A hoisting operation path planning method and system for multi-target collaborative optimization

Abstract

The application discloses a kind of multi-objective cooperative optimization hoisting operation path planning method and system, belong to hoisting operation path planning field, comprising: obtaining real-time environmental information, based on wind speed and obstacle height dynamic calculation safety distance and construct dynamic prohibited area;Path parameterization is key control point, with path length, wind-induced risk, hoisting swing and lifting energy consumption as optimization target, introduce the penalty term based on penetration depth to construct multi-objective optimization problem;Global search is carried out using NSGA-II, then elite solution is selected from Pareto frontier, L-BFGS-B local refinement is used and vertical lifting repair strategy is applied, finally, the solution set is combined and reordered to output non-dominated solution set.The application reduces wind-induced risk by 34.2% through dynamic safety distance model, generates diversified solution set using NSGA-II, combines "soft penalty + hard repair" to make feasible solution rate reach 98.0%, mixed optimization framework convergence speed is improved by 2 times and solution quality is improved by 15% on average, enhances hoisting operation safety and robustness under complex sea conditions.

Description

Technical Field

[0001] This invention relates to the field of offshore lifting path planning technology, and in particular to a multi-objective collaborative optimization method and system for lifting operation path planning. Background Technology

[0002] With the global energy structure shifting towards cleaner energy, the construction of large-scale marine infrastructure such as offshore wind power and deep-water oil and gas platforms is developing rapidly. These projects generally rely on floating crane vessels, semi-submersible installation platforms, or jack-up wind turbine installation vessels to perform high-altitude heavy lifting operations. However, the offshore operating environment is highly time-varying: the coupling effect of wind, waves, and currents causes the hull to move in six degrees of freedom, which in turn causes the suspended load to swing significantly; at the same time, the operating area often contains tall obstacles such as towers, jackets, and adjacent wind turbines, which place extremely high demands on path safety. Traditional path planning methods are mostly based on static obstacle avoidance models and a single optimization objective (such as the shortest path), which are difficult to cope with challenges such as dynamic safety boundaries, multi-physics coupling velocity reduction, and conflicts between multiple performance indicators.

[0003] The existing technology has the following main drawbacks:

[0004] (1) The static safety distance model fails to take into account the strong dynamics of the marine environment.

[0005] For example, Chinese invention patent CN114911240A discloses a "dynamic obstacle avoidance path planning method for unmanned surface vessels assisted by unmanned aerial vehicles." Although it involves a maritime operation scenario, its obstacle avoidance strategy still mainly relies on preset geometric safety boundaries or relatively fixed dynamic windows, failing to fully couple the load swing caused by instantaneous gusts and surges at sea, as well as the coupled dynamic characteristics of the unmanned aerial vehicle-lifting system. In actual maritime lifting operations, if such fixed or semi-fixed safety margin settings are used: in strong crosswinds... In certain environments, the model cannot compensate for trajectory deviations caused by wind loads in real time, which can easily lead to collision risks. For tall obstacles (such as wind turbine towers and transmission towers), the pendulum effect of the hoisted cargo in the wind field is not considered, and the fixed safety distance is insufficient to cover the maximum swing envelope of the cargo. In open and calm sea areas, the model lacks an adaptive contraction mechanism, resulting in an overly conservative detour path, which significantly increases the energy consumption and time cost of operations.

[0006] (2) The multi-objective optimization method is rigid and difficult to adapt to the nonlinear conflicts of the hoisting system.

[0007] For example, Chinese invention patent CN112781592A discloses "A Path Planning Method and System for Low-Altitude Logistics UAVs," which, in order to solve the multi-objective problem of path length, energy consumption, and safety, adopts a linear weighting method to assign weights to the multi-dimensional objectives by manually setting weights (e.g., ...). This method transforms the problem into a single objective function for solution. However, offshore lifting scenarios are highly non-convex and strongly coupled, leading to significant drawbacks in this approach.

[0008] ① Omission of non-convex Pareto front: The linear weighted method can only search for convex Pareto fronts and cannot capture high-quality solutions in non-convex regions. In nonlinear regions where there is a sharp conflict between "shortest path" and "lowest energy consumption / minimum sway" (such as when traversing narrow tower gaps), this method will miss a large number of potential non-dominated solutions, making it impossible to find the optimal balance path that combines high safety and low energy consumption under strong wind disturbances.

[0009] ② Weight Sensitivity and Subjectivity: This method is highly sensitive to weight settings; even small changes in weights can lead to huge jumps in the solution. Under complex sea conditions, the optimal weights change in real time with wind speed and load mass. Fixed weights cannot adapt to dynamic environments and it is difficult to quantify the dynamic trade-offs between different objectives.

[0010] (3) Local search fails under non-smooth approximation conditions.

[0011] Traditional gradient descent or sequential quadratic programming (SQP) is non-differentiable at obstacle constraint boundaries, causing local optimizers to get trapped in invalid regions or diverge (Liu, Y., et al. “Safe Trajectory Planning for UAVs in Cluttered Environments: A Non-Smooth Optimization Approach.” IEEE Trans. Ind. Inform., 2022). This paper points out that traditional smooth optimization methods suffer from gradient discontinuities at obstacle boundaries, leading to convergence failures or the generation of suboptimal solutions. In dense obstacle environments, the failure rate of traditional methods can reach as high as 37%.

[0012] In summary, existing path planning methods have significant shortcomings in terms of dynamic environment adaptability, multi-objective collaborative optimization, and handling of non-smooth constraints, making it difficult to meet the high safety and efficiency requirements of offshore lifting operations under complex sea conditions. Therefore, there is an urgent need for a dynamic multi-objective path planning method that integrates environmental perception modeling, nonlinear dynamic constraint handling, and hybrid intelligent optimization strategies to achieve dynamic adjustment of safety distance, Pareto collaborative optimization of multiple performance indicators, and efficient local refinement under non-smooth constraints. Summary of the Invention

[0013] The technical problem to be solved by this invention is to provide a multi-objective collaborative optimization method and system for hoisting operation path planning. By deeply integrating environmental perception modeling, nonlinear dynamic constraint processing and hybrid intelligent optimization strategies, a verifiable and deployable path planning system is constructed to solve the technical challenges faced by offshore hoisting operations under complex sea conditions.

[0014] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0015] A multi-objective collaborative optimization method for hoisting operation path planning includes the following steps: Step 1: Obtain real-time environmental information, including at least wind speed, spatial location and height of obstacles; Step 2: Based on the wind speed and obstacle height, dynamically calculate the safety distance, and construct a dynamic prohibited zone for each obstacle based on the safety distance;

[0016] Step 3: Parameterize the hoisting path into a decision vector composed of the coordinates of key control points. Use path length, wind-induced risk, load swing amplitude, and lifting energy consumption as optimization objectives. Introduce a penalty term based on the penetration depth of path points and dynamic prohibited areas to construct a multi-objective optimization problem with a penalty term. Step 4: Solve the multi-objective optimization problem with a penalty term using a hybrid optimization strategy.

[0017] First pass The algorithm performs a global search to generate an initial Pareto front; then it selects elite solutions from the initial Pareto front and utilizes... The algorithm performs local refinement and applies a vertical lift repair strategy in real time during the local optimization process to forcibly lift path points that fall into the dynamic prohibited area to a safe height.

[0018] Step 5: Merge the original solution obtained from the global search with the optimized solution obtained from the local optimization, re-sort the non-dominated solutions, and select the final set of non-dominated solutions as the output of the alternative path schemes.

[0019] A further improvement to the technical solution of the present invention is that, in step 2, the dynamic calculation of the safety distance is specifically calculated according to the following formula:

[0020]

[0021] in, The height of the obstacle; Wind speed; This is the wind-induced offset coefficient; ; Based on the basic safety margin, .

[0022] A further improvement to the technical solution of this invention is that, in step 3, the formula for calculating the path length is:

[0023]

[0024] in, This represents the total number of critical control points. For the first Spatial coordinates of the control points; , for the first Spatial coordinates of the control points.

[0025] A further improvement to the technical solution of this invention is that, in step 3, the formula for calculating the wind-induced risk is:

[0026]

[0027] in, Wind speed; This represents the total number of critical control points. For the first Lateral obstacle avoidance distance for each control point; For the first Lateral obstacle avoidance distance for each control point; For the first The height coordinates of each control point; For the first The height coordinates of each control point.

[0028] A further improvement to the technical solution of the present invention is that, in step 3, the calculation formula for the swing amplitude of the suspended object is:

[0029]

[0030] in, This is the time step under the assumption of uniform motion; For the first Spatial coordinates of the control points; For the first Spatial coordinates of the control points; For the first The space of each control point is marked; This represents the total number of critical control points. Let i be the height coordinates of the i-th control point; This refers to the deck height.

[0031] A further improvement to the technical solution of the present invention is that, in step 3, the calculation formula for increasing energy consumption is:

[0032]

[0033] in, For the weight of the suspended object, It is the acceleration due to gravity; For the first The height coordinates of each control point; For the first The height coordinates of each control point; This represents the total number of critical control points.

[0034] A further improvement to the technical solution of the present invention is that, in step 3, the penalty term is specifically:

[0035]

[0036] in, This is the penalty coefficient; The average penetration depth of all path points relative to each obstacle;

[0037] Penetration depth is defined as:

[0038]

[0039] in, For the first The height of the obstacle; For the first The elevation coordinates of each path point.

[0040] A further improvement to the technical solution of this invention is that, in step 4, the local search optimization objective function is:

[0041]

[0042] in, For path length, For wind-induced risk points, For the swing amplitude of the suspended object, To improve energy efficiency, This is a penalty item.

[0043] A further improvement to the technical solution of the present invention is that, in step 4, the vertical lifting repair strategy is as follows:

[0044] When path point Satisfy horizontal distance constraints And height constraints Immediately correct the height of that point to:

[0045]

[0046] in, The coordinates of the center of the bottom surface of the obstacle are: This refers to the height of the obstacle.

[0047] A multi-objective collaborative optimization hoisting operation path planning system includes: An environmental perception module is used to collect operational environment information in real time, including at least wind speed, spatial location and height of obstacles; The dynamic modeling module, connected to the environmental perception module, is used to calculate the dynamic safety distance based on wind speed and obstacle height, and to construct the dynamic prohibited area for each obstacle based on the safety distance.

[0048] Multi-objective optimization engine, built-in Global search algorithm and Local optimization algorithms are used to solve multi-objective optimization problems that introduce penalty terms;

[0049] The dynamic repair unit, coupled with the multi-objective optimization engine, is used to execute the vertical lift repair strategy during the local optimization process, forcibly lifting path points that fall into the dynamic prohibited area to a safe height; The path output module is used to merge the solution sets obtained from global and local optimizations, and output the final optional path scheme after non-dominated sorting.

[0050] The model export interface supports encapsulating planning logic into ONNX models for deployment and inference on embedded platforms.

[0051] The technological advancements achieved by this invention due to the adoption of the above technical solutions are as follows:

[0052] 1. This invention overcomes the shortcomings of traditional fixed safety distances, such as increased risk in strong winds and insufficient forward margin for tall obstacles, by establishing a dynamic safety distance model based on wind speed and obstacle height. Experiments show that compared to traditional fixed obstacle avoidance methods, this invention reduces wind-induced risk by an average of 34.2%, and the minimum obstacle avoidance distance for all planned paths strictly meets the dynamic safety threshold (under typical conditions). This significantly enhances operational safety in complex sea conditions.

[0053] 2. This invention adopts The algorithm replaces the linear weighted method, solving the problems of traditional methods failing to capture non-convex Pareto fronts and being sensitive to weights, leading to solution set distortion. This invention can generate a diverse set of equilibrium solutions (such as shortest path, most stable path, and equilibrium path), allowing operators to flexibly choose according to actual working conditions (urgency, sea state level), avoiding the one-sidedness of a single objective and truly achieving synergistic optimization of multiple performance indicators.

[0054] 3. This invention addresses the problem of traditional gradient methods failing at obstacle boundaries due to non-differentiability. This invention pioneers a "soft penalty + hard repair" strategy. By introducing... The high penalty term incorporates collision depth into the objective function and performs vertical lift repair in real time during local search. ), ensuring The algorithm achieves stable convergence under non-smooth constraints, increasing the feasible solution rate to 98.0% and significantly enhancing its robustness.

[0055] 4. This invention constructs " Global Search The hybrid optimization framework of "local refinement" shows that, compared with the pure evolutionary algorithm, the convergence speed is improved by more than 2 times and the quality of the solution (the objective function values ​​of each item) is improved by an average of 15%. It effectively balances computational efficiency and optimal solution accuracy and is suitable for complex path planning problems with high dimensions, non-convexity and multiple constraints.

[0056] 5. By encapsulating the planning logic into a standard ONNX model, this invention enables the system to achieve fast inference of less than 50 milliseconds (taking 15 key point outputs as an example) on embedded platforms such as NVIDIA Jetson. The average time for a single complete planning is 8.3 seconds, which meets the engineering requirements of real-time closed-loop in shipboard control systems. It truly realizes the technical implementation from theoretical algorithms to actual equipment and has significant engineering application value. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of a multi-objective collaborative optimization hoisting operation path planning method provided in an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of the dynamic safety distance model in an embodiment of the present invention;

[0060] Figure 3 This is a schematic diagram illustrating the optimal result of the objective function in an embodiment of the present invention;

[0061] Figure 4 This is an embodiment of the present invention. Graph showing the convergence iteration count;

[0062] Figure 5 These are box plots comparing the performance of different algorithms in embodiments of the present invention;

[0063] Figure 6 This is a comparison chart of the success rates of different algorithms in this invention embodiment;

[0064] Figure 7 This is a schematic diagram of the algorithm evolution process in an embodiment of the present invention. Detailed Implementation

[0065] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0066] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments:

[0067] like Figure 1 As shown, a multi-objective collaborative optimization method for hoisting operation path planning is proposed, which considers environmental input, dynamic safety modeling, and multi-objective optimization. Global Search The complete technology chain from local refinement to the final Pareto solution output includes real-time collision detection, penalty mechanisms, repair strategies, and the ONNX deployment interface; specifically, it includes the following:

[0068] (I) Mathematical Modeling and Derivation

[0069] To ensure the theoretical rigor and engineering feasibility of the path planning method, this invention establishes a complete mathematical model. The following sections elaborate on this from coordinate system setting, environment modeling, objective function construction to constraint handling.

[0070] 1.1 Coordinate System and Path Parameterization

[0071] Set up a world coordinate system Fixed to the center of the installation vessel deck, Pointing towards the zenith. The hoisting path is... The key control points consist of:

[0072]

[0073] in, For the lifting point, The target installation point is defined. The actual motion trajectory is obtained as a smooth curve through cubic B-spline interpolation. → This is to ensure continuous velocity and acceleration and avoid vibration.

[0074] The optimization variable is the coordinates of all intermediate key points (with the first and last points fixed), i.e., the decision vector:

[0075]

[0076] 1.2 Obstacle Modeling and Dynamic Safety Domain

[0077] Offshore obstacles (such as wind turbine towers and jackets) are approximately vertical cylinders. The j-th obstacle... By triplet Full description, in which h is the center of its base. j Its top height.

[0078] Under wind load, the center of gravity of the suspended mass shifts in the horizontal plane. Based on potential flow theory and field measurement data, the maximum lateral shift... With height and wind speed Positive correlation. To balance computational efficiency and security, a linear empirical model is used to define the "dynamic safety radius":

[0079]

[0080] in, This is the wind-induced offset coefficient. This is the basic safety margin (covering measurement errors and model uncertainties).

[0081] Therefore, obstacles The corresponding "prohibited area" is a dynamic cylinder:

[0082]

[0083] 1.3 Collision Penetration Depth and Violation Measurement

[0084] For any path point, If it falls into any If the collision is not complete, it is considered a potential collision. The "penetration depth" is defined as:

[0085]

[0086] Notice: This indicates that the point is "not high enough" and needs to be raised by at least [a certain amount]. Only rice can save them from danger.

[0087] The overall violation severity is the average of all non-zero penetration depths:

[0088]

[0089] in, This is the collision index set.

[0090] 1.4 Formal Definition of Multi-Objective Functions

[0091] This invention simultaneously optimizes the following four objectives, all of which are functions of x:

[0092] (1) Path length (minimize):

[0093]

[0094] (2) Wind-induced risk (minimization): defined as the weighted sum of the projected lengths of the path onto the plane perpendicular to the wind direction. Let the wind direction be along... The risk is proportional to In-plane movement:

[0095]

[0096] (3) Integral of the suspended object's swing amplitude (minimization): Based on the simple pendulum approximation, the local swing angle ,in For tangential acceleration, For the suspended height. After discretization:

[0097]

[0098] in, This is the time step (assuming uniform motion).

[0099] (4) Lifting energy consumption (minimize): Only the work done during the lifting process is considered (the descent is completed by gravity):

[0100]

[0101] in, For the weight of the suspended object, This is the acceleration due to gravity. Because... Since it is a constant, it can be simplified to the following during optimization:

[0102] .

[0103] After computer calculations, the optimized result is as follows: Figure 3 As shown.

[0104] 1.5 Penalty Mechanism and Optimization Compatibility Analysis

[0105] The original multi-objective problem is a constrained optimization problem:

[0106]

[0107] To facilitate the solution, we transform it into an unconstrained problem and introduce a large number penalty term:

[0108]

[0109] in, Ensure that any infeasible solution is The value is much larger than the feasible solution.

[0110] although exist It is not differentiable at a given point (due to the max function), but it has the following properties:

[0111] (1) Within the feasible region ;

[0112] (2) Within the infeasible region, It is a piecewise linear function, and its subgradient exists;

[0113] (3) In local search, Although the objective function is required to be continuous and differentiable, the algorithm will quickly achieve this as long as the initial point is close to the feasible boundary. Push to ,make This allows it to enter a smooth region.

[0114] In addition, the "vertical lifting repair strategy" can be used to force the strong feasibility to be met after each iteration, effectively avoiding non-differentiable points and ensuring the stability of local optimization.

[0115] 1.6 Mathematical Framework for Hybrid Optimization

[0116] Global phase ( Searching for Pareto approximations in the discrete population space Local phases Middle Elite Subset implement:

[0117]

[0118] in, For scalarized aggregation objectives (used for single-objective local search), the optimization variable x is constrained by the physical boundary:

[0119]

[0120] For example rice, Within the operating radius.

[0121] The final solution set is ,in This indicates a non-dominated sort.

[0122] This framework combines "global exploration capability" with "local convergence accuracy," making it an effective paradigm for handling high-dimensional, non-convex, and multi-objective problems with complex constraints. The comparison chart of convergence iterations is shown below. Figure 4 As shown.

[0123] (II) Implementation Examples

[0124] 2.1 Setting up the work scenario

[0125] Imagine a large wind turbine installation vessel anchored 50 kilometers offshore, needing to lift an 80-meter-high tower from the transport barge and precisely install it onto the pre-drilled pile foundation. At this time, the wind speed at sea is 10 meters per second (about force 5), and there are already other wind turbines or jacket structures around, forming several towering obstacles.

[0126] In this environment, the suspended load (i.e., the tower) does not move in a straight line—it swings back and forth like a pendulum due to the wind; at the same time, the ship itself also rises and falls with the waves. If the path planning only considers the "shortest distance," the suspended load may collide with a nearby wind turbine during the swing, causing a serious accident.

[0127] Therefore, the objective of this invention is to automatically plan an optimal trajectory that balances path length, wind risk, load stability, and energy consumption, while ensuring absolute safety. This invention provides a complete technical solution to this complex multi-objective problem.

[0128] 2.2 Dynamic safety distance

[0129] Traditional methods typically set a fixed obstacle avoidance distance, such as "at least 20 meters from the obstacle." However, at sea, this approach is either too conservative or dangerous: when the wind is light, 20 meters may be far beyond what is actually needed, resulting in a longer, more time-consuming, and energy-intensive route; when the wind is strong, 20 meters may not be enough—because strong winds can cause the suspended load to deviate significantly, and even if the waypoint itself is not in the obstacle zone, the end of the suspended load may still collide with it.

[0130] Therefore, such as Figure 2 As shown, obstacle height The higher the wind speed The larger the value, the greater the required lateral avoidance distance; this invention "proposes a dynamic safety distance model": The specific formula for the margin is:

[0131]

[0132] Where h is the height of the obstacle (in meters). This refers to wind speed (unit: m / s). For example, when... , When the speed is 10 m / s, the calculation yields... = 41.5m. This means that any point on the path, if its horizontal position is less than 41.5 meters from the center of the obstacle, must ensure that its height is greater than 80 meters, otherwise it is considered a potential collision.

[0133] This design allows the safety boundary to "change with the wind," avoiding excessive conservatism while ensuring safety under extreme working conditions.

[0134] Listing 1: Wind speed and obstacle parameter settings (actual engineering configuration)

[0135]

[0136] 2.3 Four Optimization Objectives

[0137] This invention considers four key performance indicators simultaneously, which often conflict with each other and require trade-offs through multi-objective optimization:

[0138] (1) Path length (the shorter the better): directly affects the operation time. However, blindly pursuing a short path may lead to obstacles and increase the risk.

[0139] (2) Wind-induced risk (lower is better): measures the "exposure" of the path to the wind direction. For example, making large lateral movements in the direction of strong winds will significantly increase the risk of swaying.

[0140] (3) Sway amplitude of the suspended object (the smaller the better): reflects the stability of the hoisting. Severe swaying is not only dangerous, but may also lead to docking failure. The sway amplitude is closely related to the curvature of the path and the acceleration.

[0141] (4) Lifting energy consumption (the lower the better): This mainly refers to the work done in lifting the suspended object. The descent process is dominated by gravity and is not included in the energy consumption.

[0142] These four objectives cannot be simultaneously optimized, so instead of seeking a "unique optimal solution", we generate a set of Pareto optimal solutions (i.e. "equilibrium solutions") for operators to choose from based on the current task priority - for example, choosing the shortest path in an emergency and the most stable path in high wind speeds.

[0143] 2.4 Collision Penalty and Repair Mechanism

[0144] During the optimization process, many candidate paths inadvertently pass through obstacle areas. A dual strategy of "soft penalty + hard fix" is employed, and the performance comparison graphs and success rate comparison graphs of different algorithms are shown below. Figure 5 , Figure 6 As shown:

[0145] (1) Soft penalty: Once a path point is detected to be below the top of an obstacle and the horizontal distance is too short, the "penetration depth" (i.e. how many meters short it is to be high enough) is calculated and multiplied by a large penalty factor (e.g., 10,000) and added to all objective functions. In this way, even if the path is very short and energy-efficient, as long as it hits an obstacle, the total score will be extremely poor, and it will be eliminated in the evolution.

[0146] (2) Hard fix: During the local optimization phase (using During algorithm execution, if a collision occurs in a certain iteration, the system immediately performs a "vertical lift"—forcibly setting the height of that point to the "obstacle height". "meters". Although this temporarily disrupts the path smoothness, it can quickly pull the solution back to the feasible region, and smoothness can be restored subsequently through spline interpolation.

[0147] This combined strategy ensures both the diversity of the global search (allowing brief exploration of infeasible areas) and the 100% safety of the final output path.

[0148] Listing 2: The Core Logic of Collision Penalty and Repair

[0149]

[0150] 2.5” Hybrid optimization

[0151] It is a classic multi-objective evolutionary algorithm that excels at finding diverse Pareto solutions in complex, non-convex, and discontinuous spaces. However, it converges slowly and has limited accuracy. It is a gradient-based local optimizer that converges quickly and is highly accurate, but it can only handle single objectives and requires the objective function to be "sufficiently smooth".

[0152] This invention cleverly combines the advantages of both:

[0153] (1) First use Perform a "wide net" global search to generate an initial solution set of 120 individuals across 100 generations;

[0154] (2) Select the top 30% of the best performing paths (approximately 30–40 paths);

[0155] (3) For each high-quality path, use its key point coordinates as variables, and... Perform fine polishing;

[0156] (4) During the local optimization process, vertical lifting repair is called in real time to ensure safety at every step;

[0157] (5) Finally, the original solution and the optimized solution are merged, and the non-dominated solutions are re-selected to form the final output.

[0158] Experiments have shown that this hybrid strategy is more than twice as fast as the pure evolutionary algorithm, and the quality of the solution (objective function value) is improved by an average of 15%. The intuitive results are shown in the following graph. Figure 7 As shown.

[0159] 2.6 ONNX Model Export

[0160] To enable the algorithm to run in real time on the shipboard industrial control computer (which typically has limited resources), after planning is completed, the selected optimal path generation logic is encapsulated into a standardized computation graph and exported in ONNX format, which facilitates efficient deployment on embedded platforms that support ONNXRuntime (such as NVIDIA Jetson).

[0161] The exported model has the following advantages:

[0162] (1) Input: starting point, target point, current wind speed, list of obstacles;

[0163] (2) Output: Coordinates of 15 critical path points;

[0164] (3) Inference speed: <50 milliseconds (on Jetson AGX Orin);

[0165] (4) No Python environment required, can be used Direct call.

[0166] This means that in the future, the system can generate a new path in seconds simply by updating the input parameters, truly realizing a closed loop of "perception-planning-control".

[0167] 2.7 Experimental Results

[0168] One hundred repeated experiments were conducted in a simulated environment, and the results are as follows:

[0169]

[0170] Note: Minimum obstacle avoidance distance for all solutions. (Dynamic security threshold)

[0171] Key conclusions:

[0172] (1) The minimum obstacle avoidance distance of all schemes is greater than 41.5 meters, which meets the dynamic safety requirements;

[0173] (2) Compared with traditional fixed safety distance methods (such as 20 meters), the present invention reduces wind-induced risk by an average of 34.2%;

[0174] (3) The feasible solution rate is as high as 98%, which proves that the penalty and repair mechanism is very effective;

[0175] (4) The average time for a single planning is 8.3 seconds, which meets the real-time requirements of the project.

[0176] In summary, this invention is not only theoretically rigorous, but also possesses high reliability, high efficiency, and strong adaptability in engineering practice, and can be widely applied to various types of heavy-duty offshore lifting operations.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective collaborative optimization method for hoisting operation path planning, characterized in that, Includes the following steps: Step 1: Obtain real-time environmental information, including at least wind speed, obstacle spatial location, and height; Step 2: Based on the wind speed and obstacle height, dynamically calculate the safety distance, and construct a dynamic prohibited zone for each obstacle based on the safety distance; Step 3: Parameterize the hoisting path into a decision vector composed of the coordinates of key control points. Use path length, wind-induced risk, load swing amplitude, and lifting energy consumption as optimization objectives. Introduce a penalty term based on the penetration depth of path points and dynamic prohibited areas to construct a multi-objective optimization problem with a penalty term. Step 4: Solve the multi-objective optimization problem with a penalty term using a hybrid optimization strategy. First pass The algorithm performs a global search to generate an initial Pareto front; then it selects elite solutions from the initial Pareto front and utilizes... The algorithm performs local refinement and applies a vertical lift repair strategy in real time during the local optimization process to forcibly lift path points that fall into the dynamic prohibited area to a safe height. Step 5: Merge the original solution obtained from the global search with the optimized solution obtained from the local optimization, re-sort the non-dominated solutions, and select the final set of non-dominated solutions as the output of the alternative path schemes.

2. The method according to claim 1, characterized in that, In step 2, the dynamic calculation of the safety distance is specifically calculated according to the following formula: in, It is the height of the obstacle; It's wind speed; This is the wind-induced offset coefficient; ; Based on the basic safety margin, .

3. The method according to claim 1, characterized in that, In step 3, the formula for calculating the path length is: Where N is the total number of critical control points; For the first Spatial coordinates of the control points; , for the first Spatial coordinates of the control points.

4. The method according to claim 1, characterized in that, In step 3, the formula for calculating the wind-induced risk is: in, Wind speed; This represents the total number of critical control points. For the first The x-coordinates of each control point; For the first The x-coordinates of each control point; For the first The height coordinates of each control point; For the first The height coordinates of each control point.

5. The method according to claim 1, characterized in that, In step 3, the formula for calculating the swing amplitude of the suspended object is: in, This is the time step under the assumption of uniform motion; For the first Spatial coordinates of the control points; For the first Spatial coordinates of the control points; For the first The space of each control point is marked; This represents the total number of critical control points. For the first The height coordinates of each control point; This refers to the deck height.

6. The method according to claim 1, characterized in that, In step 3, the formula for calculating the energy consumption increase is: in, For the weight of the suspended object, It is the acceleration due to gravity; For the first The height coordinates of each control point; For the first The height coordinates of each control point; This represents the total number of critical control points.

7. The method according to claim 1, characterized in that, In step 3, the penalty item is specifically as follows: in, This is the penalty coefficient; The average penetration depth of all path points relative to each obstacle; Penetration depth is defined as: in, For the first The height of the obstacle; For the first The elevation coordinates of each path point.

8. The method according to claim 1, characterized in that, In step 4, the local search optimization objective function is: in, For path length, For wind-induced risk points, For the swing amplitude of the suspended object, To improve energy efficiency, This is a penalty item.

9. The method according to claim 1, characterized in that, In step 4, the vertical lifting repair strategy is as follows: When path point Satisfy horizontal distance constraints And height constraints Immediately correct the height of that point to: in, Let h be the coordinates of the center of the bottom surface of the obstacle, and h be the height of the obstacle.

10. A multi-objective collaborative optimization hoisting operation path planning system implementing the method of any one of claims 1-9, characterized in that, include: An environmental perception module is used to collect operational environment information in real time, including at least wind speed, spatial location and height of obstacles; The dynamic modeling module, connected to the environmental perception module, is used to calculate the dynamic safety distance based on wind speed and obstacle height, and to construct the dynamic prohibited area for each obstacle based on the safety distance. A multi-objective optimization engine, with built-in NSGA-II global search algorithm and L-BFGS-B local optimization algorithm, is used to solve multi-objective optimization problems with penalty terms. The dynamic repair unit, coupled with the multi-objective optimization engine, is used to execute the vertical lift repair strategy during the local optimization process, forcibly lifting path points that fall into the dynamic prohibited area to a safe height; The path output module is used to merge the solution sets obtained from global and local optimizations, and output the final optional path scheme after non-dominated sorting. The model export interface supports encapsulating planning logic into ONNX models for deployment and inference on embedded platforms.

Images