Unmanned surface vessel energy optimal path planning method and device in complex marine environment

Abstract

The invention provides an energy optimal path planning method and device for an unmanned surface vessel in a complex marine environment, and belongs to the technical field of path planning. The method provided by the invention comprises the following steps: constructing a comprehensive energy consumption model; constructing a hybrid enhanced particle swarm optimization algorithm; constructing a safety space set and an adjacent graph through a grid method, generating an initial optimal path by adopting an A * algorithm, and carrying out bounded random disturbance and safety correction on target particle waypoints; smoothing the path generated by iteration by adopting a B-spline technology, resampling the smoothed optimal solution, and then reinjecting the optimal solution into the population; adjusting an inertia weight and a learning factor based on the number of iterations, and introducing double guide factors to carry out secondary adjustment on a cognitive item of particle speed updating; and embedding the comprehensive energy consumption model as a fitness function into a hybrid enhanced particle swarm optimization algorithm, performing real-time energy consumption evaluation, individual and global optimal path updating and path smooth optimization on each candidate path in an algorithm iteration process, and outputting an optimal navigation path of the unmanned surface vessel.

Description

Technical Field

[0001] This application relates to the field of path planning technology, and in particular to a method and apparatus for energy-optimal path planning of unmanned surface vessels in complex marine environments. Background Technology

[0002] Unmanned surface vessels (USVs) have been widely used in underwater exploration, maritime security, and marine surveying due to their advantages such as small size, low cost, high safety, and high degree of autonomy. However, energy depletion is one of the core causes of mission failure and platform loss. The limitations of the propulsion system have become a key bottleneck restricting their long-term reliable operation in complex marine environments. Therefore, conducting research on energy-optimal path planning in complex marine environments is of great practical significance for improving mission success rates, ensuring asset safety, and extending endurance.

[0003] Currently, metaheuristic optimization algorithms are the mainstream technology for path planning of unmanned surface vessels (USVs). Among them, particle swarm optimization (PSO) is widely used due to its simple structure and ease of implementation. At the same time, to address the shortcomings of traditional algorithms, various hybrid path planning frameworks that integrate different optimization paradigms are constantly emerging. By balancing the exploration and development capabilities of algorithms, they improve the convergence efficiency and navigation performance of path planning. However, existing research still has significant shortcomings: most schemes focus on minimizing the geometric length of the path, rely on simplified environmental models, and do not fully consider the complexity, dynamics, and multiphysics characteristics of the real ocean environment. Even if some studies incorporate ocean current factors, they do not comprehensively integrate the three key environmental elements of wind, waves, and currents. Furthermore, there is a gap in the accurate modeling of additional resistance and propulsion efficiency attenuation under severe sea conditions, which means that the geometrically shortest path in simulation may become the most energy-intensive route in actual applications.

[0004] Therefore, there is an urgent need for a method that integrates path planning technology with a high-fidelity environmental coupling model to solve the problem of synergistic balance between geometric optimization and energy efficiency optimization, so as to achieve autonomous, efficient and energy-saving navigation of unmanned surface vessels in real marine environments. Summary of the Invention

[0005] In view of this, this application provides an energy-optimal path planning method and apparatus for unmanned surface vessels in complex marine environments. By integrating path planning technology with a high-fidelity environmental coupling model, the method solves the problem of synergistic balance between geometric optimization and energy efficiency optimization, thereby enabling unmanned surface vessels to navigate autonomously, efficiently and energy-savingly in real marine environments.

[0006] Specifically, this application is implemented through the following technical solution:

[0007] The first aspect of this application provides an energy-optimal path planning method for unmanned surface vessels in complex marine environments, the method comprising:

[0008] A comprehensive energy consumption model is constructed that couples the spatiotemporal effects of wind, waves, and currents. The comprehensive energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. Among them, the ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism.

[0009] This paper proposes a multi-dimensional improvement to the traditional particle swarm optimization algorithm, constructing a hybrid enhanced particle swarm optimization algorithm. Specifically, a safe space set and adjacency graph are built using a grid method; the A* algorithm is used to generate the initial optimal path and apply bounded random perturbation and safety correction to the target particle waypoints; B-spline technology is used to smooth the iteratively generated paths, and the smoothed optimal solution is resampled and re-injected into the population; the inertia weight and learning factor are dynamically adjusted based on the number of iterations, and a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term for particle velocity updates.

[0010] The comprehensive energy consumption model is embedded as a fitness function into the hybrid enhanced particle swarm optimization algorithm. During the algorithm iteration process, real-time energy consumption evaluation, individual and global optimal path updates, and path smoothing optimization are performed on each candidate path to output the optimal navigation path of the unmanned surface vessel.

[0011] A second aspect of this application provides an energy-optimal path planning device for unmanned surface vessels in complex marine environments, the device comprising a construction module, an improvement module, and a determination module;

[0012] The construction module is used to construct an energy consumption model that couples the spatiotemporal effects of wind, waves, and currents. The energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. The ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism.

[0013] The improved module is used to improve the traditional particle swarm optimization algorithm in multiple dimensions and construct a hybrid enhanced particle swarm optimization algorithm. Specifically, a safe space set and adjacency graph are constructed using a grid method; the A* algorithm is used to generate the initial optimal path and perform bounded random perturbation and safety correction on the target particle waypoints; B-spline technology is used to smooth the iteratively generated path, and the smoothed optimal solution is resampled and re-injected into the population; the inertia weight and learning factor are dynamically adjusted based on the number of iterations, and a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term for particle velocity updates.

[0014] The determining module is used to embed the energy consumption model as a fitness function into the hybrid enhanced particle swarm optimization algorithm, and output the optimal navigation path of the unmanned surface vessel through real-time energy consumption evaluation of each candidate path, individual and global optimal path updates, and path smoothing optimization during the algorithm iteration process.

[0015] The energy-optimal path planning method and device for unmanned surface vessels in complex marine environments provided in this application, in the first aspect, constructs a comprehensive energy consumption model that couples the spatiotemporal effects of wind, waves, and currents, and integrates the ocean current effect into the quantification of various resistance energy consumptions through a coupling mechanism. By linking the unmanned surface vessel's ground velocity and water velocity through vector relationships, the ocean current effect directly acts on the water velocity, thereby accurately affecting the quantification results of the basic water resistance energy consumption, wind-induced additional resistance energy consumption, and wave-induced additional resistance energy consumption. At the same time, it covers the energy consumption of the control system, realizing the comprehensive and accurate quantification of energy consumption in complex marine environments, and avoiding the energy consumption estimation deviations caused by traditional models that ignore dynamic environmental changes or ocean current coupling effects. Secondly, by making multi-dimensional improvements to the traditional particle swarm optimization algorithm, a safe space set and adjacency graph are constructed using the grid method, and the initial path is generated by the A* algorithm. Combined with a heuristic initialization method of bounded random perturbation and safety correction, the safety of the initial population is guaranteed and the population diversity is improved, laying the foundation for the algorithm to converge quickly. The application of B-spline technology ensures the continuity of path curvature and satisfies navigation constraints. Dynamically adjusting the inertia weight and learning factor enables the algorithm to maintain strong global search capability in the early stage and enhance local search accuracy in the later stage. The secondary adjustment of the cognitive term by the dual guiding factor further optimizes the particle search strategy and effectively avoids the algorithm from getting trapped in local optima. Thirdly, a precise quantified comprehensive energy consumption model is embedded as a fitness function into the improved hybrid enhanced particle swarm optimization algorithm. This enables the algorithm to select the optimal path based on a real and comprehensive energy consumption assessment during the iteration process. At the same time, through path smoothing optimization and dynamic updates of individual and global optimal paths, the algorithm achieves synergistic optimization of energy consumption optimization, path safety, and smoothness. The final output of the optimal navigation path for the unmanned surface vessel not only meets the minimum energy consumption requirements in complex marine environments but also has good navigation feasibility and obstacle avoidance safety. This solves the problems of mismatch between geometric optimality and actual energy consumption optimality, poor adaptability to complex environments, and the algorithm's tendency to get trapped in local optima in traditional path planning. Attached Figure Description

[0016] Figure 1 A flowchart of the energy-optimal path planning method for unmanned surface vessels in complex marine environments provided in Embodiment 1 of this application;

[0017] Figure 2 This is a schematic diagram of the energy-optimal path planning device for unmanned surface vessels in complex marine environments provided in Embodiment 2 of this application. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0020] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0021] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0022] Figure 1 This is a flowchart illustrating the energy-optimal path planning method for unmanned surface vessels in complex marine environments, provided in Embodiment 1 of this application. Please refer to... Figure 1 The method provided in this embodiment may include:

[0023] S101. Construct a comprehensive energy consumption model that couples the spatiotemporal effects of wind, waves, and currents.

[0024] The comprehensive energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. Among them, the ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism.

[0025] Specifically, the integrated energy consumption model is a mathematical model designed for unmanned surface vessels (USVs) navigation scenarios. It can accurately quantify the total energy consumption in complex marine environments. Its core feature is the coupling of the spatiotemporal dynamic effects of wind, waves, and currents, rather than calculating energy consumption based on static simplified conditions. It can be directly used as the fitness function of path planning algorithms. The integrated energy consumption model is composed of four core energy consumption components. Among them, the water resistance basic energy consumption refers to the energy consumption of the USV in overcoming the basic frictional resistance and pressure difference resistance of the seawater on the hull during navigation. It is the core basic energy consumption of the hull's navigation, and its magnitude is directly related to the hull's speed relative to the water.

[0026] Wind-induced additional drag energy consumption refers to the energy consumption corresponding to the additional drag generated by air (wind) on the hull during navigation. Its magnitude is related to the relative speed and wind direction between the hull and the wind, and it is an additional energy consumption source that cannot be ignored in complex environments.

[0027] Wave-induced additional drag energy consumption refers to the energy consumption corresponding to the additional drag such as impact drag and wave-making drag generated by ocean waves on the hull. Its magnitude is related to the wave height, wave frequency, wave direction and the hull's navigation state, and is a key factor in the increase of energy consumption under severe sea conditions.

[0028] Control system energy consumption refers to the energy consumed by an unmanned surface vessel to maintain course stability, perform trajectory corrections, and drive servo motor / propeller responses, including control module operating energy consumption and actuator action energy consumption. It is a necessary energy consumption to ensure navigation stability.

[0029] Furthermore, the spatiotemporal effects of wind, waves, and currents refer to the dynamic changes in the intensity, direction, and mode of action of these three marine environmental elements with time and spatial location, rather than being fixed static parameters. From a temporal perspective, wind speed / direction changes with weather systems (such as gusts and sustained winds), wave height / frequency changes with tidal cycles and wind duration, and current speed / direction changes with tidal rise and fall and ocean current movement. From a spatial perspective, wind speed, wave height, and current speed differ in different navigation areas (e.g., wave field characteristics differ between nearshore and offshore areas, and current field intensity differs between straits and open seas).

[0030] The ocean current effect refers to the direct or indirect impact of ocean currents (including tidal currents, coastal currents, ocean currents, etc.) on the energy consumption of unmanned surface vessels. Its core manifestations are: ocean currents change the hull's velocity relative to the water through vector relationships (velocity relative to the water = velocity relative to the ground - ocean current velocity), thereby directly affecting the basic energy consumption of water resistance (water resistance is positively correlated with the square of the velocity relative to the water); ocean currents indirectly affect the relative velocity between the hull and the wind by changing the hull's velocity relative to the ground (vehicle-wind relative velocity = wind speed relative to the ground - velocity relative to the ground), thereby affecting the additional drag energy consumption caused by wind.

[0031] In specific implementation, a comprehensive energy consumption model coupling the spatiotemporal effects of wind, waves, and currents is constructed, including: establishing a dynamic environmental field model by decomposing different frequency components of wind, waves, and currents through spatiotemporal dependency functions; the dynamic environmental field model outputs the spatiotemporal distribution parameters of ocean currents, wind fields, and wave fields; based on the hydrodynamic parameters of the dynamic environmental field model, the correlation between the unmanned surface vessel's ground velocity and water velocity is established through vector relationships; based on the water velocity, the energy consumption corresponding to the basic water resistance during navigation is quantified, and a water resistance energy consumption model is constructed; based on the wind field parameters of the dynamic environmental field model, wind speed and direction parameters at different spatiotemporal locations are extracted, combined with the ground velocity obtained through the vector relationships. The relative velocity between the unmanned surface vessel and the wind is calculated. Based on this relative velocity, the energy consumption corresponding to wind-induced additional drag under the wind field environment is quantified, and a wind-induced additional energy consumption model is constructed. Based on the wave field parameters of the dynamic environmental field model, wave height, wave frequency, and wave direction parameters at different spatiotemporal locations in the dynamic environmental field are extracted. Wave spectrum theory is used to quantify the energy consumption corresponding to wave-induced additional drag under the wave field environment, and a wave-induced additional energy consumption model is constructed. The energy consumption during the unmanned surface vessel's heading control, track maintenance, and actuator response processes is integrated to construct the control system energy consumption. The water resistance energy consumption model, wind-induced additional energy consumption model, wave-induced additional energy consumption model, and control system energy consumption are summarized to form a comprehensive energy consumption model.

[0032] Specifically, firstly, the low-frequency deterministic components and high-frequency random components of wind, waves, and currents are decomposed using spatiotemporal dependency functions. The low-frequency components reflect large-scale predictable environmental mechanisms such as tidal cycles and weather-scale wind patterns, while the high-frequency components reflect unanalyzed small-scale environmental processes such as turbulent fluctuations. A dynamic environmental field model is established using this decomposition method. This model outputs the spatiotemporal distribution parameters of ocean currents (velocity, direction), wind fields (wind speed, direction), and wave fields (wave height, wave frequency, direction) at different spatiotemporal locations. Then, based on the hydrodynamic parameters output by the dynamic environmental field model, the equation "velocity relative to water = velocity relative to land - sea velocity" is used to... The vector relationship between "velocity" and "current velocity" is used to establish the correlation between the unmanned surface vessel's (USV) velocity relative to the ground and its velocity relative to the water. Combined with the USV's hull parameters (length, beam, draft), the energy consumption corresponding to the basic water resistance during navigation is quantified using a water resistance calculation formula, thus constructing a water resistance energy consumption model. Subsequently, based on the wind field parameters output by the dynamic environmental field model, wind speed and direction parameters at each spatiotemporal node along the USV's trajectory are extracted. Combined with the aforementioned velocity relative to the ground obtained through vector relationships (velocity relative to the ground = velocity relative to the water + ocean current velocity), the relative velocity of the USV to the wind is calculated using the vector operation "velocity relative to wind = wind speed relative to ground - velocity relative to the ground". The relative speed between the vessel and the wind is calculated, and then, based on the wind-induced additional drag calculation method and combined with parameters such as the vessel's wind-receiving area, the energy consumption corresponding to the wind-induced additional drag under this wind field environment is quantified to construct a wind-induced additional energy consumption model. Simultaneously, based on the wave field parameters output by the dynamic environmental field model, parameters such as effective wave height, peak wave frequency, and main wave direction at each spatiotemporal node on the navigation trajectory are extracted. An appropriate wave spectrum model (such as the JONSWAP wave spectrum model) is selected according to the actual sea state type. The extracted wave field parameters are used as input to construct a wave spectrum matching the current spatiotemporal wave field characteristics. This is then combined with the unmanned surface vessel's hull type parameters and navigation state parameters, and calculated through wave spectrum integration. The additional drag force generated by waves of different frequencies on the hull is calculated, and the energy consumption corresponding to the additional drag under wave field conditions is quantified to construct a wave-induced additional energy consumption model. Then, the control behavior during the unmanned surface vessel's navigation is analyzed, and the energy consumption of servo motor actions during heading control, attitude adjustment during track maintenance, and drive energy consumption during actuator response are statistically analyzed. These energy consumption data are integrated to construct the control system energy consumption. Finally, the quantitative results of the above water resistance energy consumption model, wind-induced additional energy consumption model, wave-induced additional energy consumption model, and control system energy consumption are summarized by summation to form a comprehensive energy consumption model that can comprehensively reflect the spatiotemporal coupling effect of wind, waves, and currents. The formulas for calculating water resistance and wind-induced additional drag can be found in relevant technical descriptions and will not be repeated here.

[0033] Optionally, a wave-induced additional energy consumption model is constructed, including: extracting wave field parameters corresponding to each spatiotemporal node on the unmanned surface vessel's (USV) navigation trajectory from a dynamic environmental field model; the wave field parameters include significant wave height, peak frequency, and main wave direction; selecting an appropriate JONSWAP wave spectrum model based on the sea state type of the actual navigation area, using the significant wave height and peak frequency as input parameters to construct a wave spectrum matching the wave field characteristics of the current spatiotemporal location; based on the constructed JONSWAP wave spectrum, combined with the USV's hull type parameters and navigation state parameters, calculating the additional resistance force generated by waves of different frequencies on the hull through wave spectrum integration; calculating the energy consumption corresponding to the wave-induced additional resistance at the corresponding spatiotemporal node based on the additional resistance force, combined with the USV's navigation speed and navigation time at the corresponding spatiotemporal node; and summarizing the wave-induced additional energy consumption quantification results of all spatiotemporal nodes on the navigation trajectory to form a wave-induced additional energy consumption model that changes dynamically with time and space.

[0034] In practice, wave field parameters corresponding to each spatiotemporal node on the preset navigation trajectory of the unmanned surface vessel (USV) are accurately extracted from the dynamic environmental field model. These parameters include significant wave height, peak frequency, and main wave direction. During the extraction process, abnormal fluctuation parameters caused by environmental disturbances are simultaneously removed to ensure the validity and accuracy of the input parameters. First, the sea state type of the actual navigation area of ​​the USV is determined through marine monitoring data or sea state forecast information (with a focus on whether it is a sea state with limited wind). Based on this sea state type, a suitable JONSWAP wave spectrum model is selected, and the filtered significant wave height is then used to determine the wave field parameters. Peak frequency is used as the core input parameter and substituted into the expression of the JONSWAP spectrum. By adjusting parameters such as the peak enhancement factor, a personalized spectrum that precisely matches the wave field energy distribution and frequency characteristics at the current spatiotemporal location is constructed. Based on the constructed JONSWAP spectrum, combined with the specific hull parameters of the unmanned surface vessel (including length, beam, draft, wetted surface area, etc.) and real-time navigation parameters (including water velocity, angle between the navigation direction and the main wave direction, etc.), the frequency range of the spectrum integral (covering the actual sea wave area) is determined. The main frequency range is used to perform numerical integration (such as trapezoidal integral and Simpson integral) to integrate the wave energy and hull interaction terms at different frequencies, and calculate the additional resistance force generated by waves at different frequencies on the hull. Then, the total additional resistance force of the waves on the hull is obtained by vector synthesis. Based on the calculated additional resistance force of the waves, combined with the real-time navigation speed of the unmanned surface vessel at the corresponding spatiotemporal node, the real-time power required to overcome the additional resistance of the waves at that node is first calculated using the power calculation formula (power = resistance × speed). Then, combined with the navigation time of the unmanned surface vessel at that spatiotemporal node (calculated from the path length and navigation speed between nodes), the energy consumption corresponding to the additional resistance of the waves at the corresponding spatiotemporal node is calculated using the energy calculation formula (energy = power × time). Finally, the wave-induced additional energy consumption quantification results of all spatiotemporal nodes on the unmanned surface vessel's navigation trajectory are summarized by summation. At the same time, the mapping relationship between energy consumption and spatiotemporal node coordinates is established to form a dynamic wave-induced additional energy consumption model that can reflect the energy consumption changes in different navigation positions and wave field environments at different times.

[0035] For example, in one embodiment, water resistance is expressed as:

[0036] ;

[0037] in, The water resistance of the unmanned surface vessel; , , This is the water resistance coefficient; This refers to the speed of the unmanned surface vessel in the water.

[0038] Wind-induced additional drag is expressed as:

[0039] ;

[0040] in, Add wind-induced drag to unmanned surface vessels; air density; This refers to the wind-induced drag coefficient. The relative angle of attack of the boat against the wind; The windward area of ​​the unmanned surface vessel; The relative speed between the unmanned surface vessel and the wind.

[0041] The calculation formula is:

[0042] ;

[0043] in, The relative velocity vector between the unmanned surface vessel and the wind; This is the wind's velocity vector relative to the ground; This is the ground velocity vector of the unmanned surface vessel.

[0044] Due to accurate calculation A large amount of wind tunnel data is required. To ensure the computational feasibility of the path planning algorithm, this application does not directly perform calculations. Instead, it uses parametric expressions to quantify the resulting energy consumption. :

[0045] ;

[0046] in, Added energy consumption due to wind for unmanned surface vessels; Energy conversion efficiency of unmanned surface vessel propulsion systems; air density; The wind-induced coefficient; The windward area of ​​the unmanned surface vessel; The angle between the wind and the unmanned surface vessel's direction of travel; For reference wind speed; The speed at which the unmanned surface vessel travels; This represents the travel time of the unmanned surface vessel in the corresponding wind field environment.

[0047] Wave-induced additional resistance is expressed as:

[0048] ;

[0049] ;

[0050] ;

[0051] in, The effective wave angle frequency; The peak wave angular frequency; The speed at which the unmanned surface vessel travels; It is the acceleration due to gravity; For significant wave height; This results in an additional increase in wave-induced resistance. The density of seawater; Captain of an unmanned surface vessel; The angle between the wind and the unmanned surface vessel's direction of travel.

[0052] The above formula shows that the additional resistance is approximately the same as... Proportional to, with Inversely proportional, this proportional relationship applies to most ship types. Furthermore, to capture the propulsion efficiency loss caused by wave motion, a model is established:

[0053] ;

[0054] ;

[0055] in, Energy conversion efficiency of unmanned surface vessel propulsion systems; To improve the basic efficiency of the system; The coefficient for wave-induced efficiency attenuation; For significant wave height; This leads to additional energy consumption; This results in an additional increase in wave-induced resistance. The speed at which the unmanned surface vessel travels; This represents the travel time of the unmanned surface vessel in the corresponding wind field environment.

[0056] In addition to altering water velocity through indirect coupling, the energy consumption caused by ocean currents can also be quantified separately:

[0057] ;

[0058] in, The ocean currents cause additional energy consumption for unmanned surface vessels; Energy conversion efficiency of unmanned surface vessel propulsion systems; The magnitude of the relative speed between the unmanned surface vessel and the ocean current; The angle between the direction of the ocean current and the direction of the boat's travel; The speed at which the unmanned surface vessel travels; Captain of an unmanned surface vessel; This represents the travel time of the unmanned surface vessel in the corresponding wind field environment.

[0059] The integrated energy consumption model can be expressed as:

[0060] ;

[0061] in, This represents the real-time total propulsion power of the unmanned surface vessel. Energy conversion efficiency of unmanned surface vessel propulsion systems; Water resistance; Added drag due to wind; This causes additional resistance due to waves; For the velocity of water.

[0062] Integrating over time yields the total energy consumption:

[0063] ;

[0064] in, The total energy consumption of the unmanned surface vessel from the start to the end of the journey; The starting time of the voyage; The end time of the voyage; This represents the real-time total propulsion power of the unmanned surface vessel. The time step for each time sub-interval.

[0065] To integrate with practical path planning algorithms, this application employs an equivalent modular parameterization method to discretize the energy consumption components:

[0066] ;

[0067] in, The total energy consumption of the unmanned surface vessel from the start to the end of the journey; Energy consumption due to water resistance; Added energy consumption due to wind; This leads to additional energy consumption; To control the system's energy consumption.

[0068] The method provided in this embodiment is based on a dynamic environmental field model established by decomposing wind, wave, and current components at different frequencies using a spatiotemporal dependency function. Subsequent models for water resistance, wind-induced additional energy consumption, and wave-induced additional energy consumption directly utilize the spatiotemporal distribution parameters of ocean current, wind field, and wave field output by this model. This ensures the consistency and timeliness of the environmental parameters upon which each energy consumption component is calculated, avoiding the energy consumption estimation bias caused by neglecting dynamic environmental changes in traditional static environmental models. Furthermore, centralized environmental parameter modeling reduces redundancy from multiple data sources, improving the overall computational efficiency and accuracy of the model. Simultaneously, by establishing a vector relationship between the unmanned surface vessel's velocity relative to the ground and its velocity relative to the water, the core impact of ocean currents is fully considered. Ocean currents not only directly alter the velocity relative to the water, thus affecting the quantification of water resistance energy consumption, but also... By indirectly influencing the relative speed of the vessel to the wind through ground velocity calculation, the quantification of wind-induced additional energy consumption can reflect the linkage effect of ocean currents. Furthermore, the calculation of wave-induced additional energy consumption also indirectly reflects the auxiliary influence of ocean currents due to the coupling relationship between ocean current and wave field parameters in the dynamic environmental field model. This achieves comprehensive penetration of ocean current effects in the calculation of various resistance energy consumption. Compared with traditional energy consumption models that ignore ocean currents or only consider the direct influence of ocean currents, this significantly improves the authenticity and completeness of energy consumption quantification in complex marine environments. Finally, by summarizing the four types of energy consumption to form a comprehensive energy consumption model, it can accurately capture the energy consumption variation law of unmanned surface vessels under the spatiotemporal coupling effect of wind, waves, and currents. This provides high-fidelity fitness function support for subsequent energy-optimal path planning and effectively solves the problem of mismatch between geometrically optimal path and actual energy-optimal path.

[0069] S102. Improve the traditional particle swarm optimization algorithm in multiple dimensions and construct a hybrid enhanced particle swarm optimization algorithm.

[0070] Specifically, a safe space set and adjacency graph are constructed using the grid method; the A* algorithm is used to generate the initial optimal path and perform bounded random perturbation and safety correction on the target particle waypoints; the B-spline technique is used to smooth the iteratively generated path, and the smoothed optimal solution is resampled and re-injected into the population; the inertia weight and learning factor are dynamically adjusted based on the number of iterations, and a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term of particle velocity update.

[0071] In specific implementation, a safe space set and adjacency graph are constructed using a grid method. The A* algorithm is used to generate an initial optimal path and perform bounded random perturbation and safety correction on the target particle waypoints. This includes: dividing the unmanned surface vessel's navigation search space into regular grids; performing obstacle collision detection on each grid point; selecting grid points that do not intersect with any obstacles to form a collision-free safe space set; based on the safe space set, constructing an adjacency graph for path search using grid points as nodes and Euclidean distances between nodes as edge weights; and running the A* algorithm on the adjacency graph with the starting and target points of the navigation mission as input to search for the starting and target paths. The discrete initial path with the minimum cost between the point and the target point is generated by spline interpolation to form a preset number of intermediate waypoints, thus forming the initial path of the first particle. For the target particle to be initialized, a bounded random perturbation within a preset range is applied based on the generated initial path waypoints to generate a set of initial waypoints for the target particle. The waypoints of the target particle generated after perturbation are checked one by one. If a waypoint falls into an obstacle area, the nearest neighbor safe point of the waypoint is searched in the safe space set and replaced with the nearest neighbor safe point. If no valid nearest neighbor safe point is found, the original waypoint is retained, and the initial path of the particle population is finally obtained.

[0072] Specifically, firstly, based on the unmanned surface vessel's navigation mission range, obstacle distribution density, and path planning accuracy requirements, the side length of the regular grid is set, and the entire navigation search space is uniformly divided into several adjacent regular grids. A unique coordinate identifier (such as row and column number or Cartesian coordinates) is defined for each grid point. For each grid point, a preset obstacle collision detection algorithm is invoked, inputting the spatial geometric parameters of known obstacles (such as vertex coordinates of polygonal obstacles, center coordinates and radius of circular obstacles). The algorithm determines whether the grid point coordinates fall within the spatial coverage area of ​​any obstacle (if it is a polygonal obstacle, the ray method is used to determine if the point is inside the polygon; if it is a circular obstacle, the distance from the point to the circle is calculated). (Whether the center distance is less than the radius) filters out grid points that do not intersect with any obstacles. The coordinates and identification information of these grid points are stored in a set to form a collision-free safe space set. Based on this safe space set, each grid point is used as a node for path search. The adjacency relationship between nodes is defined (e.g., only adjacent grid nodes in the vertical, horizontal, left, right, and diagonal directions are allowed to establish connections). The straight-line distance between each pair of adjacent nodes is calculated using the Euclidean distance formula. This distance is used as the weight of the corresponding edge to construct an adjacency graph containing the node set, edge set, and edge weight information, and stored as a data structure (such as an adjacency matrix or adjacency list) for easy use by the A* algorithm. The starting coordinates and target coordinates of the navigation mission are converted into the corresponding grid points. The identification information is used as input parameters for the A* algorithm. A heuristic function is set (e.g., based on the Euclidean distance or Manhattan distance from the node to the target point). During algorithm execution, the starting point is first added to the open list. Each time, the node with the smallest cost function (f(n) = g(n) + h(n), where g(n) is the actual cost from the starting point to node n, and h(n) is the estimated cost from node n to the target point) is selected from the open list. If this node is the target point, the search terminates; otherwise, its adjacent nodes are added to the open list according to the rules, and the cost information is updated. At the same time, the processed nodes are moved to the closed list until the discrete initial path with the smallest cost between the starting point and the target point is found. For this discrete initial path, the number of particle waypoints is required ( (Assuming each particle contains 20 waypoints), cubic spline interpolation is used to generate intermediate waypoints that meet the smoothness requirements between adjacent nodes of the discrete path. The starting point, intermediate waypoints, and target point are combined in sequence to form the complete initial path of the first particle. For the remaining target particles to be initialized, the coordinates of all waypoints of the generated initial path are read first, and the perturbation amplitude range of each waypoint is set (e.g., the maximum perturbation distance in the x-axis and y-axis directions does not exceed 3 grid side lengths). A pseudo-random number generator is used to generate a random perturbation value in the range of [-perturbation amplitude, perturbation amplitude] for the x-coordinate and y-coordinate of each waypoint. The original waypoint coordinates are added to the corresponding perturbation value to generate the initial waypoint set of the target particles.For each perturbed target particle waypoint, the obstacle collision detection process described above is repeated. If a waypoint is detected to have fallen into an obstacle area, a nearest neighbor safe point search is initiated. The Euclidean distance between the waypoint and all grid points in the safe space is calculated using the k-nearest neighbor algorithm (k=1). The grid point with the smallest distance is selected as the nearest neighbor safe point, and its coordinates replace the original perturbed waypoint. The replacement information is recorded. If, after the search, the distance between the waypoint and all grid points in the safe space is found to be greater than a preset threshold (e.g., 5 grid side lengths), meaning no effective nearest neighbor safe point is found, the original perturbed waypoint is retained to avoid path deviation from the reasonable search range due to over-correction. After all waypoints of the target particle have been verified and replaced, the initial path of the particle is formed according to the waypoint order. The target particle initialization process described above is repeated until an initial path for a particle population of a preset size (e.g., 50 particles) is generated. All paths satisfy the waypoint safety constraints and cover a reasonable search space.

[0073] For example, in one embodiment, a "safety point set" is first constructed using a raster scan map: the search space is... Divided into rules For each grid point, collision detection is performed, and all grid points that do not intersect with obstacles are recorded as a free space set. Based on this, an adjacency graph is constructed:

[0074] ;

[0075] in, This is an adjacency graph used for pathfinding; For free space sets; Let be the set of edges of the graph; , for Any two nodes in; This is the adjacency threshold.

[0076] The edge weights are Euclidean distances. For a given starting point... and target point In the diagram Run the A* algorithm to determine the discrete path with the minimum cost:

[0077] ;

[0078] in, The initial path obtained by searching using the A* algorithm; The first node of the path; Starting point; For the path's first One node; The target point.

[0079] The path is resampled using spline interpolation. These intermediate waypoints form the initial path of the first particle:

[0080] ;

[0081] in, To represent the first particle The coordinate vectors of each waypoint; , For the first The rectangular coordinates of each waypoint in a two-dimensional plane; Waypoint number; The total number of waypoints contained in a single particle.

[0082] For the remaining particles, regarding waypoints Apply a bounded random perturbation:

[0083] ;

[0084] in, For the first Coordinate vectors after waypoint disturbance To represent the first particle The coordinate vectors of each waypoint; For the first Random disturbance coefficients for each waypoint; For the first Disturbance direction vectors for each waypoint; The disturbance coefficients are assumed to follow a uniform distribution over the interval.

[0085] Optionally, B-spline technology is used to smooth the iteratively generated paths. The smoothed optimal solution is then resampled and reinjected into the population. This includes: selecting the individual optimal path and the population-wide optimal path from the iteratively generated next-generation candidate paths; extracting the discrete waypoint coordinates corresponding to the two types of optimal paths as the original input nodes for B-spline smoothing; setting the order of the B-spline; using the extracted discrete waypoint coordinates as control vertices; calculating and determining the curve parameters through B-spline basis functions; and constructing a B-spline smoothing curve that fits the trend of the original optimal path and has continuous curvature; and based on the navigation constraints of the unmanned surface vessel, the curve is then smoothed. The feasibility of the constructed B-spline smoothed curve is verified. If there are regions where the curvature of the curve exceeds the limit, the coordinates of the corresponding control vertices are adjusted, and the B-spline curve is re-optimized until all navigation constraints are met. For the verified B-spline smoothed curve, the smoothed curve is resampled according to the preset sampling interval using the equal arc length sampling method to generate a new discrete waypoint set with the same number as the original population particle waypoints, forming the smoothed optimal path. The smoothed optimal path obtained by resampling replaces the particle paths in the population with fitness values ​​lower than the preset value, or is directly injected into the population as new particles to update the population.

[0086] In practical implementation, from the new generation of candidate paths generated by particle swarm optimization iteration, the fitness value of each path is calculated (using a comprehensive energy consumption model), and the individual optimal path and the global optimal path with the best fitness are selected. The coordinates of all discrete waypoints corresponding to these two types of optimal paths are extracted and arranged in path order, serving as the original input nodes for B-spline smoothing. The order of the B-spline is set according to the smoothness requirements of the unmanned surface vessel (e.g., a 3rd-order B-spline to ensure the curve is second-order continuous and differentiable). The extracted discrete waypoint coordinates are directly used as the control vertices of the B-spline, based on the de Boer algorithm or Cox-de... Boor's recursive formula calculates the B-spline basis functions for each interval. Combining the basis functions with the linear combination of control vertices determines the curve parameters, constructing a smooth B-spline curve that closely matches the original optimal path trend and has continuous curvature. Based on the unmanned surface vessel's navigation constraints (including maximum turning angle, maximum curvature, minimum turning radius, etc.), the curvature values ​​at each point of the smooth B-spline curve are calculated using numerical differentiation, and the curve is verified segment by segment for feasibility. If a region with excessive curvature is detected, the corresponding control vertex is located, and the B-spline curve is recalculated by fine-tuning the x and y coordinates of these control vertices (e.g., translating in the direction of decreasing curvature). This verification and adjustment process is repeated until the curvature of all segments of the curve meets the navigation constraints. For the verified B-spline smoothed curve, the coordinates of the sampling points on the curve are calculated using an equal arc length sampling algorithm according to a preset sampling interval (consistent with the waypoint spacing of the original population particles). By accumulating the curve arc length and recording the corresponding point coordinates when the preset interval is reached, a new discrete waypoint set with the same number as the waypoints of the original population particles is generated. The waypoints are arranged in the sampling order to form the smoothed optimal path. The fitness values ​​of all particles in the population are calculated and compared with a preset threshold. The smoothed optimal path obtained by resampling replaces the paths of particles in the population with fitness values ​​lower than the threshold, or directly adds them to the population as new particles, updates the population size and keeps the total number of particles unchanged, and completes the re-injection of the smoothed path into the population.

[0087] Optionally, the inertia weight and learning factor are dynamically adjusted based on the number of iterations, including: determining the total number of iterations of the algorithm, as well as the initial values ​​of the inertia weight, cognitive learning factor, and social learning factor; based on the ratio of the current iteration number to the total number of iterations, for the inertia weight, starting from the initial value, gradually decreasing as the number of iterations increases by multiplying the initial value by 1 and subtracting 0.5 times the ratio; for the cognitive learning factor, starting from the initial value, gradually decreasing as the number of iterations increases by multiplying the initial value by 1 and subtracting the ratio; for the social learning factor, starting from the initial value, gradually increasing as the number of iterations increases by multiplying the initial value by the ratio.

[0088] Specifically, firstly, based on the path optimization requirements of the particle swarm optimization algorithm, the total number of iterations (e.g., 100, 200) is pre-set, along with initial values ​​for the inertia weight (e.g., 0.9), the cognitive learning factor (e.g., 2.0), and the social learning factor (e.g., 0.5). During each iteration, the ratio of the current iteration to the total number of iterations (referred to as the iteration progress ratio) is calculated in real-time. Regarding the inertia weight, based on the set initial value, the inertia weight is gradually reduced as the number of iterations increases, following the calculation rule of "current inertia weight = initial value × [1 - (iteration progress ratio × 0.5)]", to maintain strong global search capability in the early stages and enhance local search accuracy in the later stages. Regarding cognitive learning... The cognitive learning factor, based on a set initial value, is calculated according to the rule "current cognitive learning factor = initial value × (1 - iteration progress ratio)". As the number of iterations increases, the cognitive learning factor gradually decreases to reduce the particle's dependence on its historical best position and avoid getting trapped in local optima later. The social learning factor, also based on a set initial value, is calculated according to the rule "current social learning factor = initial value × iteration progress ratio". As the number of iterations increases, the social learning factor gradually increases to enhance the particle's ability to learn the global best position of the population, guiding the population to converge quickly to the optimal solution later. Each iteration updates the values ​​of these three parameters in real time and substitutes the updated parameters into the iteration formulas for particle velocity and position until the algorithm completes the total number of iterations.

[0089] For example, in one embodiment, the adjustment of the inertia weight and the learning factor can be expressed as:

[0090] ;

[0091] ;

[0092] ;

[0093] in, The adjusted inertia weight; This is the initial value for the inertia weight; This represents the current iteration number; This represents the total number of iterations. This refers to the adjusted cognitive learning factor; This is the initial value for the cognitive learning factor; The adjusted social learning factor; This is the initial value for the social learning factor.

[0094] Optionally, a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term for particle velocity update, including: calculating the distance between the current position of each particle and the optimal position of the individual particle; using the ratio of the distance to a preset reference value as input, calculating the output of the Sigmoid function; combining the ratio of the distance to the preset reference value, a preset adjustment coefficient, the product of the first guiding factor and the output of the Sigmoid function, and the correlation value of the second guiding factor and the original calculation result of the cognitive term to obtain a scaling factor; in the cognitive term calculation stage of particle velocity update, the scaling factor is integrated into the cognitive term calculation process to perform a secondary correction on the cognitive term; during the iteration process, as the number of iterations increases, the value of the first guiding factor is gradually decreased and the value of the second guiding factor is increased.

[0095] Specifically, in each iteration of the particle swarm optimization algorithm, the average distance between the set of waypoint coordinates corresponding to the current position of each particle and the set of waypoint coordinates corresponding to the best position of that particle in its history is first calculated using the Euclidean distance formula, thus obtaining the distance between the current position of each particle and the best position of that particle. A preset reference value for this distance is set in advance (determined based on the range of the navigation search space and the path optimization accuracy). The ratio of the above distance of each particle to the preset reference value is calculated, and this ratio is used as an input parameter to the Sigmoid function to calculate the output result of the Sigmoid function for each particle. An adjustment coefficient is set in advance (e.g., a fixed value between 0.8 and 1.2), and the initial values ​​of the first guiding factor and the second guiding factor are set simultaneously (e.g., the initial value of the first guiding factor is 1.5, and the initial value of the second guiding factor is 0.5). The scaling factor for each particle is calculated using a formula. In the cognitive term calculation stage for particle velocity updates (the original calculation logic for the cognitive term is "cognitive learning factor × random number × (individual optimal position - current position)"), the calculated scaling factor is multiplied by the original calculation result of the cognitive term to perform a secondary correction, resulting in the adjusted cognitive term. During the algorithm iteration process, the value of the guiding factor is gradually adjusted according to a preset rule as the number of iterations increases. For example, each time it is iterated, the first guiding factor is decreased by multiplying the initial value by (1 - number of iterations / total number of iterations), while the second guiding factor is increased by multiplying the initial value by (1 + number of iterations / total number of iterations) (or a fixed step size is set for gradual adjustment) until the iteration ends. The scaling factor is calculated in real time and the cognitive term is corrected according to the above process in each iteration.

[0096] For example, in one embodiment, the distance between the current position of each particle and the optimal position of the individual particle is calculated:

[0097] ;

[0098] in, The distance between the current position and the optimal position of each particle; The optimal position for each individual particle; The current position of each particle.

[0099] Calculate the Sigmoid function:

[0100] ;

[0101] in, This is the output of the Sigmoid function; The distance between the current position and the optimal position of each particle; This is a preset reference value.

[0102] Calculate the scaling factor:

[0103] ;

[0104] in, This is the scaling factor; The distance between the current position and the optimal position of each particle; This is a preset reference value; Basic adjustment items; As the first guiding factor; For random disturbance terms; This is the output of the Sigmoid function; As the second guiding factor; For cognitive learning factors; The optimal position for each individual particle; The current position of each particle.

[0105] Optionally, the method further includes: processing the fitness evaluation task of each particle in parallel through multi-threading, pre-generating the random number matrix required for algorithm iteration; firstly screening candidate paths through geometric bounding boxes and region affiliation judgment, and performing collision detection and path correction only on paths suspected of having collisions; generating a random step size that conforms to the Lévy β stable distribution through the Mantenia method, and adding the random step size to the velocity component of the particle pointing to the global optimal position of the population according to the dimension.

[0106] Specifically, firstly, a multi-threaded processing environment is configured. The number of parallel threads is set according to the population size and the number of hardware cores. Each particle in the particle population is assigned to an independent thread. A random number matrix (containing random parameters for velocity updates, perturbation calculations, etc.) required for the entire algorithm iteration process is pre-generated and stored in shared memory. Each thread reads the random numbers as needed. For all candidate paths generated in the iteration, the minimum geometric bounding box of each path is constructed (the bounding box boundary is determined by calculating the extreme values ​​of the x and y coordinates of all waypoints on the path). At the same time, based on the regional classification of obstacles (the navigation space is divided into safe areas and suspected collision areas according to the distribution of obstacles), the candidate paths are initially screened by judging whether the bounding box of the path overlaps with the bounding box of the obstacle and whether the path falls into the suspected collision area. Paths that do not overlap with any obstacle bounding box and are located in the safe area are retained (directly judged as non-collision paths, no further detection is required). Only the bounding boxes are selected. Identifying potential collision paths that overlap or fall into suspected collision areas; for each potential collision path, verifying the collision relationship between each waypoint and obstacle on the path (using ray casting or distance judgment method); if a collision is detected, correcting it by replacing the nearest neighbor safe point or replanning the path segment; employing the Mantenia method, generating two independent sets of random numbers u and v that follow a standard normal distribution, calculating a random step size that conforms to the Lévy β stable distribution based on the formula, and decomposing the random step size into corresponding components along the x and y dimensions; in the calculation stage of the social term (the velocity component pointing to the global optimal position of the population) for particle velocity updates, adding the decomposed random step size component to the social term velocity component to complete the adjustment of the social term velocity; each parallel thread synchronously completes the fitness evaluation of the assigned particles, and after the evaluation is completed, summarizes the fitness results of all particles; throughout the process, the initial screening, collision detection correction, and velocity component adjustment processes are executed cyclically until the algorithm iteration ends.

[0107] For example, in one embodiment, the Lévy β-stable random step size is generated for each dimension of each generation using the Mantenia method:

[0108] ;

[0109] in, The generated Lévy β-stable random step size; Let be a random variable that follows a normal distribution; Let be a random variable that follows a standard normal distribution; for Follows a mean of 0 and a variance of The normal distribution; for It follows a standard normal distribution; Let be the shape parameter of the Lévy distribution.

[0110] Add the random step size along the dimension to the velocity component pointing to the global optimum:

[0111] ;

[0112] in, For the first The particle velocity in the next iteration; For the first The particle velocity in the next iteration; The weighting coefficients for the Levy step size; The generated Lévy β-stable random step size; This represents the globally optimal position for the population. This represents the particle's current position.

[0113] S103. The comprehensive energy consumption model is embedded as a fitness function into the hybrid enhanced particle swarm optimization algorithm. During the algorithm iteration process, real-time energy consumption evaluation, individual and global optimal path updates, and path smoothing optimization are performed on each candidate path to output the optimal navigation path of the unmanned surface vessel.

[0114] In specific implementation, the core parameters of the hybrid enhanced particle swarm optimization algorithm are set, and multiple initial candidate paths are obtained based on heuristic initialization of the particle swarm. The comprehensive energy consumption model is used as the fitness function, and the spatiotemporal information of the flight trajectory and dynamic environmental field parameters corresponding to each initial candidate path are input. The total energy consumption of each initial candidate path is calculated in real time through the comprehensive energy consumption model to obtain the fitness value of each path. The fitness value of each particle's current candidate path is compared with the particle's historical best fitness value. If the current value is better, the current path is updated to the particle's individual best path. The individual best paths of all particles are summarized, and the fitness values ​​are compared to select the best paths. The path with the optimal stress value is taken as the global optimal path for the population. Based on the dynamic parameter adjustment method and the dual guiding factor adjustment mechanism, the velocity and position of each particle are updated by combining the current individual optimal path and the global optimal path to generate a new generation of candidate paths. For the new generation of candidate paths, B-spline technology is used for curve fitting to control the path curvature to continuously meet the unmanned surface vessel's navigation constraints. At the same time, the feasibility of path obstacle avoidance is verified by a fast collision detection strategy, and paths that do not meet the constraints are corrected. It is determined whether the current iteration number has reached the preset maximum iteration number. After the iteration terminates, the final global optimal path obtained is taken as the optimal navigation path for the unmanned surface vessel.

[0115] Specifically, the core parameters of the hybrid enhanced particle swarm optimization algorithm are first set, including particle swarm size, maximum number of iterations, initial value and adjustment range of inertia weight, initial values ​​of cognitive learning factor and social learning factor, B-spline order, collision detection threshold, etc. Simultaneously, a safe space set and adjacency graph are constructed based on the grid method. The initial optimal path is generated using the A* algorithm, and bounded random perturbations and safety corrections are applied to the target particle waypoints to complete the heuristic initialization of the particle swarm, resulting in multiple initial candidate paths. A comprehensive energy consumption model coupling the spatiotemporal effects of wind, waves, and current is used as the fitness function. For each initial candidate path, its corresponding spatiotemporal trajectory information (including coordinates of each waypoint, navigation...) is extracted. The energy consumption model is input into the integrated energy consumption model, which calculates and summarizes the energy consumption of each initial candidate path in real time through the separate calculations of water resistance energy consumption, wind-induced additional energy consumption, wave-induced additional energy consumption, and control system energy consumption. This total energy consumption is used as the fitness value of each path (the lower the energy consumption, the better the fitness). The fitness value of each particle's current candidate path is compared with the particle's historical best fitness value (the historical best fitness value is the current path fitness value at the initial iteration). If the current fitness value is better, the current candidate path is updated to the particle's individual optimal path. This process continues until all particles have achieved their individual optimal paths. After path determination, the individual optimal paths of all particles are summarized, and the fitness values ​​of each path are compared one by one. The path with the highest fitness value (lowest total energy consumption) is selected as the global optimal path for the population. Based on a preset dynamic parameter adjustment method (dynamically adjusting inertia weights and learning factors with the number of iterations) and a dual-guidance factor adjustment mechanism (calculating scaling factors to perform secondary corrections on cognitive terms), the individual optimal path of the current particle and the global optimal path for the population are combined. Substituting these into the particle velocity update formula (incorporating the velocity update formula of the Lévy step size) and the position update formula, the new velocity and new position of each particle are calculated to generate a new generation of candidate paths. From the new generation of candidate paths, the individual optimal paths and the population are selected. The global optimal path is extracted, and its discrete waypoint coordinates are used as control vertices. The B-spline order is set, and a smooth B-spline curve is constructed through basis function calculation. The curvature of each point of the curve is calculated by numerical differentiation to verify whether it meets the navigation constraints such as the maximum turning angle and minimum turning radius of the unmanned surface vessel. If there are areas with excessive curvature, the coordinates of the corresponding control vertices are adjusted and the curve is re-optimized to ensure the continuity of the path curvature. At the same time, a fast collision detection strategy is adopted. Candidate paths are initially screened by geometric bounding boxes and region attribution judgment. Only path-by-path collision detection is performed on suspected collision paths. Paths with detected collisions are corrected by replacing the nearest neighbor safe point or replanning the path segment to ensure the feasibility of obstacle avoidance.After each iteration, it checks if the current iteration count has reached the preset maximum iteration count. If not, it returns to the process of "calculating fitness values ​​- updating individual and global optimal paths - updating particle velocity and position - path smoothing and collision detection correction" to continue iterating. If the maximum iteration count has been reached, the iteration terminates, and the final globally optimal path of the population is taken as the optimal navigation path for the unmanned surface vessel.

[0116] The method provided in this embodiment, in the first aspect, in the process of constructing the comprehensive energy consumption model, establishes a dynamic environmental field model by decomposing wind, wave, and current components of different frequencies through spatiotemporal dependency functions, outputs real-time spatiotemporal distribution parameters of ocean current, wind field, and wave field, and then associates the unmanned surface vessel's ground velocity and water velocity based on vector relationships, so that the ocean current effect directly acts on the water velocity to accurately quantify the basic energy consumption of water resistance. Combined with wind field parameters, the relative speed of the vessel to the wind is calculated to quantify wind-induced additional energy consumption. The JONSWAP spectral model and spectral integral are used to quantify wave-induced additional energy consumption. At the same time, the energy consumption of the control system is integrated. The resulting comprehensive energy consumption model can fully capture the dynamic changes of each energy consumption component in a complex marine environment, avoid the energy consumption estimation deviation caused by the neglect of environmental coupling effects in traditional static models, provide a high-fidelity fitness function for path optimization, and ensure that the subsequently selected path meets the actual optimal energy consumption requirements.

[0117] Secondly, in the multi-dimensional improvement of the particle swarm optimization algorithm, a safe space set and adjacency graph are constructed by using the grid method, and the initial path is generated by the A* algorithm. Heuristic initialization is achieved by combining bounded random perturbation and safety correction, which not only ensures the safety of the initial population but also improves diversity, laying the foundation for the algorithm to converge quickly. The application of B-spline technology makes the path curvature continuous and meets the navigation constraints of the unmanned surface vessel. The inertia weight and learning factor are dynamically adjusted based on the number of iterations (the inertia weight is gradually reduced, the cognitive learning factor is gradually reduced, and the social learning factor is gradually increased) to balance the global search and local search capabilities of the algorithm. The secondary adjustment of the cognitive term by the dual guiding factors (combining distance and Sigmoid function output to calculate the scaling factor) further optimizes the particle search strategy, effectively avoiding the algorithm from getting trapped in local optima and improving the accuracy and efficiency of path optimization.

[0118] Thirdly, in generating random step sizes to optimize velocity components, a random step size conforming to the Lévy β stable distribution is generated using the Mantenia method. This random step size is then added dimensionally to the velocity component of the particle pointing to the global optimum position of the population. The "heavy-tailed" characteristic of the Lévy distribution is used to enhance the particle's exploration ability in the solution space, enabling the particle to jump out of the local optimum region to search for a better path. At the same time, multi-threaded parallel processing of fitness evaluation tasks and a fast collision detection strategy (first screening suspected collision paths through geometric bounding boxes and then accurately detecting them) are used to improve the algorithm's computational efficiency while ensuring the feasibility of path obstacle avoidance. Ultimately, the optimal navigation path output by the algorithm has the characteristics of lowest energy consumption, safe navigation, and smooth feasibility, solving the problem of mismatch between geometric optimum and actual energy consumption optimum in traditional path planning.

[0119] Corresponding to the aforementioned embodiment of the energy-optimal path planning method for unmanned surface vessels in complex marine environments, this application also provides an embodiment of an energy-optimal path planning device for unmanned surface vessels in complex marine environments.

[0120] Figure 2 This is a schematic diagram of the energy-optimal path planning device for unmanned surface vessels in complex marine environments provided in Embodiment 2 of this application. Please refer to... Figure 2 The apparatus provided in this embodiment includes a construction module 210, an improvement module 220, and a determination module 230;

[0121] The construction module 210 is used to construct an energy consumption model that couples the spatiotemporal effects of wind, waves, and currents. The energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. The ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism.

[0122] The improved module 220 is used to improve the traditional particle swarm optimization algorithm in multiple dimensions and construct a hybrid enhanced particle swarm optimization algorithm. Specifically, it constructs a safe space set and adjacency graph using a grid method, generates an initial optimal path using the A* algorithm, and applies bounded random perturbation and safety correction to the target particle waypoints. It uses B-spline technology to smooth the iteratively generated path, resamples the smoothed optimal solution, and re-injects it into the population. It dynamically adjusts the inertia weight and learning factor based on the number of iterations and introduces a dual-guiding factor to perform a secondary adjustment to the cognitive term for particle velocity updates.

[0123] The determining module 230 is used to embed the energy consumption model as a fitness function into the hybrid enhanced particle swarm optimization algorithm, and output the optimal navigation path of the unmanned surface vessel through real-time energy consumption evaluation of each candidate path, individual and global optimal path updates, and path smoothing optimization during the algorithm iteration process.

[0124] The apparatus of this embodiment can be used to perform... Figure 1The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0125] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0126] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0127] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for energy-optimal path planning of unmanned surface vessels in complex marine environments, characterized in that, The method includes: A comprehensive energy consumption model is constructed that couples the spatiotemporal effects of wind, waves, and currents. The comprehensive energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. Among them, the ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism. This paper proposes a multi-dimensional improvement to the traditional particle swarm optimization algorithm, constructing a hybrid enhanced particle swarm optimization algorithm. Specifically, a safe space set and adjacency graph are built using a grid method; the A* algorithm is used to generate the initial optimal path and apply bounded random perturbation and safety correction to the target particle waypoints; B-spline technology is used to smooth the iteratively generated paths, and the smoothed optimal solution is resampled and re-injected into the population; the inertia weight and learning factor are dynamically adjusted based on the number of iterations, and a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term for particle velocity updates. The comprehensive energy consumption model is embedded as a fitness function into the hybrid enhanced particle swarm optimization algorithm. During the algorithm iteration process, real-time energy consumption evaluation, individual and global optimal path updates, and path smoothing optimization are performed on each candidate path to output the optimal navigation path of the unmanned surface vessel.

2. The method according to claim 1, characterized in that, A safe space set and adjacency graph are constructed using a grid method. The A* algorithm is used to generate the initial optimal path, and bounded random perturbations and safety corrections are applied to the target particle waypoints, including: The navigation search space of the unmanned surface vessel is divided into regular grids. Obstacle collision detection is performed on each grid point, and grid points that do not intersect with any obstacles are selected to form a collision-free safe space set. Based on the aforementioned safe space set, an adjacency graph for path search is constructed, with grid points as nodes and the Euclidean distance between nodes as edge weights. Using the starting point and target point of the navigation mission as input, the A* algorithm is run on the adjacency graph to search for the discrete initial path with the minimum cost between the starting point and the target point. A preset number of intermediate waypoints are generated through spline interpolation to form the initial path of the first particle. For the target particle being initialized, a bounded random perturbation within a preset amplitude range is applied based on the generated initial path points to generate the initial path point set of the target particle. The waypoints of the target particles generated after the disturbance are checked one by one. If a waypoint falls into the obstacle area, the nearest safe neighbor point of the waypoint is searched in the safe space and the original waypoint is replaced by the nearest safe neighbor point. If no effective nearest safe neighbor point is found, the original waypoint is retained, and the initial path of the particle population is finally obtained.

3. The method according to claim 1, characterized in that, A second adjustment is made to the cognitive terms for particle velocity updates by introducing dual guiding factors, including: Calculate the distance between the current position and the optimal position of each particle; Using the ratio of the distance to a preset reference value as input, the output of the Sigmoid function is calculated. The scaling factor is obtained by combining the ratio of the distance to the preset reference value, the preset adjustment coefficient, the product of the first guiding factor and the output of the Sigmoid function, and the correlation value of the second guiding factor and the original calculation result of the cognitive item. In the cognitive term calculation stage of particle velocity update, the scaling factor is incorporated into the cognitive term calculation process to perform secondary correction on the cognitive term; during the iteration process, as the number of iterations increases, the value of the first guiding factor is gradually reduced and the value of the second guiding factor is increased.

4. The method according to claim 1, characterized in that, The inertia weight and learning factor are dynamically adjusted based on the number of iterations, including: Determine the total number of iterations for the algorithm, as well as the initial values ​​for the inertia weight, cognitive learning factor, and social learning factor; Based on the ratio of the current iteration number to the total iteration number, the inertia weight is gradually reduced starting from the initial value, following the rule of multiplying the initial value by 1 and subtracting 0.5 times the ratio as the number of iterations increases; For the cognitive learning factor, starting from the initial value, it gradually decreases as the number of iterations increases by multiplying the initial value by 1 and subtracting the ratio. For the social learning factor, starting from the initial value, it gradually increases with the number of iterations by multiplying the initial value by the ratio mentioned above.

5. The method according to claim 1, characterized in that, Construct a comprehensive energy consumption model that couples the spatiotemporal effects of wind, waves, and currents, including: A dynamic environmental field model is established by decomposing different frequency components of wind, waves, and currents using a spatiotemporal dependency function; the dynamic environmental field model outputs the spatiotemporal distribution parameters of ocean currents, wind fields, and wave fields. Based on the hydrodynamic parameters of the dynamic environmental field model, the relationship between the unmanned surface vessel's velocity relative to the ground and its velocity relative to the water is established through vector relationships. Based on the velocity relative to the water, the energy consumption corresponding to the basic water resistance during navigation is quantified, and a water resistance energy consumption model is constructed. Based on the wind field parameters of the dynamic environmental field model, wind speed and wind direction parameters at different spatiotemporal locations are extracted. Combined with the ground velocity obtained through the vector relationship, the relative velocity between the unmanned surface vessel and the wind is calculated. Based on the relative velocity, the energy consumption corresponding to wind-induced additional drag under the wind field environment is quantified, and a wind-induced additional energy consumption model is constructed. Based on the wave field parameters of the dynamic environmental field model, the wave height, wave frequency, and wave direction parameters at different spatiotemporal locations in the dynamic environmental field are extracted. The energy consumption corresponding to the additional resistance of waves under the wave field environment is quantified by wave spectrum theory, and a wave-induced additional energy consumption model is constructed. Integrate the energy consumption of unmanned surface vessels in the processes of course control, track maintenance, and actuator response to construct a control system energy consumption model; The energy consumption models of water resistance, wind-induced additional energy consumption, wave-induced additional energy consumption, and control system energy consumption are summarized to form a comprehensive energy consumption model.

6. The method according to claim 5, characterized in that, Constructing a model for the additional energy consumption caused by the wave, including: Wave field parameters corresponding to each spatiotemporal node on the unmanned surface vessel's trajectory are extracted from the dynamic environmental field model; the wave field parameters include significant wave height, peak frequency, and dominant wave direction. Based on the sea state type of the actual navigation area, a suitable JONSWAP wave spectrum model is selected, and the effective wave height and peak wave frequency are used as input parameters to construct a wave spectrum that matches the wave field characteristics of the current spatiotemporal location. Based on the constructed JONSWAP wave spectrum, combined with the hull type parameters and navigation status parameters of the unmanned surface vessel, the additional resistance force generated by waves of different frequencies on the hull is calculated by wave spectrum integration. Based on the combined additional resistance force, and combined with the unmanned surface vessel's speed and travel time at the corresponding time and space node, calculate the energy consumption corresponding to the additional resistance of the waves at the corresponding time and space node. The wave-induced additional energy consumption quantification results of all spatiotemporal nodes along the navigation trajectory are summarized to form a wave-induced additional energy consumption model that changes dynamically with time and space.

7. The method according to claim 1, characterized in that, Output the optimal navigation path for the unmanned surface vessel, including: The core parameters of the hybrid enhanced particle swarm optimization algorithm are set, and multiple initial candidate paths are obtained based on the heuristic initialization of the particle swarm. The comprehensive energy consumption model is used as the fitness function. The spatiotemporal information of the flight trajectory and the dynamic environmental field parameters corresponding to each initial candidate path are input. The total energy consumption of each initial candidate path is calculated in real time through the comprehensive energy consumption model to obtain the fitness value of each path. Compare the fitness value of each particle's current candidate path with the particle's historical best fitness value. If the current value is better, update the current path as the particle's individual best path. Summarize the individual best paths of all particles, compare their fitness values, and select the path with the best fitness value as the global best path for the population. Based on the dynamic parameter adjustment method and the dual guiding factor adjustment mechanism, the velocity and position of each particle are updated by combining the current individual optimal path and the global optimal path, and a new generation of candidate paths are generated. For the next generation of candidate paths, B-spline technology is used for curve fitting to ensure that the curvature of the path continuously meets the navigation constraints of the unmanned surface vessel. At the same time, a fast collision detection strategy is used to verify the feasibility of obstacle avoidance of the path and to correct paths that do not meet the constraints. Determine whether the current iteration count has reached the preset maximum iteration count. After the iteration terminates, use the final global optimal path obtained by the population as the optimal navigation path for the unmanned surface vessel.

8. The method according to claim 1, characterized in that, The B-spline technique is used to smooth the iteratively generated paths, and the smoothed optimal solution is resampled and re-injected into the population, including: From the iteratively generated new generation of candidate paths, the individual optimal path and the population global optimal path are selected, and the discrete waypoint coordinates corresponding to the two types of optimal paths are extracted as the original input nodes for B-spline smoothing. The order of the B-spline is set, and the extracted discrete waypoint coordinates are used as control vertices. The curve parameters are determined by calculating the B-spline basis functions to construct a smooth B-spline curve that fits the original optimal path trend and has continuous curvature. Based on the navigation constraints of the unmanned surface vessel, the feasibility of the constructed B-spline smooth curve is verified; if there are regions where the curvature of the curve exceeds the limit, the coordinates of the corresponding control vertex are adjusted and the B-spline curve is re-optimized until all navigation constraints are met. For the B-spline smoothed curve that has passed the verification, the smoothed curve is resampled according to the preset sampling interval using the equal arc length sampling method, generating a new discrete waypoint set with the same number as the original population particle waypoints, forming the smoothed optimal path; The smoothed optimal path obtained from resampling replaces the particle path in the population with a fitness value lower than the preset value, or is directly injected into the population as a new particle to update the population.

9. The method according to claim 1, characterized in that, A hybrid enhanced particle swarm optimization algorithm is constructed by making multi-dimensional improvements to the traditional particle swarm optimization algorithm, including: The fitness evaluation task of each particle is processed in parallel by multi-threading, and the random number matrix required for algorithm iteration is generated in advance. First, candidate paths are initially screened by geometric bounding boxes and region affiliation judgment. Collision detection and path correction are only performed on paths that are suspected of having collisions. A random step size conforming to the Lévy β stable distribution is generated using the Mantenia method, and this random step size is added dimensionally to the velocity component of the particle pointing to the global optimal position of the population.

10. An energy-optimal path planning device for unmanned surface vessels in complex marine environments, characterized in that, The device includes a construction module, an improvement module, and a determination module; The construction module is used to construct an energy consumption model that couples the spatiotemporal effects of wind, waves, and currents. The energy consumption model comprehensively quantifies the basic energy consumption of water resistance, wind-induced additional resistance energy consumption, wave-induced additional resistance energy consumption, and control system energy consumption. The ocean current effect is integrated into the quantification process of various resistance energy consumptions through a coupling mechanism. The improved module is used to improve the traditional particle swarm optimization algorithm in multiple dimensions and construct a hybrid enhanced particle swarm optimization algorithm. Specifically, a safe space set and adjacency graph are constructed using a grid method; the A* algorithm is used to generate the initial optimal path and perform bounded random perturbation and safety correction on the target particle waypoints; B-spline technology is used to smooth the iteratively generated path, and the smoothed optimal solution is resampled and re-injected into the population; the inertia weight and learning factor are dynamically adjusted based on the number of iterations, and a dual guiding factor is introduced to perform a secondary adjustment on the cognitive term for particle velocity updates. The determining module is used to embed the energy consumption model as a fitness function into the hybrid enhanced particle swarm optimization algorithm, and output the optimal navigation path of the unmanned surface vessel through real-time energy consumption evaluation of each candidate path, individual and global optimal path updates, and path smoothing optimization during the algorithm iteration process.

Images