics and empirical experience.

[0168] S53. Assuming that weather conditions suddenly deteriorate beyond the forecast range during navigation, calculate the emergency safety margin at each location on the candidate route and test whether the emergency reachability time from any point on the route to the nearest emergency shelter anchorage meets the preset response time limit requirements.

[0169] Extreme scenario stress tests were conducted on the flight path schemes that passed the first two layers of verification. It was assumed that during flight path execution, weather conditions suddenly deteriorated beyond the forecast range (e.g., gust wind speeds). The speed increased rapidly from 10.8 m / s to 25 m / s within 10 minutes, testing whether the vessel could remain safe under emergency operations (full rudder turn, emergency deceleration, and berthing at the nearest berth).

[0170] Define emergency safety margin: .

[0171] in: — Emergency safety margin, dimensionless, can be negative (negative value indicates that the overturning criticality has been exceeded). — Minimum capsizing moment of a ship under given loading conditions, in N·m; — Wind tilting moment at position s along the flight path under extremely adverse conditions, in N·m; — Take the minimum value along the entire route, which is the safety margin for the most dangerous position.

[0172] Dimensionality Consistency: Both the numerator and denominator are N·m, and the ratio is dimensionless.

[0173] Simultaneously test the accessibility of emergency shelters: calculate the sailing time from any point on the route to the nearest safe emergency shelter anchorage. .

[0174] in: — Emergency navigation time from the route position s to the nearest emergency shelter anchorage, in h (or min). — Study the known set of emergency shelter anchorages within the river section; — The distance along the shipping channel from the current position s to the anchorage p, in km; — Emergency navigation speed, in km / h, is 1.2 times the minimum maneuvering speed of the vessel type; — Find the minimum sailing time among all anchorages.

[0175] Require If the emergency response time limit is less than 1 minute (recommended value by the Three Gorges Navigation Administration), the route is considered to lack sufficient safety and risk avoidance capabilities in extreme scenarios.

[0176] Step S53 serves as the final embodiment of this method's application to complex sea state attributes. The core threat of complex sea states lies in their suddenness and cumulative nature—downbursts can cause wind speeds to jump from Force 5 to Force 8 or higher within minutes, accompanied by visibility dropping to almost zero. The optimization algorithm generates routes based on forecast fields, but forecasts always contain uncertainties. Emergency safety margin testing ensures that even with significant forecast deviations, the route still possesses a last line of defense—the vessel can safely evacuate or hold out until the weather improves. This step elevates route optimization from static path planning to dynamic safety and survivability design.

[0177] Step S5 serves to perform a third-party safety verification of the route plan generated in step S4, independent of the optimization algorithm. While the optimization algorithm itself contains risk objective functions and constraints, it is essentially an approximate calculation based on a simplified physical model and may contain unforeseen safety risks. This step, through dual verification using three layers of physical simulation and navigation experience, ensures that the final route plan is safe and reliable in actual navigation.

[0178] Step S6: Based on the latest observation data, the route plan is updated on a rolling basis. Through online learning and full life cycle incremental learning mechanisms, the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model are continuously optimized.

[0179] In this embodiment, step S6 may specifically include the following steps: S61 adopts a model predictive control framework, taking the ship's current position as a new starting point, and performs replanning within the rolling optimization time window in combination with the latest short-term weather forecasts. It also sets an event triggering mechanism based on the real-time changes in wind speed or precipitation intensity, and immediately triggers replanning when the changes exceed the threshold.

[0180] A model predictive control (MPC) framework is used to implement rolling time-domain replanning. During actual ship navigation, at each interval... (Use 10 minutes, consistent with the weather forecast update frequency in step S1) Execute the following replanning loop: (1) Get the current time The latest meteorological observations and short-term forecasts (Latest output of the model in step S1).

[0181] (2) Based on the ship's current actual position A new beginning, with the original destination As the endpoint, within the time window The complete optimization process from step S2 to step S4 is executed internally, where The length of the rolling time domain (taken as 1 hour).

[0182] (3) Only replanning routes are implemented. Inland segments ( (Take 20 minutes), after execution, in This could trigger a replanning process at any moment.

[0183] The replanning trigger condition is based not only on time, but also on the event trigger mechanism: if real-time observation data indicates wind speed exist The change within 5 minutes exceeds the threshold. (Take 3 m / s), or the change in precipitation intensity R exceeds If the speed is 10 mm / h, then replanning will be triggered immediately, without waiting for the next fixed time window.

[0184] Step S61 serves to address the paradox of weather forecast timeliness through rolling time-domain replanning—0-1 hour forecasts offer the highest accuracy but have a short lead time, while 3-hour forecasts offer a long lead time but have low accuracy. By updating every 10 minutes, the system always makes decisions based on the latest and most accurate short-term forecasts, while maintaining the ability to predict changes within the next hour through rolling time-domain updates. The event triggering mechanism ensures a rapid response to sudden strong winds (such as downbursts), avoiding potential delays in critical reaction time that might occur with fixed-period updates.

[0185] S62 compares the ship response data observed during actual navigation with the simulation prediction values ​​of the MMG split ship motion model, uses the recursive least squares algorithm to update the wind pressure coefficient and roll damping ratio in the model online, and stores the corrected parameters in the ship type parameter knowledge base.

[0186] The ship response data (actual speed) observed during actual navigation Actual roll angle Actual wind tilt moment The simulation prediction values ​​of the MMG-separated ship motion model in step five are compared with those of the model in step five, and the key parameters of the model are corrected online.

[0187] Define prediction error: .

[0188] in: — Speed ​​prediction error, in m / s; — Actual observed speed, in m / s; — MMG simulation predicts the speed, in m / s; — Roll angle prediction error, in rad or degrees; — Actual observed roll angle, in rad or degrees; — MMG simulation predicts the roll angle, in rad or degrees; — Wind tilt moment prediction error, in N·m; — Estimated wind tilt moment, in N·m; — MMG simulation predicts wind tilt moment, in N·m.

[0189] The recursive least squares (RLS) algorithm is used to update the key parameters in the MMG model, namely the wind pressure coefficient, in real time. And roll damping ratio The RLS recursive formula is: ; ; .

[0190] in: — The parameter estimation vector at time t, specifically including the wind pressure coefficient. And roll damping ratio Both are dimensionless parameters; — Kalman gain vector, dimensionless; — Actual observed output at time t (e.g., roll angle) The dimensions depend on the observed variables; — The regression vector at time t contains current meteorological and hydrological conditions (such as wind speed, ship speed, etc.), with different dimensions for each dimension; — Transpose of the regression vector; — Covariance matrix, whose dimensions are the combination of variances of the elements of the parameter vector; — Forgetting factor, dimensionless We set it to 0.95 to control the decay rate of historical data; — Identity matrix, dimensionless.

[0191] After each voyage, all recorded correction parameters are stored in the ship type parameter knowledge base for reference by subsequent voyages of similar ships.

[0192] The purpose of step S62 is to address the issue that while the MMG model introduced in step S5 has a solid physical foundation, its empirical coefficients (such as wind pressure coefficient) are somewhat flawed. Roll damping ratio Because the values ​​may vary depending on the ship and the loading conditions, laboratory calibration values ​​may deviate from actual navigation results. This step uses the RLS algorithm to correct these two key parameters in the MMG model online, enabling the calibration model to continuously learn the ship's true dynamic characteristics during actual navigation, achieving personalized accuracy improvement with one model per ship and one update per voyage. The recursive nature of the RLS algorithm results in extremely low computational cost, allowing it to run in real time on the ship's existing computing platform.

[0193] S63 archives complete data from each voyage to a central database, triggers an incremental learning process according to a preset cycle, retrains the conditional generative adversarial network scenario generation model and the enhanced random forest short-term forecast model, updates the neural network weights of the conditional generative adversarial network and the ensemble learning parameters of the enhanced random forest, refits the local calibration coefficients in the precipitation-visibility empirical relationship model, optimizes the hyperparameters of the evolutionary algorithm, and adopts an elastic weight consolidation strategy to maintain the memory of old knowledge while learning new data.

[0194] Establish a closed-loop data mechanism covering the entire lifecycle, archiving all data from each voyage (meteorological observations, AIS trajectory, ship response, final route plan, actual emergencies encountered, and pilot evaluations) to a central database. Incremental learning is triggered quarterly. (1) Merge the new data with the historical data, retrain the cGAN scenario generation model in step S1 and the enhanced random forest short-term forecast model input in step S2, update the neural network weights of the generator and discriminator in the conditional generative adversarial network, and the integrated learning parameters such as decision tree structure, split threshold and feature importance weight in the enhanced random forest, so that the model can adapt to the latest changes in meteorological patterns (e.g., the increase in the frequency of extreme events caused by climate change).

[0195] (2) Perform Bayesian hyperparameter optimization on the NSGA-III optimization parameters (crossover probability, mutation probability, population size) in step three. Based on the optimization convergence speed and solution quality index of the past quarter, automatically adjust the algorithm parameters.

[0196] (3) Refit the visibility model in step S2 The local calibration coefficients a and b in the data were refitted using the weighted least squares method based on the latest accumulated synchronous precipitation-visibility observation pairs, with the new data having a higher weight than the old data (reflecting the long-term trend of climate change).

[0197] Incremental learning employs an Elastic Weight Consolidation (EWC) strategy to prevent catastrophic forgetting of old knowledge while learning new data. EWC adds a regularization term to the loss function: .

[0198] in: — Total loss function after EWC regularization, dimensions and Consistent; — The current parameter vector of the model (for cGAN, it is the neural network weights of the generator and discriminator; for enhanced random forest, it is the decision tree structure parameters and feature weights; for visibility model, it is the coefficients a and b). The physical dimensions of each parameter dimension are different. The regularization term only constrains the parameter offset and does not change the balance of dimensions. — The loss function on new data (such as mean squared error) is measured in units of the square of the target variable; — Parameter index, dimensionless integer; — EWC regularization strength, dimensionless, ranging from 10 to 100; — Diagonal elements of the Fisher information matrix, dimensionless, measurement parameters The importance of old tasks; — The current model's first One parameter; — The first of the old model Each parameter (the baseline value saved after training the old task).

[0199] This strategy ensures that the model does not forget historically validated forecasting models when learning new meteorological patterns.

[0200] Step S63 serves the purpose of addressing the fact that meteorological models and navigational environments are not static—climate change may lead to an increase in the frequency of strong winds, river dredging may alter local flow characteristics, and the introduction of new ship types may change the distribution of the ship type parameter library. This step updates the core parameters of the three types of models (the neural network weights of cGAN, the ensemble learning parameters of the enhanced random forest, and the local calibration coefficients of the precipitation-visibility model) to enable the lifecycle incremental learning mechanism to self-evolve. As runtime and data accumulation increase, the model's forecast accuracy and optimization performance continuously improve. This fundamentally differs from the traditional development-deployment-static software model—this method is a continuously growing intelligent system whose value increases over time.

[0201] Step S6 aims to establish a continuous improvement mechanism for route optimization. Weather forecasting is inherently probabilistic, with shorter lead times generally leading to greater accuracy and longer lead times to greater uncertainty. This step, through a three-step design, enables the optimization method to continuously update the route as new observation data becomes available during the voyage, while simultaneously feeding back actual data and experience from each voyage to the model, thus achieving continuous evolution.

[0202] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0203] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0204] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps. Example

[0205] Further reference Figure 2As a response to the above Figure 1 The present invention provides an embodiment of a multi-objective optimization device for ship routes under complex sea state scenarios, which is similar to the method shown. Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0206] like Figure 2 As shown, the multi-objective optimization device 70 for ship routes under complex sea conditions described in this embodiment includes: an identification module 71, a construction module 72, an adjustment module 73, a generation module 74, a testing module 75, and an optimization module 76. Wherein: The identification module 71 is used to identify and classify complex sea state event types based on multi-source meteorological observation data and generate high-resolution meteorological scene fields; Module 72 is used to establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and to construct differentiated safety constraints under complex sea states; The adjustment module 73 is used to adaptively adjust the weights of risk objectives and economic objectives based on the real-time identified scenario types using an improved multi-objective evolutionary algorithm, and to search for the Pareto optimal route solution set. The generation module 74 is used to perform diversity maintenance and sparsification processing on the Pareto solution set to generate a set of optimal route options for decision-makers to choose from. Test module 75 is used to perform independent physical simulation verification, historical flight mode verification, and emergency safety margin testing for candidate route schemes, and eliminate unsafe schemes. Optimization module 76 is used to continuously update the route plan based on the latest observation data, and continuously optimize the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model through online learning and full life cycle incremental learning mechanisms.

[0207] Example 3 To address the aforementioned technical problems, embodiments of the present invention also provide an electronic device. Please refer to [link / reference needed]. Figure 3 , Figure 3 This is a basic structural block diagram of the electronic device in this embodiment.

[0208] The aforementioned electronic device 8 includes a memory 81, a processor 82, and a network interface 83 that are interconnected via a system bus. It should be noted that only the electronic device 8 with components 81, 82, and 83 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the electronic device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0209] The aforementioned electronic devices can be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These electronic devices can interact with users via keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0210] The aforementioned memory 81 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the aforementioned memory 81 may be an internal storage unit of the aforementioned electronic device 8, such as the hard disk or memory of the electronic device 8. In other embodiments, the aforementioned memory 81 may also be an external storage device of the aforementioned electronic device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 8. Of course, the aforementioned memory 81 may also include both internal storage units and external storage devices of the aforementioned electronic device 8. In this embodiment, the aforementioned memory 81 is typically used to store the operating system and various application software installed on the aforementioned electronic device 8, such as computer-readable instructions for a multi-objective optimization method for ship routes under complex sea conditions. In addition, the aforementioned memory 81 can also be used to temporarily store various types of data that have been output or will be output.

[0211] In some embodiments, the processor 82 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 82 is typically used to control the overall operation of the electronic device 8. In this embodiment, the processor 82 is used to execute computer-readable instructions stored in the memory 81 or to process data, such as executing computer-readable instructions for the multi-objective optimization method for ship routes under complex sea state scenarios.

[0212] The aforementioned network interface 83 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the aforementioned electronic device 8 and other electronic devices. Example

[0213] The present invention also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the above-described method for multi-objective optimization of ship routes for complex sea state scenarios.

[0214] The beneficial effects of implementing the above embodiments are as follows: (1) It has the function of high-fidelity wind and rain complex scenario characterization: through multi-source observation fusion and conditional generation adversarial network, it realizes the type identification of complex sea state events and high-resolution scenario field reconstruction, which significantly improves the spatial details of environmental input and the accuracy of disaster coupling characterization.

[0215] (2) The adaptive dynamic weights have been optimized: the improved multi-objective evolutionary algorithm can dynamically adjust the risk and economic weights based on the disaster type identified in real time, so that the route solution set can achieve a scenario-adaptive balance between safety redundancy and economy, avoiding one-size-fits-all decision-making.

[0216] (3) It has multi-level safety assurance: It introduces a triple screening mechanism of independent physical simulation verification, historical navigation mode comparison and emergency margin test in extreme scenarios, and eliminates unsafe schemes from three dimensions of dynamics, statistics and extreme working conditions, which greatly improves the engineering credibility of the solution set.

[0217] (4) It has the ability to continuously evolve throughout the entire life cycle: It integrates online learning and incremental learning mechanisms to drive the ship motion model parameters, generator network weights, random forest forecast parameters and empirical relationship local coefficients to be updated with new observation data, so that the system has the adaptive ability to optimize as it is used, and its long-term timeliness is significantly better than that of static models.

[0218] (5) The solution set is highly practical: Through diversity maintenance and sparsification, it outputs Pareto front schemes with moderate size and obvious differences, which makes it easy for decision-makers to quickly select the best according to the cruise preference, taking into account both theoretical optimality and practical ease of operation.

[0219] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0220] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A multi-objective optimization method for ship routes under complex sea state scenarios, characterized in that, Includes the following steps: Based on multi-source meteorological observation data, we identify and classify complex sea state event types and generate high-resolution meteorological scene fields. Establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and construct differentiated safety constraints under complex sea states; An improved multi-objective evolutionary algorithm is used to adaptively adjust the weights of risk and economic objectives based on the real-time identified scenario types, and to search for the Pareto optimal route solution set. The Pareto solution set is subjected to diversity preservation and sparsification processing to generate a set of optimal route options for decision-makers to choose from. Independent physical simulation verification, historical flight pattern verification, and emergency safety margin testing for extreme scenarios are conducted on candidate route schemes to eliminate unsafe schemes. The route plan is updated on a rolling basis based on the latest observation data, and the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model are continuously optimized through online learning and full life cycle incremental learning mechanisms.

2. The multi-objective optimization method for ship routes under complex sea state scenarios according to claim 1, characterized in that, The steps for identifying and classifying complex sea state event types based on multi-source meteorological observation data and generating high-resolution meteorological scene fields specifically include: Spatiotemporal matching, outlier detection and removal, missing value imputation and adaptive filtering smoothing are performed on data from ground automatic weather stations, weather radars and GNSS water vapor inversion or reanalysis to obtain a standardized meteorological dataset. Based on wind speed and precipitation intensity threshold-triggered event detection, a multi-dimensional feature vector including gust wind speed, average wind speed, rainfall intensity, wind direction standard deviation, and wind speed rise rate is extracted. An unsupervised clustering algorithm is used to classify the events into pure strong wind type, pure heavy rainfall type, and wind and rain composite type scenarios. Using the classified event type labels as conditions, the low-resolution meteorological forecast field is reconstructed in super-resolution, and the high spatiotemporal resolution wind speed, gust, wind direction and precipitation intensity forecast fields of key flight segment grid points in the study area are output.

3. The multi-objective optimization method for ship routes under complex sea state scenarios according to claim 1, characterized in that, The steps of establishing a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and constructing differentiated safety constraints under complex sea states, specifically include: The route safety risk is defined as the cumulative integral of risk exposure over time periods along the entire route. The wind tilting moment ratio is calculated based on the forecast gust speed, the ship's windward area, the wind pressure coefficient, the wind pressure center height, and the wind direction angle. The visibility deviation is determined based on the ratio of the forecast precipitation intensity to the minimum safe operating visibility of the route segment. The comprehensive risk is the higher of the two. A weighted economic objective function is established, which includes the expected total flight time and the expected total fuel consumption. The expected total flight time is obtained by integrating the ratio of the range infinitesimal element to the ground speed, and the expected total fuel consumption is obtained by calculating the relationship between propulsion power and speed and wind-induced additional drag correction. Establish upper and lower limits for hard constraints on the wind tilt moment ratio and the visibility deviation, respectively, establish hard constraints on the safe distance of the waterway boundary, and establish coupled constraints for the combined wind and rain scenario, requiring that the joint index of the wind tilt moment ratio and the visibility deviation does not exceed a predetermined threshold.

4. The multi-objective optimization method for ship routes under complex sea state scenarios according to claim 1, characterized in that, The steps of using an improved multi-objective evolutionary algorithm to adaptively adjust the weights of risk and economic objectives based on the real-time identified scenario types, and searching for the Pareto optimal route solution set, specifically include: The third generation of non-dominated sorting genetic algorithm is used as the optimization engine. Through adaptive reference point generation, constraint dominance priority strategy and population initialization strategy based on scenario switching, Pareto optimal route solution set that meets the constraints is searched. Based on the real-time identified scenario types, the priority weights of risk objectives and economic objectives in the optimization process are dynamically adjusted, and a smooth transition function is used to avoid abrupt weight changes at scenario boundaries. The similarity between the current scenario and historical scenarios is calculated. Elite individuals are extracted from the Pareto front solution set of similar historical scenarios as prior seeds and mixed with randomly generated individuals to form the initial population, thereby realizing cross-scenario knowledge transfer and accelerating optimization convergence.

5. The multi-objective optimization method for ship routes under complex sea state scenarios according to claim 1, characterized in that, The steps of performing diversity preservation and sparsification processing on the Pareto solution set to generate the optimal route scheme set for decision-makers to choose from specifically include: An adaptive grid clustering method is used to sparsify the final non-dominated solution set, retaining representative solutions in each non-empty cell and interpolating virtual solutions to keep the size of the solution set within a preset range and maintain the continuity of the Pareto front. Calculate the distance between each solution in the sparsified Pareto solution set and the positive and negative ideal solutions, determine the relative proximity and sort them in descending order, and output the recommended solution ranking based on user preferences. The selected route plan is visualized by overlaying a meteorological risk layer on a 3D geospatial scene, and the recommended course, recommended speed, estimated arrival time and maximum heel angle parameters for each segment are automatically extracted from the optimal route plan for direct use by the ship's navigator.

6. The multi-objective optimization method for ship routes under complex sea state scenarios according to claim 1, characterized in that, The steps of conducting independent physical simulation verification, historical flight pattern verification, and emergency safety margin testing for extreme scenarios on candidate route schemes, and eliminating unsafe schemes, specifically include: The hull-propeller-rudder separation modeling method is used to perform time-domain simulation of the six-degree-of-freedom motion of the ship on the candidate route. If the roll angle exceeds the limit, the rudder effect is lost, or the heading deviation exceeds the correction capability during the simulation, the route plan is determined to be unsafe. Retrieve historical tracks with similar shapes to the candidate route from the historical AIS database, extract the actual navigation status of similar tracks under corresponding weather conditions, and issue a navigability warning if most historical tracks choose to decelerate or drift, or have records of running aground, or have no records of passing through. Assuming that weather conditions suddenly deteriorate beyond the forecast range during navigation, calculate the emergency safety margin at each location on the candidate route, and test whether the emergency reachability time from any point on the route to the nearest emergency shelter anchorage meets the preset response time limit requirements.

7. The multi-objective optimization method for ship routes under complex sea state scenarios according to any one of claims 1 to 6, characterized in that, The steps of continuously updating the route plan based on the latest observation data and continuously optimizing the wind pressure coefficient and roll damping ratio of the MMG-separated ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model through online learning and full life-cycle incremental learning mechanisms specifically include: A model predictive control framework is adopted, taking the ship's current position as a new starting point, and combining the latest short-term meteorological forecasts to perform replanning within the rolling optimization time window. An event triggering mechanism is set based on the real-time changes in wind speed or precipitation intensity, and replanning is triggered immediately when the changes exceed the threshold. The ship response data observed during actual navigation were compared with the simulation prediction values ​​of the MMG split ship motion model. The wind pressure coefficient and roll damping ratio in the model were updated online using the recursive least squares algorithm, and the corrected parameters were stored in the ship type parameter knowledge base. Complete data from each voyage is archived to a central database. An incremental learning process is triggered at preset intervals to retrain the Conditional Generative Adversarial Network (CGAN) scenario generation model and the Enhanced Random Forest (ARFR) short-term forecast model. The neural network weights of the CGAN and the ensemble learning parameters of the ARFR are updated. The local calibration coefficients in the precipitation-visibility empirical relationship model are refitted. The hyperparameters of the evolutionary algorithm are optimized. An elastic weight consolidation strategy is adopted to retain the memory of old knowledge while learning new data.

8. A multi-objective optimization device for ship routes under complex sea state scenarios, characterized in that, include: The identification module is used to identify and classify complex sea state event types based on multi-source meteorological observation data, and generate high-resolution meteorological scene fields; The module is used to establish a multi-objective optimization function that includes navigation safety risk objectives and economic objectives, and to construct differentiated safety constraints under complex sea states; The adjustment module is used to adaptively adjust the weights of risk objectives and economic objectives based on the real-time identified scenario types using an improved multi-objective evolutionary algorithm, and to search for the Pareto optimal route solution set. The generation module is used to maintain diversity and sparsify the Pareto solution set to generate a set of optimal route options for decision-makers to choose from. The testing module is used to perform independent physical simulation verification, historical flight pattern verification, and emergency safety margin testing for candidate route schemes, and to eliminate unsafe schemes. The optimization module is used to continuously update the route plan based on the latest observation data, and continuously optimize the wind pressure coefficient and roll damping ratio of the MMG split ship motion model, the neural network weights in the conditional generative adversarial network scenario generation model, the ensemble learning parameters in the enhanced random forest short-term forecast model, and the local calibration coefficients in the precipitation-visibility empirical relationship model through online learning and full life cycle incremental learning mechanisms.

9. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the multi-objective optimization method for ship routes in complex sea state scenarios as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the multi-objective optimization method for ship routes in complex sea state scenarios as described in any one of claims 1 to 7.

Images