Deep-sea mining vehicle full-coverage path planning method based on improved biological activation neural network

Abstract

The invention relates to the technical field of deep-sea mining vehicle path planning, and discloses a deep-sea mining vehicle full-coverage path planning method based on an improved bio-activation neural network, and the method comprises the following steps: (1) analyzing the climbing dynamics of a deep-sea mining vehicle, and determining the passing time of each deep-sea mining region; (2) establishing a deep-sea mining operation environment model based on the determined passing time of each region; (3) establishing a grid heuristic function for deep-sea mining coverage operation requirements; and (4) in the deep-sea mining operation process, improving an escape path planning algorithm for dead zone escape demands, optimizing a target search mechanism and a fitness function, and performing point-to-point intelligent path planning on the mining vehicle. According to the method, an improved biological activation neural network algorithm and an escape algorithm are combined, deep-sea mining operation can be autonomously, accurately and efficiently completed under the condition that various seabed obstacles exist, and safety, energy conservation and mining efficiency are guaranteed.

Description

Technical Field

[0001] The present invention relates to a method for deep-sea mining vehicle (DSMV) path planning, belonging to the technical field of deep-sea mining vehicle (DSMV) path planning. Background Art

[0002] Cobalt nodules are a crucial component of deep-sea mineral resources, and deep-sea mining vehicles (DSMVs) are key equipment in their mining. Under the complex working conditions of the deep sea, relying solely on remote control to operate a mining vehicle is extremely difficult. The importance of deep-sea mineral resources is self-evident, and deep-sea mining vehicles (DSMVs) are essential and crucial equipment in their extraction. However, manual control is extremely challenging in the complex and ever-changing deep-sea environment.

[0003] In recent years, extensive research has been conducted on coverage path planning methods. This research primarily encompasses traditional coverage path planning methods (Traditional Algorithms), deep reinforcement learning methods (Deep Reinforcement Learning), and heuristic algorithms (Heuristic Algorithms). Traditional methods primarily include random coverage methods, spiral coverage methods, and reciprocating coverage methods based on unit decomposition. These methods are computationally efficient and easy to implement, but struggle with complex, irregular seabed terrain. Deep reinforcement learning methods can learn environmental characteristics and path planning patterns from extensive training data, generating more accurate and efficient paths and exhibiting good generalization capabilities. However, these methods require prohibitive computational costs. Heuristic algorithms, including swarm intelligence algorithms, evolutionary computation methods, and biologically activated neural networks, offer high computational accuracy, overcome the high computational costs of data-driven methods, effectively improve mining efficiency, and are suitable for complex, irregular seabed terrain.

[0004] There are still some unresolved issues in the coverage path planning algorithm currently used in DSMV, such as: 1. Most existing path planning methods focus only on factors that affect path performance, such as path length and number of turns, but ignore the interference of soil resistance and water flow during the operation of deep-sea mining vehicles. 2. Traditional coverage path planning is prone to dead zones, where the surrounding area is surrounded by obstacles or covered areas, resulting in low mining efficiency. 3. Traditional coverage path planning algorithms have low coverage efficiency and high operation time costs.

[0005] Therefore, it is extremely urgent to develop a new DSMV path planning system that can autonomously, safely, accurately and efficiently complete deep-sea mining operations in the presence of various seabed obstacles. Summary of the Invention

[0006] To address the current low efficiency of deep-sea mining coverage operations, this paper proposes a full-coverage path planning method for a deep-sea mining vehicle based on an improved biologically activated neural network. This method enables autonomous, safe, accurate, and efficient deep-sea mining operations in the presence of various seabed obstacles. The deep-sea mining vehicle is tracked.

[0007] The present invention provides a method for planning a full-coverage path for a deep-sea mining vehicle based on an improved biologically activated neural network, comprising the following steps: (1) Analyze the climbing dynamics of deep-sea mining vehicles and determine the travel time in each deep-sea mining area; (2) Establish a deep-sea mining operation environment model based on the determined transit time for each area; (3) Establishing a grid-inspired function for deep-sea mining coverage operations to improve the biologically activated neural network (BINN) and enhance the coverage efficiency of deep-sea mining vehicles; (4) During deep-sea mining operations, the escape path planning algorithm (ESC) is improved to meet the needs of escaping from the dead zone, the target search mechanism and fitness function are optimized, and point-to-point intelligent path planning is performed for the mining vehicle to solve the problem of escaping from the dead zone during the covering operation.

[0008] The process of analyzing the climbing dynamics of the deep-sea mining vehicle in step (1) and establishing the travel time of the deep-sea mining area is as follows: First, define the distribution law of ground pressure: ; in: is the ground pressure; G is the weight of the mining vehicle; B and L are the width and length of the deep-sea mining vehicle tracks respectively; α is the slope angle; In the actual climbing process, the center of gravity of the crawler deep-sea mining vehicle moves backward when climbing. The maximum shear stress and participating shear stress of the soft bottom are analyzed and defined as follows: ; in: τ(j m ) is the actual shear stress; is the maximum shear stress; λ is a scaling factor less than 1; ω is a constant; j m =i m x ,i m is the track slip ratio, x is the track ground contact length; k m is the substrate shear modulus; Theoretical driving force behind the neglect of grouser effect in deep-sea mining vehicles The analysis is performed and defined as follows: ; in: is the theoretical driving force; B and L are the width and length of the deep-sea mining vehicle tracks respectively; for substrate cohesion; is the ground pressure; is the internal friction angle of the substrate; λ is a scaling factor less than 1; ω is a constant; i m is the track slip ratio; x is the track ground contact length; is the substrate shear modulus; Calculate the difference between actual driving force and theoretical driving force , defined as follows: ; in, n , m is a non-zero constant; The ultimate driving force The definition is as follows: ; During the climbing process, the resistance encountered by the deep-sea mining vehicle is divided into ground squeezing resistance and water gravity; ground squeezing resistance The definition is as follows: ; in: is the cohesive deformation modulus; is the friction deformation modulus; B and L are the width and length of the deep-sea mining vehicle track, respectively is the longitudinal eccentricity; Water gravity component The definition is as follows: ; Where: G is the gravity of the mining vehicle; α is the slope angle; The deep-sea mining vehicle will be affected by the hydrodynamics during its travel. F h It is composed of water resistance and inertia force and is defined as follows: ; in: is the water resistance coefficient; is the frontal area of the mining vehicle; and are the mining vehicle's travel speed and the ocean current speed; is the added mass coefficient; is the size of the mining cart.

[0009] The final actual driving force of the above deep-sea mining vehicle , ground compression resistance , water gravity component and hydrodynamics F h Substitute the formula into Newton's second law, the formula is as follows: ; Where: G is the gravity of the deep-sea mining vehicle, g is the gravitational acceleration coefficient; , dv and dt Represent the differential forms of velocity and time respectively; By solving a first-order linear separable variable differential equation , we can get the running speed of the deep-sea mining vehicle under different slopes, and then use the following slope model formula to calculate the over-slope time of different slopes: ; in: α is the slope angle; h is the slope height; v m is the vehicle's travel speed; After calculating the slope time, the slope less than 5° is defined as a flat area, the slope between 5° and 10° is defined as a steep slope area, and the slope greater than 10° is defined as an infeasible area.

[0010] The process of establishing the deep-sea mining operation environment model in step (2) is as follows: The seabed topography model established by the numerical model is used to simulate the real deep-sea mining operation environment. The natural topography in the seabed environment is described by the following exponential function: ; in:( x d , y d ) represents the actual coordinates of the seabed; Z ( x d , y d ) represents the height of the seabed topography; ( x di , y di) represents the center coordinates of the hillock; h i It is a topographic parameter that determines the height of the sea hill; s is the number of sea mounds; x si and y si They are i A hillock in x Axis and y The decay rate along the axis.

[0011] The three-dimensional path is projected onto a two-dimensional plane, and the terrain of the mining area is divided into obstacle areas and barrier-free areas. The obstacle areas include high obstacle areas (such as steep slopes, steps, and other obstacles that mining vehicles cannot cross) and restricted access areas (such as mud and sand mixtures or ditches, etc.). Grids with slopes greater than 10° are set as high obstacle areas. The barrier-free areas are divided into flat areas and steep areas. Flat areas refer to barrier-free areas with slope angles between 0° and 5°, and steep areas refer to barrier-free areas with slope angles between 5° and 10°.

[0012] The process of establishing a grid heuristic function for deep-sea mining coverage operation requirements in step (3) is as follows: A. Establishing coverage status grid activity indicators : Set the grid status of the traveled grid to 0, the grid status of the untraveled grid to 500, and the grid status of the grid with obstacles to -500, and define the coverage status grid activity indicator. as follows: ; B. Establishing distance grid activity indicators : Set the distance grid activity index for: ; in: are the coordinates of the surrounding grid points; are the coordinates of the current point; C. Establishing Steering Grid Activity Indicators : Defining steering grid activity indicators for: ; in: Represents the value from the previous coordinate point ( ) points to the current coordinate point ( ) direction vector; Represents the value from the previous coordinate point ( ) points to the grid coordinate point ( ) direction vector; D. Synthesize the final grid heuristic function: Comprehensively consider the above three grid activity indicators , the final grid heuristic function is expressed as the following formula: ; in: Corresponding to three grid activity indicators The weight of .

[0013] According to the above heuristic function, the coverage path planning of the mining vehicle can comprehensively consider the coverage efficiency, driving distance and steering angle of the mining vehicle to ensure the safety and high efficiency of the mining operation.

[0014] The process of improving the escape path planning algorithm (ESC) in step (4) is as follows; A. Search for the escape target grid point: Assume that the mining vehicle is located at the dead zone position ( ), the position coordinates of the target grid point are ( ), then the distance from the dead zone position point to the target grid point The definitions are as follows: ; According to the above formula, the distance between all suitable target grids and the current dead zone grid is determined, and the grid point with a relatively small distance is selected as the final target grid point; B. Initialize the algorithm and population and establish the panic index and iterative process: a. Initialization algorithm and population: First initialize N A population of individuals, each of which consists of D dimensional vector Description, individual i In the j The value of dimension is given by: ; in: Represents an individual i In the j The location of the dimension, and Respectively represent j The lower and upper bounds of the dimension ensure that the initial position of each individual is randomly distributed in the feasible space; the random variable It is uniformly distributed between 0 and 1, reflecting the randomness of the initial decision during the evacuation process; After the population is initialized, the fitness function is used f Evaluate the fitness of each individual =f ( ), then sort the population in ascending order of fitness and store the individuals with the highest fitness in the elite pool E In the formula below, the parameter represents the number of potential safe exits identified by the crowd; ; These elite individuals represent the best potential solutions identified by the population and serve as reference points for subsequent iterations; b. Establish panic index and iterative process: The Escape Path Planning (ESC) algorithm simulates the iterative process to reflect the behavior changes of the crowd as the evacuation progresses. t The panic index P(t) is calculated at the beginning of the generation: ; The panic index P(t) reflects the overall panic level of the crowd. The higher the value, the more chaotic the behavior. t Gradually increase from 0 to the maximum number of iterations T , the index gradually decreases, simulating the crowd’s adaptation to the evacuation environment; C. When the number of iterations t ≤ T / 2 starts to explore the optimal solution: In the number of iterations t ≤ T / 2 exploration phase, in which T is the maximum number of iterations of the algorithm, t is the current iteration number, and the individuals are divided into three groups according to their fitness: calm, herd, and panic (specifically, the population is sorted in ascending order of fitness and the proportion is c =0.15, h =0.35 and p =0.5 is stratified into calm, follow-the-herd, and panic groups), which reflects the diverse reactions of the crowd during the evacuation, with some people remaining calm, some following the herd, and some falling into panic; Update calm group: Individuals in the calm group act rationally and move towards the center of group decision-making move: ; in: Represents the updated individual i In the j Dimensional location; Represents an individual i In the j Dimensional location; binary variable Determined by the Bernoulli distribution, partial updates are allowed to simulate the non-update of some dimensions due to crowding. Specifically, Generates 0 or 1 with equal probability; is the adaptive Levian weight, which uses the Levian distribution to simulate the step size of the exploration phase, defined as formula; For the calm group j The center of the dimension, that is, the average value of all calm individuals in the dimension; vector It is defined as follows: ; in: v c,j express j Dimensional calm individual movement vector; Represents an individual i In the j Dimensional location; is a randomly generated position within the calm group boundary, and Respectively indicate the The minimum and maximum values of all calm individuals in the dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, is a random variable that follows the standard normal distribution N(0,1); Update the herd group: The conforming individuals are influenced by the calm and panic groups' behaviors, and their position updates are based on the combined influence of the two groups: ; in, m 2 is through m 1 binary variables generated by the same mechanism; A randomly selected individual from the panic group represents the potential movement direction driven by panic; vector It is defined as follows: ; in, v h,j express j Dimensional movement vectors of conforming individuals; is a randomly generated position within the population group boundary, and Respectively represent the minimum and maximum values of all conforming individuals in this dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with a value of (0,1); Update panic group: Panic-driven individuals explore the solution space more randomly, influenced by potential exits (from the elite pool) and random directions from other individuals: ; in: An individual is randomly selected from the elite pool to represent a potential exit to which the panicking individual might go; is an individual randomly selected from the population, introducing randomness; the vector It is defined as follows: ; in, v p,j express j Movement vectors of dimensionally panicked individuals; is a randomly generated position within the panic group boundary, and Respectively represent the minimum and maximum values of all panic individuals in this dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with a value of (0,1); D. When the number of iterations t>T / 2, it enters the development phase: As the number of iterations exceeds T / 2, the algorithm enters the development phase, where all individuals are considered calm. The focus shifts to fine-tuning their positions based on the currently identified optimal solution. During this phase, individuals adjust their positions by moving closer to members of the elite pool (representing possible safe exits and the best solution identified in the previous iteration) and randomly selected individuals from the population, simulating a population gradually converging toward the identified optimal exit. The position update formula for this phase is: ; in, Represents an individual i In the j Dimensional location; is the position of a member of the elite pool, representing both a possible safe exit and one of the best solutions identified so far; is the position of a randomly selected individual; individuals adjust their positions by approaching elite pool members and randomly selected individuals, simulating the crowd gradually converging towards the identified optimal exit; E. Adaptive Levian Weights and Behavior Simulation: The individual step size is controlled by the adaptive Lev weight, simulating the different degrees of exploration and development at different stages of the algorithm; each dimension j Levy weights The calculation is as follows: ; in: Γ is the gamma function; β is a dynamic adjustment parameter that changes with the algorithm process and is defined as follows: ; in, is the initial value of β, which is set to 1.5. This adjustment allows the algorithm to make a wider range of exploratory moves in the initial stage (when β is small) and gradually transition to more refined exploitative moves (as β increases), reflecting the natural progression from panic-driven exploration to calm, rational decision-making; F. Determine the fitness function f And update the elite pool: Point-to-point path planning for deep-sea mining vehicles in deep-sea mining operations is a multi-objective nonlinear optimization problem that requires comprehensive consideration of factors such as distance cost, turning cost, and obstacle zone constraints. f The design is as follows: ; in: It is the time cost; is the cost of turning; is the obstacle constraint cost.

[0015] Time cost J L , is the path length of the deep-sea mining vehicle from the starting point to the end point, and the calculation formula is as follows: ; in:( ) indicates the previous coordinate point; ( ) represents the current coordinate point; v m represents the driving speed of DSMV, which is set to 0.83 m / s; k ( i ) means passing through a slope ( The ratio of the travel time of the grid to the travel time of the flat grid is obtained from the slope model formula through climbing dynamics analysis; Turning costs J A , is the sum of the steering angles of each path point of the deep-sea mining vehicle: ; in, Indicates the i -1 coordinate point to i The steering angle of the coordinate point.

[0016] Obstacle constraint cost J O Once a pathpoint is found within an obstacle area, the cost is set to infinity to reject the path: ; In the process of point-to-point static path planning for deep-sea mining vehicles, the above costs should be optimized as much as possible.

[0017] In each iteration, the fitness of each updated individual is recalculated. A greedy selection process is used to retain the better solutions: ; individual i The update rule can be described as: ; If the new fitness is better (i.e., smaller for the minimization problem), then the individual position update is Otherwise, keep the original position ; After each iteration, the elite pool E It will also be updated to ensure it contains the best solutions found so far.

[0018] Aiming at the needs of deep-sea mining operations, the present invention proposes a new full-coverage path planning method for deep-sea mining vehicles. This method combines an improved biologically activated neural network algorithm (BINN) and an escape strategy (ESC), taking into account the safety, energy saving, and mining efficiency of deep-sea mining operations. It can complete deep-sea mining operations autonomously, accurately, and efficiently in the presence of various seabed obstacles, ensuring safety, energy saving, and mining efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is the overall structural diagram of the coverage path planning motion framework in the deep-sea mining vehicle full coverage path planning method based on the improved biological activated neural network of the present invention.

[0020] Figure 2 This is a flow chart of the ESC algorithm in the full coverage path planning method for deep-sea mining vehicles based on the improved biologically activated neural network of the present invention. DETAILED DESCRIPTION

[0021] Aiming at the needs of deep-sea mining operations, this paper proposes a new full-coverage path planning method for deep-sea mining vehicles based on an improved biologically activated neural network. By combining the improved biologically activated neural network algorithm (BINN) and the escape strategy (ESC), it can achieve safe, energy-saving and efficient deep-sea mining operations.

[0022] The following combination Figure 1 The method of the present invention is described in detail.

[0023] 1. Dynamic analysis of deep-sea mining vehicles and establishment of seabed environment model.

[0024] 1. Calculation of regional travel time based on the climbing dynamics analysis of deep-sea mining vehicles.

[0025] In the soft soil environment of polymetallic nodule mining areas, the traction performance of mining vehicles is affected by the height of the track teeth, the distribution of ground pressure, and the adhesion characteristics of the substrate. This paper analyzes the climbing dynamics of mining vehicles and first defines the distribution of ground pressure: ;(1) in: is the ground pressure; G is the weight of the mining vehicle; B and L are the width and length of the mining vehicle tracks respectively; α is the slope angle.

[0026] In the actual climbing process, the center of gravity of the crawler mining vehicle moves backward when climbing. The maximum shear stress and participating shear stress of the soft bottom are analyzed and defined as follows: ; (2) in: τ(j m ) is the actual shear stress; is the maximum shear stress; λ is a scaling factor less than 1; ω is a constant; j m =i m x , i m is the track slip ratio, x is the track ground contact length; k m is the substrate shear modulus.

[0027] Theoretical driving force for ignoring the role of grouser in mining vehicles The analysis is performed and defined as follows: ;(3) in: is the theoretical driving force; B and L are the width and length of the mining vehicle track respectively; for substrate cohesion; is the ground pressure; is the internal friction angle of the substrate; λ is a scaling factor less than 1; ω is a constant; i m is the track slip ratio; x is the track ground contact length; is the substrate shear modulus.

[0028] Calculate the difference between actual driving force and theoretical driving force , defined as follows: ;(4) in, n , m is a nonzero constant.

[0029] The ultimate driving force The definition is as follows: (5) During the climbing process, the resistance encountered by the mining vehicle can be mainly divided into ground compression resistance and water gravity force. The ground compression resistance is defined as follows: ;(6) in: is the cohesive deformation modulus; is the friction deformation modulus; is the longitudinal eccentricity.

[0030] The water gravity component is defined as follows: ;(7) Where: G is the gravity of the mining vehicle; α is the slope angle.

[0031] Unlike land-based tracked vehicles, deep-sea mining vehicles are larger and have a frontal surface in seawater, which is much denser than air. Therefore, they are affected by hydrodynamic forces during their travels. Hydrodynamic resistance is composed of water resistance and inertial force, and is defined as follows: ;(8) in: F h It is water powered; is the water resistance coefficient; is the frontal area of the mining vehicle; and are the mining vehicle's travel speed and the ocean current speed; is the added mass coefficient; is the size of the mining cart.

[0032] Substituting the above force formulas (5), (6), (7) and (8) into Newton's second law, the formula is as follows: ;(9) Where: G represents the gravity of the mining vehicle, g is the gravitational acceleration coefficient; , dv and dt Represent the differential forms of velocity and time respectively. By solving the first-order linear separable variable differential equation The running speed of the deep-sea mining vehicle under different slopes can be obtained, and then the over-slope time of different slopes can be calculated using the following slope model formula: ;(10) in: h is the slope height; α is the slope angle; v m is the vehicle's travel speed.

[0033] In the thin, soft base of polymetallic nodules, the slope affects the ground pressure of mining vehicles, which in turn affects climbing performance. Therefore, the time required to climb a slope varies depending on the slope. Slope time calculations define slopes less than 5° as flat, those between 5° and 10° as steep, and those greater than 10° as infeasible.

[0034] 2. Deep-sea mining operation environment model.

[0035] The topography of deep-sea cobalt crust mining areas is very complex, consisting of exposed bedrock and a mixture of underwater mud, sand, and gravel. This paper uses a numerical model to simulate the actual deep-sea mining environment using a seabed topography model. The natural topography of the seabed environment is described by the following exponential function: ;(11) in:( x di , y di ) represents the center coordinates of the hillock; h i It is a topographic parameter that determines the height of the sea hill; s is the number of sea hills; further processing of the acquired terrain data can obtain the slope angle corresponding to each grid.

[0036] Deep-sea mining vehicles need to minimize repeated paths to ensure higher economic efficiency, thereby improving mining efficiency during seabed operations. Projecting a three-dimensional path onto a two-dimensional plane can effectively simplify the model for better full-coverage path planning. In order to enable mining vehicles to operate safely and efficiently, the terrain is divided into obstacle areas and obstacle-free mining areas. Obstacle areas mainly include the following two categories: one is high obstacles, such as steep slopes, steps, and other obstacles that mining vehicles cannot cross; the other is areas that restrict the passage of mining vehicles, such as mud and sand mixtures or ditches. In the present invention, grids with slopes greater than 10° are set as high-obstacle areas. Obstacle-free mining areas are divided into two categories: one is a flat mining area, that is, an obstacle-free area with a slope angle between 0° and 5°; the other is a steep mining area, that is, an obstacle-free area with a slope angle between 5° and 10°.

[0037] 2. Establishment of a full-coverage path planning motion framework based on an improved biologically activated neural network.

[0038] The requirements for mining coverage path planning are as follows: traverse all grids while avoiding obstacles and minimizing the repetition rate in the coverage operation. In this invention, a mining operation walking method based on a round-trip mode is adopted, which has high coverage rate, simple and easy-to-implement rules, and the ability to adapt to complex environments.

[0039] 1. Establish a grid heuristic function for deep-sea mining coverage operations.

[0040] Considering the actual working conditions, mining operation requirements and driving control problems of deep-sea mining vehicles, it is necessary to try to plan a non-repetitive straight line path for the mining vehicle to carry out deep-sea mining work. Taking into account the feasibility, efficiency, driving distance and turning factors of the mining vehicle, this paper proposes three grid activity indicators: .

[0041] (1) Establish coverage status grid activity indicators: First, in order to ensure the safety and mining efficiency of deep-sea mining vehicles, it is necessary to ensure that the mining vehicles avoid obstacles during the movement and try not to walk on repeated grids. A dynamic attribute grid is used to add a coverage status flag to each grid. The grid status of the grid that has been traveled is set to 0, the grid status of the grid that has not been traveled is set to 500, and the grid status of the grid with obstacles is set to -500, defining the grid activity index. as follows: ;(12) (2) Establishing distance grid activity indicators: Grid distance is also an important factor to consider in the mining vehicle motion planning process. The mining vehicle will give priority to the grid with the closest distance to walk, and gradually plan the global path from local to global. Therefore, set the grid activity index for: ;(13) in: are the coordinates of the surrounding grid points; are the coordinates of the current point.

[0042] (3) Establishing steering grid activity indicators: The mining vehicle motion control and steering consumes a lot of energy, so the path planning should ensure that the steering angle of the mining vehicle is as small as possible, defining the grid activity index for: ;(14) in: Represents the value from the previous coordinate point ( ) points to the current coordinate point ( ) direction vector; Represents the value from the previous coordinate point ( ) points to the grid coordinate point ( ) direction vector.

[0043] (4) Synthesize the final grid heuristic function.

[0044] Comprehensively consider the above three grid activity indicators , the final grid heuristic function can be expressed as the following formula: ;(15) in: Corresponding to three grid activity indicators According to the above heuristic function, the coverage path planning of the mining vehicle can fully consider the coverage efficiency, driving distance and steering angle of the mining vehicle to ensure the safety and high efficiency of the mining operation.

[0045] 2. Adopt escape path planning methods to meet the needs of escaping from deep-sea mining dead zones.

[0046] Deep-sea mining vehicles often encounter obstacles and covered areas during mining operations, often referred to as dead zones. To address this issue, the present invention utilizes an escape path planning (ESC) algorithm to intelligently plan point-to-point paths for mining vehicles, helping them escape these dead zones and improving overall mining efficiency.

[0047] (1) Search for the escape target grid point.

[0048] The target grid point search requires two characteristics: first, it must be untraversed; second, it must be close to a mining dead zone and be conducive to subsequent mining traversal. This selection of target grid points ensures the lowest repetition rate and maximizes economic benefits.

[0049] Assume that the mining vehicle is located at the dead zone position ( ), the position coordinates of the target grid point are ( ), then the distance from the dead zone position point to the target grid point The definitions are as follows: ; (16) According to the above formula, the distances between all suitable target grids and the current dead zone grid can be determined, and the grid point with a relatively small distance can be selected as the final target grid point.

[0050] (2) Use the escape algorithm to escape from the dead zone.

[0051] The ESC algorithm is unique in its approach, drawing on subtle dynamic changes in crowd behavior, which has not been fully explored in existing meta-heuristic algorithms. Specifically, the ESC algorithm utilizes the behaviors of different groups of people during the evacuation process (calm groups, herd groups, and panic groups) and incorporates these behaviors into the exploration and development phase of the algorithm. This approach not only provides an effective way to simulate the optimization process, but also enhances the algorithm's ability to avoid local optimal solutions and obtain better global solutions. The ESC algorithm flow chart is shown below. Figure 2 shown.

[0052] The ESC algorithm is particularly inspired by the "leader-follower" system observed in crowds. In this system, individuals naturally assume the role of guiding collective movement: static or dynamic leaders influence the direction and pace of evacuation, while followers, who make up the bulk of the crowd, are influenced by the individuals around them. This phenomenon is reflected in the algorithm's exploration phase, which categorizes agents into calm, follower, and panic-prone groups. The unique behaviors exhibited by each type of agent collectively drive the search process toward the optimal solution.

[0053] Calm individuals: These are the calm individuals within a group who assess situations with clarity and make rational decisions. These actors systematically search the problem space, much like the calm individuals in an evacuation seeking efficient paths, guiding others through their steady influence.

[0054] Crowd-following: This manifests as a herd mentality, where individuals blindly follow the group's actions without thinking. This is mapped into the behavior of herd-following agents in our algorithm. This behavior strengthens the algorithm's development phase by aggregating agents toward promising areas of the search space—much like how individuals in a crowd might follow others toward a perceived exit or safe area.

[0055] Panic Crowds: Panic-stricken individuals (whose unpredictable and erratic movements can hinder escape route discovery or unexpectedly facilitate new routes) inspire the diversification mechanism in our algorithm. This property is replicated in panic agent behavior, introducing randomness to prevent the algorithm from prematurely converging to a local optimum—just as panic in a crowd can lead to the discovery of unconventional exits.

[0056] Through the ESC algorithm, this paper leverages the inherent intelligence of crowd behavior in emergency situations, transforming the interplay of calmness, conformity, and panic into a computational model. This approach not only provides valuable insights into algorithm design but also highlights the potential of natural and human phenomena as a source of inspiration for developing advanced problem-solving strategies.

[0057] (3) Initialize the algorithm and population and establish the panic index and iterative process.

[0058] The ESC algorithm is designed to simulate the behavior of a crowd during an emergency evacuation, where individuals must navigate towards an exit in a dynamic and uncertain environment. ESC introduces the concept of an elite pool, representing potential exits identified by the crowd. This enhances the algorithm's ability to explore the solution space, avoid local optima, and consider multiple directions simultaneously.

[0059] The ESC algorithm is first initialized N A population of individuals, each of which consists of D dimensional vector Description. j The value of dimension is given by: ; (17) in: and Respectively represent j The lower and upper bounds of the dimension ensure that the initial position of each individual is randomly distributed in the feasible space; the random variable It is uniformly distributed between 0 and 1, reflecting the randomness of the initial decision during the evacuation process.

[0060] After the population is initialized, the fitness function is used f Evaluate the fitness of each individual = f( )The population is then sorted in ascending order of fitness, and the individuals with the highest fitness are stored in the elite pool. E In the formula below, the parameter represents the number of potential safe exits identified by the crowd.

[0061] ; (18) These elite individuals represent the best potential solutions identified by the population and serve as reference points for subsequent iterations.

[0062] Establish a panic index and iterative process as follows: The ESC algorithm simulates an iterative process to reflect the changes in crowd behavior as the evacuation progresses. t The panic index P(t) is calculated at the beginning of the generation: ;(19) The panic index P(t) reflects the overall panic level of the crowd, and the higher the value, the more chaotic the behavior. t Gradually increase from 0 to the maximum number of iterations T , the index gradually decreases, simulating the crowd’s adaptation to the evacuation environment.

[0063] (4) Explore the optimal solution (when t≤T / 2).

[0064] In the number of iterations t ≤ T / 2( T is the maximum number of iterations of the algorithm, t The exploration phase (where is the current iteration number) is divided into three groups according to individual fitness: calm, herd, and panic. Specifically, the population is sorted in ascending order of fitness and the proportion c =0.15, h =0.35 and p =0.5 is stratified into calm, follow-the-herd, and panic groups, which reflects the diverse reactions of the crowd during the evacuation, with some people remaining calm, some following the herd, and some falling into panic.

[0065] aUpdate calm group.

[0066] Individuals in the calm group act rationally and move towards the center of group decision-making move: ; (20) in: Represents the updated individual i In the j Dimensional location; Represents an individual i In the j Dimensional position. Binary variable Determined by the Bernoulli distribution, partial updates are allowed to simulate the non-update of some dimensions due to crowding. Specifically, Generates 0 or 1 with equal probability. is the adaptive Levian weight, which uses the Levian distribution to simulate the step size of the exploration phase, and is defined in formula (27). For the calm group j The center of the dimension is the average value of all calm individuals in the dimension. It is defined as follows: ;(twenty one) in: v c,j express j The movement vector of the dimensional calm individual. Represents an individual i In the j Dimensional location. is a randomly generated position within the calm group boundary, and Respectively indicate the The minimum and maximum values of all calm individuals in the dimension, r i,j Represents an individual i In the j dimensional motion constant. , is the fine-tuning perturbation of the individual motion, is a random variable that follows the standard normal distribution N(0,1).

[0067] b. Update the herd group.

[0068] The conforming individuals are influenced by the calm and panic groups' behaviors, and their position updates are based on the combined influence of the two groups: ;(twenty two) in: m 2 is through m 1Binary variables generated by the same mechanism. A randomly selected individual from the panic group, representing the potential direction of panic-driven movement. Vector It is defined as formula (23): ;(twenty three) in, v h,j express j Dimensional movement vectors of conforming individuals; is a randomly generated position within the population group boundary, and Respectively represent the minimum and maximum values of all conforming individuals in this dimension, ri,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with (0,1).

[0069] c. Update the panic group.

[0070] Panic-driven individuals explore the solution space more randomly, influenced by potential exits (from the elite pool) and random directions from other individuals: ;(twenty four) in: An individual is randomly selected from the elite pool to represent a potential exit point to which the panicking individual might go. is an individual randomly selected from the population, introducing randomness. The vector It is defined as formula (25): ; (25) in, v p,j express j Movement vectors of dimensionally panicked individuals; is a randomly generated location within the panic group boundaries; and Respectively represent the minimum and maximum values of all panic individuals in this dimension; r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with (0,1).

[0071] (5) Entering the development stage (when t>T / 2): As the number of iterations exceeds T / 2, the algorithm enters the development phase, where all individuals are considered calm. The focus shifts to fine-tuning their positions based on the currently identified optimal solution. During this phase, individuals adjust their positions by moving closer to members of the elite pool (representing possible safe exits and the best solution identified in the previous iteration) and randomly selected individuals from the population, simulating a population gradually converging toward the identified optimal exit. The position update formula for this phase is: ; (26) in, Represents an individual i In the j Dimensional location. is a position of an elite pool member, representing both a possible safe exit and one of the best solutions identified so far. is the position of a randomly selected individual. Individuals adjust their positions by approaching elite pool members and randomly selected individuals, simulating a crowd gradually converging toward the identified optimal exit.

[0072] (6) Adaptive Levian weights and behavioral simulation.

[0073] The individual step size is controlled by the adaptive column weight, simulating the different degrees of exploration and development at different stages of the algorithm. j Levy weights The calculation is as follows: ; (27) in: Γ is the gamma function; β is a dynamic adjustment parameter that changes with the algorithm process and is defined as follows: ; (28) in, is the initial value of β, which is set to 1.5. This adjustment allows the algorithm to make a larger range of exploratory moves in the initial stage (when β is small) and gradually transition to more refined exploitative moves (as β increases), reflecting the natural progression from panic-driven exploration to calm, rational decision-making.

[0074] (7) Determine the fitness function f And update the elite pool.

[0075] Point-to-point path planning for deep sea mining vehicles (DSMVs) in deep sea mining operations is a multi-objective nonlinear optimization problem that requires comprehensive consideration of factors such as distance cost, turning cost, and obstacle constraints. f The design is as follows: ;(29) in: It is the time cost; is the cost of turning; is the obstacle constraint cost.

[0076] Time cost J L , mainly considering the path length of DSMV from the starting point to the end point. The specific calculation formula is as follows: ; (30) in:( ) indicates the previous coordinate point; ( ) represents the current coordinate point; v mrepresents the driving speed of DSMV, which is set to 0.83 m / s; k ( i ) means passing through a slope ( The ratio of the travel time of the grid to the travel time of the flat grid can be obtained from the slope model formula (10) through climbing dynamics analysis.

[0077] Turning costs J A , mainly considering the sum of the steering angles of each path point of the DSMV: ; (31) in, Indicates the i -1 coordinate point to i The steering angle of the coordinate point.

[0078] Obstacle constraint cost J O Once a pathpoint is found within an obstacle area, the cost is set to infinity to reject the path: ; (32) During the DSMV point-to-point static path planning process, the above costs should be optimized as much as possible.

[0079] In each iteration, the fitness of each updated individual is recalculated. A greedy selection process is used to retain the better solutions: ; (33) The update rule of individual i can be described as: ; (34) If the new fitness is better (i.e., smaller for the minimization problem), then the individual position update is Otherwise, keep the original position After each iteration, the elite pool E The pool of elites plays a key role in guiding the population towards the best possible exit.

[0080] To verify the effectiveness of this motion framework, the present invention conducted coverage path planning experiments in simple and complex deep-sea mining environments. The experimental results show that the proposed method based on dynamic analysis can help deep-sea mining vehicles avoid obstacles and effectively cover all areas in complex seabed environments. When encountering a dead zone during the coverage operation, it can effectively escape the dead zone along the path that consumes the least time, and the coverage operation rate can reach 100%. Compared with the traditional method, the number of turns is reduced by 17.53%, the repetition rate is reduced by 11.6%, and the walking time is reduced by 6.83%. Supported by the above relevant simulation experiments, it can be demonstrated that the proposed mining vehicle coverage path planning motion framework based on the improved biologically activated network takes into account safety, mining efficiency, and energy saving, and well meets the operational needs of actual deep-sea mining.

Claims

1. A full-coverage path planning method for deep-sea mining vehicles based on an improved biologically activated neural network, characterized by: The following steps are involved: (1) Analyze the climbing dynamics of deep-sea mining vehicles and determine the travel time in each deep-sea mining area; (2) Establish a deep-sea mining operation environment model based on the determined transit time for each area; (3) Establishing a grid-inspired function for deep-sea mining coverage operations to improve the biologically activated neural network and enhance the coverage efficiency of deep-sea mining vehicles; (4) During deep-sea mining operations, the escape path planning algorithm is improved to meet the needs of escaping from the dead zone, the target search mechanism and fitness function are optimized, and point-to-point intelligent path planning is performed for the mining vehicle.

2. The method for full coverage path planning of deep-sea mining vehicles based on improved biologically activated neural networks according to claim 1 is characterized in that: The process of analyzing the climbing dynamics of the deep-sea mining vehicle in step (1) and establishing the travel time of the deep-sea mining area is as follows: First, define the distribution law of ground pressure: ； in: is the ground pressure; G is the weight of the mining vehicle; B and L are the width and length of the deep-sea mining vehicle tracks respectively; α is the slope angle; In the actual climbing process, the center of gravity of the crawler deep-sea mining vehicle moves backward when climbing. The maximum shear stress and participating shear stress of the soft bottom are analyzed and defined as follows: ； in: τ(j m ) is the actual shear stress; is the maximum shear stress; λ is a scaling factor less than 1; ω is a constant; j m =i m x , i m is the track slip ratio, x is the track ground contact length; k m is the substrate shear modulus; Theoretical driving force behind the neglect of grouser effect in deep-sea mining vehicles The analysis is performed and defined as follows: ； in: is the theoretical driving force; B and L are the width and length of the deep-sea mining vehicle tracks respectively; for substrate cohesion; is the ground pressure; is the internal friction angle of the substrate; λ is a scaling factor less than 1; ω is a constant; i m is the track slip ratio; x is the track ground contact length; is the substrate shear modulus; Calculate the difference between actual driving force and theoretical driving force , defined as follows: ； in, n , m is a non-zero constant; The ultimate driving force The definition is as follows: ； During the climbing process, the resistance encountered by the deep-sea mining vehicle is divided into ground squeezing resistance and water gravity; ground squeezing resistance The definition is as follows: ； in: is the cohesive deformation modulus; is the friction deformation modulus; B and L are the width and length of the deep-sea mining vehicle track, respectively is the longitudinal eccentricity; Water gravity component The definition is as follows: ； Where: G is the gravity of the mining vehicle; α is the slope angle; The deep-sea mining vehicle will be affected by the hydrodynamics during its travel. F h It is composed of water resistance and inertia force and is defined as follows: ； in: is the water resistance coefficient; is the frontal area of the mining vehicle; and are the mining vehicle's travel speed and the ocean current speed; is the added mass coefficient; is the volume of the mining vehicle; The final actual driving force of the above deep-sea mining vehicle , ground compression resistance , water gravity component and hydrodynamics F h Substitute the formula into Newton's second law, the formula is as follows: ； Where: G is the gravity of the deep-sea mining vehicle, g is the gravitational acceleration coefficient; , dv and dt Represent the differential forms of velocity and time respectively; By solving a first-order linear separable variable differential equation , we can get the running speed of the deep-sea mining vehicle under different slopes, and then use the following slope model formula to calculate the over-slope time of different slopes: ； in: α is the slope angle; h is the slope height; v m is the vehicle's travel speed; After calculating the slope time, the slope less than 5° is defined as a flat area, the slope between 5° and 10° is defined as a steep slope area, and the slope greater than 10° is defined as an infeasible area.

3. The method for full coverage path planning of deep-sea mining vehicles based on improved biologically activated neural networks according to claim 1 is characterized in that: The process of establishing the deep-sea mining operation environment model in step (2) is as follows: The seabed topography model established by the numerical model is used to simulate the real deep-sea mining operation environment. The natural topography in the seabed environment is described by the following exponential function: ； in:( x d , y d ) represents the actual coordinates of the seabed; Z ( x d , y d ) represents the height of the seabed topography; ( x di , y di ) represents the center coordinates of the hillock; h i It is a topographic parameter that determines the height of the sea hill; s is the number of sea mounds; x si and y si They are i A hillock in x Axis and y Attenuation rate in the axial direction; The three-dimensional path is projected onto a two-dimensional plane, and the terrain of the mining area is divided into obstacle areas and barrier-free areas. The obstacle areas include high obstacle areas and restricted access areas, and the grids with slopes greater than 10° are set as high obstacle areas. The barrier-free areas are divided into flat areas and steep areas. Flat areas refer to barrier-free areas with slope angles between 0° and 5°, and steep areas refer to barrier-free areas with slope angles between 5° and 10°.

4. The method for full coverage path planning of deep-sea mining vehicles based on improved biologically activated neural networks according to claim 1 is characterized in that: The process of establishing a grid heuristic function for deep-sea mining coverage operation requirements in step (3) is as follows: A. Establishing coverage status grid activity indicators : Set the grid status of the traveled grid to 0, the grid status of the untraveled grid to 500, and the grid status of the grid with obstacles to -500, and define the coverage status grid activity indicator as follows: ； B. Establishing distance grid activity indicators : Set the distance grid activity index for: ； in: are the coordinates of the surrounding grid points; is the coordinate of the current point; C. Establishing Steering Grid Activity Indicators : Defining steering grid activity indicators for: ； in: Represents the value from the previous coordinate point ( ) points to the current coordinate point ( ) direction vector; Represents the value from the previous coordinate point ( ) points to the grid coordinate point ( ) direction vector; D. Synthesize the final grid heuristic function: Comprehensively consider the above three grid activity indicators , the final grid heuristic function is expressed as the following formula: ； in: Corresponding to three grid activity indicators The weight of .

5. The method for full coverage path planning of deep-sea mining vehicles based on improved biologically activated neural networks according to claim 1 is characterized in that: The process of improving the escape path planning algorithm in step (4) is as follows; A. Search for the escape target grid point: Assume that the mining vehicle is located at the dead zone position ( ), the position coordinates of the target grid point are ( ), then the distance from the dead zone position point to the target grid point The definitions are as follows: ； According to the above formula, the distance between all suitable target grids and the current dead zone grid is determined, and the grid point with a relatively small distance is selected as the final target grid point; B. Initialize the algorithm and population and establish the panic index and iterative process: a. Initialization algorithm and population: First initialize N A population of individuals, each of which consists of D dimensional vector Description, individual i In the j The value of dimension is given by: ； in: Represents an individual i In the j The location of the dimension, and Respectively represent j The lower and upper bounds of the dimension ensure that the initial position of each individual is randomly distributed in the feasible space; the random variable It is uniformly distributed between 0 and 1, reflecting the randomness of the initial decision during the evacuation process; After the population is initialized, the fitness function is used f Evaluate the fitness of each individual = f ( ), then sort the population in ascending order of fitness and store the individuals with the highest fitness in the elite pool E In the formula below, the parameter represents the number of potential safe exits identified by the crowd; ； These elite individuals represent the best potential solutions identified by the population and serve as reference points for subsequent iterations; b. Establish panic index and iterative process: The escape path planning algorithm simulates the iterative process to reflect the behavior changes of the crowd as the evacuation progresses. t The panic index P(t) is calculated at the beginning of the generation: ； The panic index P(t) reflects the overall panic level of the crowd. The higher the value, the more chaotic the behavior. t Gradually increase from 0 to the maximum number of iterations T , the index gradually decreases, simulating the crowd’s adaptation to the evacuation environment; C. When the number of iterations t ≤ T / 2 starts to explore the optimal solution: In the number of iterations t ≤ T / 2 exploration phase, in which T is the maximum number of iterations of the algorithm, t The number of the current iteration is divided into three groups according to the individual fitness: calm, conformist and panic; Update calm group: Individuals in the calm group act rationally and move towards the center of group decision-making move: ； in: Represents the updated individual i In the j Dimensional location; Represents an individual i In the j Dimensional location; binary variable Determined by the Bernoulli distribution, Generates 0 or 1 with equal probability; is the adaptive Levian weight, which uses the Levian distribution to simulate the step size of the exploration phase, defined as formula; For the calm group j The center of the dimension, that is, the average value of all calm individuals in the dimension; vector It is defined as follows: ； in: v c,j express j Dimensional calm individual movement vector; Represents an individual i In the j Dimensional location; is a randomly generated position within the calm group boundary, and Respectively indicate the The minimum and maximum values of all calm individuals in the dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, is a random variable that follows the standard normal distribution N(0,1); Update the herd group: The conforming individuals are influenced by the calm and panic groups' behaviors, and their position updates are based on the combined influence of the two groups: ； in: m 2 is through m 1 binary variables generated by the same mechanism; A randomly selected individual from the panic group represents the potential movement direction driven by panic; vector It is defined as follows: ； in, v h,j express j Dimensional movement vectors of conforming individuals; is a randomly generated position within the population group boundary, and Respectively represent the minimum and maximum values of all conforming individuals in this dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with a value of (0,1); Update panic group: Panic-driven agents explore the solution space more randomly, influenced by potential exits and random directions from other agents: ； in: An individual is randomly selected from the elite pool to represent a potential exit to which the panicking individual might go; is an individual randomly selected from the population, introducing randomness; the vector It is defined as follows: ； in, v p,j express j Movement vectors of dimensionally panicked individuals; is a randomly generated position within the panic group boundary, and Respectively represent the minimum and maximum values of all panic individuals in this dimension, r i,j Represents an individual i In the j Dimensional movement constant; , is the fine-tuning perturbation of the individual motion, To obey the standard normal distribution N A random variable with a value of (0,1); D. When the number of iterations t > T / 2, it enters the development phase: As the number of iterations exceeds T / 2, the algorithm enters the development phase, and all individuals are considered to be in a calm state; the focus turns to fine-tuning the position based on the currently identified optimal solution; the position update formula in this phase is: ； in, Represents an individual i In the j Dimensional location; is the position of a member of the elite pool, representing both a possible safe exit and one of the best solutions identified so far; is the position of a randomly selected individual; individuals adjust their positions by approaching elite pool members and randomly selected individuals, simulating the crowd gradually converging towards the identified optimal exit; E. Adaptive Levian Weights and Behavior Simulation: The individual step size is controlled by the adaptive Lev weight, simulating the different degrees of exploration and development at different stages of the algorithm; each dimension j Levy weights The calculation is as follows: ； in: Γ is the gamma function; β is a dynamic adjustment parameter that changes with the algorithm process and is defined as follows: ； in, is the initial value of β, which is 1.5; F. Determine the fitness function f And update the elite pool: Fitness function f The design is as follows: ； in: It is the time cost; is the cost of turning; is the obstacle constraint cost; Time cost J L , is the path length of the deep-sea mining vehicle from the starting point to the end point, and the calculation formula is as follows: ； in:( ) indicates the previous coordinate point; ( ) represents the current coordinate point; v m Indicates the speed of the deep-sea mining vehicle, set to 0.83m / s; k ( i ) means passing through a slope ( The ratio of the travel time of the grid to the travel time of the flat grid is obtained from the slope model formula through climbing dynamics analysis; Turning costs J A , is the sum of the steering angles of each path point of the deep-sea mining vehicle: ； in, Indicates the i -1 coordinate point to i Steering angle of coordinate points; Obstacle constraint cost J O Once a pathpoint is found within an obstacle area, the cost is set to infinity to reject the path: ； In each iteration, the fitness of each updated individual is recalculated; a greedy selection process is used to retain the better solutions: ； individual i The update rule is described as: ； If the new fitness Better, then the individual position is updated to Otherwise, keep the original position ; After each iteration, the elite pool E Will also be updated.

Images