Dynamic adaptive amphibious vehicle path and energy consumption optimization method

Abstract

The invention discloses a dynamic adaptive amphibious vehicle path and energy consumption optimization method, and relates to the field of path and energy consumption optimization. The method comprises the following steps: acquiring information data; the information data comprises environment data and related data of the amphibious vehicle; the environment data comprises wind speed, wind direction, ocean current speed, ocean current direction, sea wave height and sea wave frequency; the related data of the amphibious vehicle comprise the position, the speed and the acceleration; the information data contains a timestamp; inputting the information data into a path and energy consumption optimization model to obtain optimized planning path information; the optimized planning path information is a planning path with lowest energy consumption determined based on a heuristic function; the path and energy consumption optimization model is obtained by adopting a reinforcement learning algorithm and an A * algorithm and updating and optimizing based on an amphibious vehicle water surface energy consumption model and a heuristic function. According to the invention, path and energy consumption optimization can be realized, and the cruising ability and task completion efficiency of the amphibious vehicle are improved.

Description

Technical Field

[0001] The present application relates to the field of path and energy consumption optimization, and in particular to a method for dynamically adaptively optimizing the path and energy consumption of an amphibious vehicle. Background Art

[0002] In modern marine environment applications, amphibious vehicles are widely used in environmental monitoring, resource exploration, military missions and rescue. With the increase in application requirements, the ability of amphibious vehicles to perform tasks in complex and dynamic marine environments has become crucial. However, the complex marine environment makes them face many challenges in the path planning process.

[0003] Traditional A * Path planning algorithms such as Dijkstra are mainly applicable to static environments and cannot cope with complex environments with real-time dynamic changes. Although methods such as Model Predictive Control (MPC) can handle dynamic environments, they rely on accurate environmental and vehicle dynamics models. In practical applications, the actual effect is often limited due to the inaccuracy of the model.

[0004] In addition, many existing path planning algorithms only focus on the length or time of the path, and do not fully consider the optimization of energy consumption. For energy-constrained systems such as amphibious vehicles, energy consumption is a key indicator. If only the length or time of the path is considered, while ignoring energy consumption, the vehicle may run out of energy during the mission execution and fail to complete the scheduled mission. Summary of the invention

[0005] The purpose of this application is to provide a dynamic adaptive amphibious vehicle path and energy consumption optimization method, which can achieve path and energy consumption optimization and improve the endurance and mission completion efficiency of the amphibious vehicle.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] The present application provides a dynamic adaptive amphibious vehicle path and energy consumption optimization method, including:

[0008] Acquire information data; the information data includes: environmental data and data related to the amphibious vehicle itself; the environmental data includes: wind speed, wind direction, ocean current speed, ocean current direction, wave height and wave frequency; the data related to the amphibious vehicle itself includes: position, speed and acceleration; the information data includes a timestamp;

[0009] The information data is input into the path and energy consumption optimization model to obtain the optimized planning path information; the optimized planning path information is the planning path with the lowest energy consumption determined based on the heuristic function; the path and energy consumption optimization model is a reinforcement learning algorithm and A *The algorithm is obtained after updating and optimizing the amphibious vehicle surface energy consumption model and the heuristic function; the amphibious vehicle surface energy consumption model is a mathematical model determined according to the dynamic characteristics of the amphibious vehicle and environmental factors.

[0010] Optionally, the amphibious vehicle surface energy consumption model includes a propulsion energy consumption model and an environmental resistance energy consumption model; the expression of the amphibious vehicle surface energy consumption model is:

[0011] E=E p +E r ;

[0012] Where E is the energy consumption of the amphibious vehicle on the water surface; E p To promote energy consumption; E r Energy consumption for environmental resistance.

[0013] Optionally, the expression of the propulsion energy consumption model is:

[0014]

[0015] Among them, F p is the propulsion force; d is the driving distance, and η is the efficiency of the propulsion system.

[0016] Optionally, the expression of the environmental resistance energy consumption model is:

[0017] E r =E r_wind +E r_current +E r_wave ;

[0018] E r_wind =w c ·k c ·(v w ·cos(θ-d w ));

[0019] E r_current =w f ·k f ·(v c cos(θ-d c ));

[0020] E r_wave =w w ·k w ·(h w ·f w );

[0021] Among them, E r_wind E is wind resistance energy consumption; r_current E is the energy consumption of ocean current resistance; r_wave is the energy consumption of wave resistance; w cis the weight coefficient of wind resistance energy consumption; k c is the wind resistance coefficient; v w is the wind speed; θ is the heading angle; d w is the wind direction; cos(θ-d w ) is the angle between the heading angle of the amphibious vehicle and the wind direction; w f is the weight coefficient of energy consumption of ocean current resistance; k f is the ocean current resistance coefficient; v c is the ocean current speed; d c is the direction of ocean current; cos(θ-d c ) is the angle between the heading angle of the amphibious vehicle and the direction of the ocean current; h w is the wave height; f w is the wave frequency; k w is the wave resistance coefficient; w w is the weight coefficient of wave resistance energy consumption.

[0022] Optionally, the heuristic function is expressed as:

[0023]

[0024] Among them, h”(n) is the heuristic function; w d is the weight of the path length; (x n ,y n ) is the node coordinate of the current position; (x g ,y g ) is the target location node coordinate; E r E is the environmental resistance energy consumption; p To promote energy consumption.

[0025] Optionally, the method for determining the path and energy consumption optimization model specifically includes:

[0026] Obtain training information data;

[0027] Constructing a deep Q network; the deep Q network includes an experience replay buffer;

[0028] Initialize the experience replay buffer and weight parameters of the deep Q network;

[0029] Determine the state space and action space based on the training information data; the expression of the state space s is: s = [v w ,d w ,v c ,d c ,h w ,f w ,x,y,v,θ]; the expression of action space a is: a=[Δw d ,Δw c,Δw f ,Δw w ,Δθ,Δv]; v w is the wind speed; d w is the wind direction; v c is the ocean current speed; d c is the direction of ocean current; h w is the wave height; f w is the wave frequency; θ is the heading angle; x is the horizontal coordinate of the position; y is the vertical coordinate of the position; v is the speed; Δw c Δw is the adjustment of wind resistance energy consumption weight coefficient; d Δw is the path length weight adjustment; f Δw is the adjustment of the energy consumption weight coefficient of ocean current resistance; w is the adjustment of the wave resistance energy consumption weight coefficient; Δθ is the heading angle adjustment; Δv is the speed adjustment.

[0030] For any time step, the corresponding action space is selected based on the current state space, and the next state space is determined according to the corresponding reward function; the time step is determined after time point alignment processing is performed according to the timestamp; the reward function is determined according to the state space and the action space;

[0031] Randomly extract sample data from the stored experience playback buffer, and then * The algorithm performs path search, and based on the heuristic function, trains and updates the weight parameters of the deep Q network with the goal of network convergence to obtain a trained deep Q network; the stored experience replay buffer stores the state space, action space, reward function and next state space corresponding to each time step;

[0032] The trained deep Q network is determined as the path and energy consumption optimization model.

[0033] Optionally, the reward function is expressed as:

[0034]

[0035] Among them, R(s,a) is the reward function; E is the energy consumption of the amphibious vehicle on the water surface; α, β, and γ are all weight parameters; a i is the action at the i-th time step; δ j is the result of the jth training update adjustment; n is the total number of time steps; m is the total number of training updates; a i+1 is the action at the i+1th time step.

[0036] Optionally, obtaining information data specifically includes:

[0037] Obtain initial information data;

[0038] Align the initial information data according to the timestamp;

[0039] A Kalman filter is used to remove noise from the aligned initial information data to obtain processed data;

[0040] The processed data are fused by using a weighted average method to obtain the information data.

[0041] According to the specific embodiments provided in this application, this application has the following technical effects:

[0042] This application provides a dynamic adaptive amphibious vehicle path and energy consumption optimization method by combining reinforcement learning algorithm and A * The algorithm optimizes the heuristic function parameters and improves the efficiency and accuracy of path search. The heuristic function comprehensively considers factors such as path length, propulsion energy consumption, and environmental resistance energy consumption to ensure path planning with the lowest energy consumption. In addition, the reinforcement learning algorithm not only optimizes the heuristic function parameters, but also dynamically adjusts according to real-time environmental data and the status of the amphibious vehicle during path execution to ensure that path planning is always optimal. This application significantly improves the autonomous operation capability and task completion efficiency of amphibious vehicles. Through real-time data processing and dynamic path adjustment, amphibious vehicles can complete more tasks under limited energy conditions. As a result, this application can achieve path and energy consumption optimization and improve the endurance and task completion efficiency of amphibious vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 A flow chart of the dynamic adaptive amphibious vehicle path and energy consumption optimization method for this application;

[0045] Figure 2 Based on the improvement of A * The overall flow chart of the algorithm's dynamic adaptive amphibious vehicle routing and energy consumption optimization method;

[0046] Figure 3 Schematic diagram for sensor data collection and environment modeling;

[0047] Figure 4 Flowchart for reinforcement learning, experience replay, and model update based on deep Q network. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] The Chinese patent document CN116560363A known to the applicant introduces a robot low-energy path planning method based on an improved ant colony algorithm in a multi-attribute grid environment. This method establishes a multi-attribute grid map of robot motion, takes into account multiple energy consumption criteria such as path length, number of turns, slope, and ground friction coefficient, and improves the heuristic function of the traditional ant colony algorithm while optimizing the pheromone model to improve the energy efficiency of path planning. However, this method is mainly applicable to pre-known grid environments, and in practical applications, it may be difficult to meet the needs when facing complex marine environments.

[0050] The Chinese patent document CN116243701A discloses an optimal energy consumption path planning method for an underwater robot based on the optimal propulsion speed. The candidate path in the form of a B-spline curve is generated by a quantum particle swarm algorithm, and the optimal propulsion speed and energy consumption of each path segment are calculated using a vector analysis method. The energy consumption of each path segment is accumulated to obtain the total energy consumption of the candidate path. This method improves the computational efficiency of the optimal energy consumption path planning. However, this method is mainly applicable to static or quasi-static environments, and has weak real-time response capabilities to dynamic changes in the environment. In addition, although the particle swarm algorithm has advantages in searching for the global optimal solution, the computational complexity is high in high-dimensional complex environments, and there are certain deficiencies in real-time performance.

[0051] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0052] In an exemplary embodiment, a dynamic adaptive amphibious vehicle path and energy consumption optimization method is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to a server as an example for explanation. Figure 1 As shown, the process includes the following steps 100 to 200.

[0053] Specifically, a dynamic adaptive amphibious vehicle path and energy consumption optimization method includes:

[0054] Step 100: Acquire information data. The information data includes: environmental data and data related to the amphibious vehicle itself; environmental data includes: wind speed, wind direction, ocean current speed, ocean current direction, wave height and wave frequency; data related to the amphibious vehicle itself includes: position, speed and acceleration; information data contains a timestamp.

[0055] Step 200: Input the information data into the path and energy consumption optimization model to obtain the optimized planning path information. The optimized planning path information is the planning path with the lowest energy consumption determined based on the heuristic function; the path and energy consumption optimization model is based on the reinforcement learning algorithm and A * The algorithm is obtained after updating and optimizing the amphibious vehicle surface energy consumption model and the heuristic function; the amphibious vehicle surface energy consumption model is a mathematical model determined according to the dynamic characteristics of the amphibious vehicle and environmental factors.

[0056] The energy consumption model of the amphibious vehicle on the water surface includes the propulsion energy consumption model and the environmental resistance energy consumption model; the expression of the energy consumption model of the amphibious vehicle on the water surface is:

[0057] E=E p +E r ;

[0058] Where E is the energy consumption of the amphibious vehicle on the water surface; E p To promote energy consumption; E r Energy consumption for environmental resistance.

[0059] The expression of propulsion energy consumption model is:

[0060]

[0061] Among them, F p is the propulsion force; d is the driving distance, and η is the efficiency of the propulsion system.

[0062] The expression of environmental resistance energy consumption model is:

[0063] E r =E r_wind +E r_current +E r_wave ;

[0064] E r_wind =w c ·k c ·(v w ·cos(θ-d w ));

[0065] E r_current =w f ·k f ·(v c ·cos(θ-d c ));

[0066] E r_wave =w w ·k w ·(h w ·f w );

[0067] Among them, E r_wind E is wind resistance energy consumption; r_current E is the energy consumption of ocean current resistance; r_wave is the energy consumption of wave resistance; w c is the weight coefficient of wind resistance energy consumption; k c is the wind resistance coefficient; v w is the wind speed; θ is the heading angle; d w is the wind direction; cos(θ-d w ) is the angle between the heading angle of the amphibious vehicle and the wind direction; w f is the weight coefficient of ocean current resistance energy consumption; k f is the ocean current resistance coefficient; v c is the ocean current speed; d c is the direction of ocean current; cos(θ-d c ) is the angle between the heading angle of the amphibious vehicle and the direction of the ocean current; h w is the wave height; f w is the wave frequency; k w is the wave resistance coefficient; w w is the weight coefficient of wave resistance energy consumption.

[0068] In one embodiment, the expression of the heuristic function is:

[0069]

[0070] Among them, h”(n) is the heuristic function; w d is the weight of the path length; (x n ,y n ) is the node coordinate of the current position; (x g ,y g ) is the target location node coordinate; E r E is the environmental resistance energy consumption; p To promote energy consumption.

[0071] As an optional implementation method, the method for determining the path and energy consumption optimization model specifically includes:

[0072] Obtain training information data; build a deep Q network; the deep Q network includes an experience replay buffer.

[0073] Initialize the experience replay buffer and weight parameters of the deep Q network.

[0074] Determine the state space and action space based on the training information data; the expression of the state space s is: s = [v w ,d w ,v c ,d c ,h w ,f w ,x,y,v,θ]; the expression of action space a is: a=[Δw d ,Δw c ,Δw f ,Δw w ,Δθ,Δv]; v w is the wind speed; d w is the wind direction; v c is the ocean current speed; d c is the direction of ocean current; h w is the wave height; f w is the wave frequency; θ is the heading angle; x is the horizontal coordinate of the position; y is the vertical coordinate of the position; v is the speed; Δw c Δw is the adjustment of wind resistance energy consumption weight coefficient; d Δw is the path length weight adjustment; f Δw is the adjustment of the energy consumption weight coefficient of ocean current resistance; w is the adjustment of the wave resistance energy consumption weight coefficient; Δθ is the heading angle adjustment; Δv is the speed adjustment.

[0075] For any time step, the corresponding action space is selected based on the current state space, and the next state space is determined according to the corresponding reward function; the time step is determined after the time point is aligned according to the timestamp; the reward function is determined based on the state space and action space.

[0076] Randomly extract sample data from the stored experience playback buffer, and then * The algorithm performs path search and trains and updates the weight parameters of the deep Q network based on the heuristic function with the goal of network convergence to obtain the trained deep Q network; the stored experience replay buffer stores the state space, action space, reward function and next state space corresponding to each time step.

[0077] The trained deep Q network is determined as the path and energy consumption optimization model.

[0078] Specifically, the reward function is expressed as:

[0079]

[0080] Among them, R(s,a) is the reward function; E is the energy consumption of the amphibious vehicle on the water surface; α, β, and γ are all weight parameters; a iis the action at the i-th time step; δ j is the result of the jth training update adjustment; n is the total number of time steps; m is the total number of training updates; a i+1 is the action at the i+1th time step.

[0081] In one embodiment, obtaining information data specifically includes: obtaining initial information data; aligning the initial information data according to timestamps; using a Kalman filter to remove noise from the aligned initial information data to obtain processed data; and using a weighted average method to fuse the processed data to obtain information data.

[0082] This application aims at the path planning and energy consumption optimization problems of amphibious vehicles in dynamic marine environments. By modifying the heuristic function, the improved A * The algorithm solves the path decision problem. At the same time, by collecting environmental data such as wind speed, wind direction, ocean current speed, ocean current direction, wave height and frequency in real time, it dynamically adjusts the heuristic function parameters of path planning to optimize path planning and energy consumption management. By combining reinforcement learning and real-time data processing technology, it dynamically adapts to environmental changes, comprehensively considers path planning and energy consumption optimization, and improves the endurance and mission completion efficiency of amphibious vehicles.

[0083] In practical applications, the overall flow chart corresponding to the method mentioned in this application is shown in Figure 2 The specific steps include:

[0084] Step 1: Real-time environmental data collection and processing. Sensors are used to collect real-time environmental data and information related to the amphibious vehicle itself, including wind speed v w , wind direction w , ocean current speed v c , ocean current direction d c 、wave height h w , wave frequency f w The sensor collects data once a second to ensure the real-time and accuracy of the data. Each collected data is timestamped for time synchronization during subsequent processing.

[0085] After collecting the required data, all data need to be time-stamped and aligned to ensure that each type of data is processed and fused at the same time point. Use the Kalman filter to process the raw data, remove noise, and improve data accuracy. After data preprocessing is completed, data fusion is performed to improve the overall accuracy of the data. The same type of data from multiple sensors can be fused by weighted averaging to improve data accuracy.

[0086] Step 2: Establish an amphibious vehicle surface energy consumption model. In order to optimize path planning and energy consumption management, it is necessary to reasonably plan the path with the lowest energy consumption based on the dynamic characteristics and environmental factors of the amphibious vehicle. The energy consumption of an amphibious vehicle when performing a mission includes propulsion energy consumption and environmental resistance energy consumption.

[0087] The propulsion energy consumption mainly depends on the propulsion force, driving distance and efficiency of the propulsion system of the amphibious vehicle. Specifically, it can be expressed as:

[0088]

[0089] Among them, F p is the propulsion force, d is the travel distance, and η is the efficiency of the propulsion system.

[0090] In a dynamic marine environment, amphibious vehicles need to overcome environmental resistance to maintain their driving state. Environmental resistance mainly includes wind resistance, ocean current resistance and wave resistance, which are the main external resistance sources when amphibious vehicles are driving in a dynamic marine environment.

[0091] The wind resistance energy consumption model represents the energy consumed by the air resistance encountered by the amphibious vehicle during driving, which can be specifically expressed as:

[0092] E r_wind =w c ·k c ·(v w ·cos(θ-d w )).

[0093] In the formula, v w Indicates wind speed, d w Indicates wind direction, cos(θ-d w ) is the angle between the heading angle θ of the amphibious vehicle and the wind direction, which is used to calculate the actual wind speed felt by the amphibious vehicle in the forward direction. c is the wind resistance coefficient, w c It is the weight coefficient of wind resistance energy consumption, which is used to adjust the influence proportion of wind resistance energy consumption in total energy consumption.

[0094] Ocean current resistance energy consumption refers to the energy consumed by the ocean current resistance encountered by the amphibious vehicle during driving. It can be specifically expressed as:

[0095] E r_current =w f ·k f ·(v c ·cos(θ-d c )).

[0096] In the formula, v c represents the ocean current speed, d c Indicates the direction of ocean current, cos(θ-d c) is the angle between the heading angle θ of the amphibious vehicle and the direction of the ocean current, which is used to calculate the actual ocean current speed felt by the amphibious vehicle in the forward direction. f is the ocean current resistance coefficient, w f It is the weight coefficient of ocean current resistance energy consumption, which is used to adjust the influence proportion of ocean current resistance energy consumption in total energy consumption.

[0097] Wave resistance energy consumption refers to the energy consumed by the wave resistance encountered by the amphibious vehicle during driving. It can be specifically expressed as:

[0098] E r_wave =w w ·k w ·(h w ·f w ).

[0099] In the formula, h w represents the wave height, f w Represents the wave frequency, the product of wave height and frequency h w ·f w It represents the comprehensive resistance effect of waves on amphibious vehicles. w is the wave resistance coefficient, w w The weight coefficient of wave resistance energy consumption is used to adjust the influence ratio of wave resistance energy consumption in the total energy consumption.

[0100] The comprehensive environmental resistance energy consumption is expressed as the sum of wind resistance, ocean current resistance and wave resistance. The specific formula is:

[0101] E r =E r_wind +E r_current +E r_wave .

[0102] In summary, the total energy consumption model is obtained by combining the propulsion energy consumption and the environmental resistance energy consumption:

[0103] E=E p +E r .

[0104] like Figure 3 As shown in Figure 2, in practical applications, the process of sensor data collection and environment modeling is as follows:

[0105] The first step is to collect environmental data. Corresponding data is obtained based on wind speed and direction sensors, ocean current speed and direction sensors, wave and frequency sensors, and GPS and IMU data are also obtained.

[0106] The acquired data is processed by Coleman filtering and data fusion, and then the environmental model is updated. The environmental model includes: ocean current model (i.e. ocean current resistance energy consumption), wind field model (i.e. wind resistance energy consumption model) and wave model (i.e. wave resistance energy consumption).

[0107] Step 3: Reinforcement learning parameter training. The deep Q network (DQN) is used to learn and adjust the parameters of the path planning heuristic function, and the path is dynamically adjusted to optimize the path planning based on real-time environmental data and energy consumption models. The input of the reinforcement learning model includes the current environmental state and the state of the amphibious vehicle, and the output is the updated heuristic function parameters and dynamic adjustment strategy. The specific process is as follows:

[0108] The state space includes the wind speed v w , wind direction w , ocean current speed v c , ocean current direction d c 、wave height h w , wave frequency f w and the current position (x, y), velocity v and heading angle θ of the amphibious vehicle. Therefore, the state can be expressed as:

[0109] s=[v w ,d w ,v c ,d c ,h w ,f w ,x,y,v,θ].

[0110] Accordingly, the action space can be expressed as:

[0111] a=[Δw d ,Δw c ,Δw f ,Δw w ,Δθ,Δv].

[0112] This vector represents all the actions that the reinforcement learning model can take, including adjusting the parameters of the heuristic function and the specific actions of dynamically adjusting the path.

[0113] The reward function is defined as the negative of the total energy consumption of the path, taking into account the path smoothness and adjustment effects:

[0114]

[0115] The steps for training updates are as follows:

[0116] Initialize the weights and experience replay buffer of the deep Q network.

[0117] At each time step, the current environment state s and the state of the amphibious vehicle are collected, and an action a is selected (adjusting the heuristic function parameters or dynamically adjusting the path).

[0118] Execute action a and obtain the new state s' and reward R(s,a).

[0119] Store the current state s, action a, reward R(s,a), and next state s' into the experience replay buffer.

[0120] Randomly extract small batches of samples from the experience replay buffer and use these samples to update the weights of the Q network. Repeat the training until the Q network converges and obtains the optimal heuristic function parameter adjustment and path dynamic adjustment strategy.

[0121] Step 4: Path planning and adjustment. Based on the current environment model and reinforcement learning strategy, dynamically calculate and adjust the path planning, use the optimized heuristic function to calculate the path cost, and combine the A* algorithm to search the path, and make dynamic adjustments during the execution process. Specifically:

[0122] The heuristic function comprehensively considers factors such as path length, environmental resistance energy consumption, and propulsion energy consumption, and is expressed as:

[0123]

[0124] Among them, w d is the weight of the path length, (x n ,y n ) is the node coordinate of the current position, (x g ,y g ) is the target location node coordinate, E r is the environmental resistance energy consumption, E p To promote energy consumption.

[0125] The A* algorithm is used to search and generate a path, and during the execution of the path, the path is dynamically adjusted according to the real-time environmental data and the status of the amphibious vehicle. The reinforcement learning model continuously updates the heuristic function parameters and calculates the optimal path in real time. If the environment changes significantly (such as wind speed, ocean current direction, etc.), the system will recalculate the path to ensure that the path planning is always optimal. For the reinforcement learning of the deep Q network and the process of experience playback and model update, see Figure 4 .

[0126] Step 5: Experience replay and model update. During the path planning process, the experience replay mechanism is used to store and utilize historical data to train the neural network. The specific process is as follows:

[0127] Experience storage: During each path planning process, the current state, action, reward, and next state are stored in the experience replay buffer.

[0128] Mini-batch sample update: randomly draw mini-batch samples from the experience replay buffer and use these samples to update the weights of the Q network.

[0129] Model update: Through repeated training and updating, the reinforcement learning model can better adjust the heuristic function parameters and dynamically adjust the path to optimize the path planning strategy.

[0130] The benefits of the present invention are:

[0131] First, this application establishes a dynamic environmental model by collecting and processing environmental data (such as wind speed, wind direction, ocean current speed, ocean current direction, wave height and frequency) in real time, ensuring that path planning can adapt to environmental changes in real time, improving accuracy and robustness. Secondly, this application combines the reinforcement learning algorithm and the A algorithm to optimize the heuristic function parameters of the A algorithm and improve the efficiency and accuracy of path search. The heuristic function comprehensively considers factors such as path length, propulsion energy consumption, and environmental resistance energy consumption to ensure path planning with the lowest energy consumption. In addition, the reinforcement learning model not only optimizes the heuristic function parameters, but also dynamically adjusts according to real-time environmental data and the status of the amphibious vehicle during path execution to ensure that path planning is always optimal. This application significantly improves the autonomous operation capability and task completion efficiency of amphibious vehicles. Through real-time data processing and dynamic path adjustment, amphibious vehicles can complete more tasks under limited energy conditions.

[0132] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A dynamic adaptive amphibious vehicle path and energy consumption optimization method, characterized in that: The dynamic adaptive amphibious vehicle path and energy consumption optimization method comprises: Acquire information data; the information data includes: environmental data and data related to the amphibious vehicle itself; the environmental data includes: wind speed, wind direction, ocean current speed, ocean current direction, wave height and wave frequency; the data related to the amphibious vehicle itself includes: position, speed and acceleration; the information data includes a timestamp; The information data is input into the path and energy consumption optimization model to obtain the optimized planning path information; the optimized planning path information is the planning path with the lowest energy consumption determined based on the heuristic function; the path and energy consumption optimization model is a reinforcement learning algorithm and A * The algorithm is obtained after updating and optimizing the amphibious vehicle surface energy consumption model and the heuristic function; the amphibious vehicle surface energy consumption model is a mathematical model determined according to the dynamic characteristics of the amphibious vehicle and environmental factors.

2. The dynamic adaptive amphibious vehicle path and energy consumption optimization method according to claim 1 is characterized in that: The amphibious vehicle surface energy consumption model includes a propulsion energy consumption model and an environmental resistance energy consumption model; the expression of the amphibious vehicle surface energy consumption model is: E=E p +E r ; Where E is the energy consumption of the amphibious vehicle on the water surface; E p To promote energy consumption; E r Energy consumption for environmental resistance.

3. The dynamic adaptive amphibious vehicle path and energy consumption optimization method according to claim 2 is characterized in that: The expression of the propulsion energy consumption model is: Among them, F p is the propulsion force; d is the driving distance, and η is the efficiency of the propulsion system.

4. The dynamic adaptive amphibious vehicle path and energy consumption optimization method according to claim 2 is characterized in that: The expression of the environmental resistance energy consumption model is: AND r =And r_wind +E r_current +E r_wave ; E r_wind =w c ·k c ·(v w ·cos(θ-d w )); E r_current =w f ·k f ·(v c ·cos(θ-d c )); E r_wave =w w ·k w ·(h w ·f w ); Among them, E r_wind E is wind resistance energy consumption; r_current E is the energy consumption of ocean current resistance; r_wave is the energy consumption of wave resistance; w c is the weight coefficient of wind resistance energy consumption; k c is the wind resistance coefficient; v w is the wind speed; θ is the heading angle; d w is the wind direction; cos(θ-d w ) is the angle between the heading angle of the amphibious vehicle and the wind direction; w f is the weight coefficient of ocean current resistance energy consumption; k f is the ocean current resistance coefficient; v c is the ocean current speed; d c is the direction of ocean current; cos(θ-d c ) is the angle between the heading angle of the amphibious vehicle and the direction of the ocean current; h w is the wave height; f w is the wave frequency; k w is the wave resistance coefficient; w w is the weight coefficient of wave resistance energy consumption.

5. The dynamic adaptive amphibious vehicle path and energy consumption optimization method according to claim 1 is characterized in that: The expression of the heuristic function is: Among them, h″(n) is the heuristic function; w d is the weight of the path length; (x n ,y n ) is the node coordinate of the current position; (x g ,y g ) is the target location node coordinate; E r E is the environmental resistance energy consumption; p To promote energy consumption.

6. The method for dynamically adaptive amphibious vehicle path and energy consumption optimization according to claim 1, characterized in that: The method for determining the path and energy consumption optimization model specifically includes: Obtain training information data; Constructing a deep Q network; the deep Q network includes an experience replay buffer; Initialize the experience replay buffer and weight parameters of the deep Q network; Determine the state space and action space based on the training information data; the expression of the state space s is: s = [v w , d w , v c , d c ,h w , f w , x, y, v, θ]; the expression of action space a is: a=[Δw d , Δw c , Δw f , Δw w , Δθ, Δv]; v w is the wind speed; d w is the wind direction; v c is the ocean current speed; d c is the direction of ocean current; h w is the wave height; f w is the wave frequency; θ is the heading angle; x is the horizontal coordinate of the position; y is the vertical coordinate of the position; v is the speed; Δw c is the adjustment of wind resistance energy consumption weight coefficient; Δw d Δw is the path length weight adjustment; f Δw is the adjustment of the energy consumption weight coefficient of ocean current resistance; w is the adjustment of the wave resistance energy consumption weight coefficient; Δθ is the heading angle adjustment; Δv is the speed adjustment; For any time step, the corresponding action space is selected based on the current state space, and the next state space is determined according to the corresponding reward function; the time step is determined after time point alignment processing is performed according to the timestamp; the reward function is determined according to the state space and the action space; Randomly extract sample data from the stored experience playback buffer, and then * The algorithm performs path search, and based on the heuristic function, trains and updates the weight parameters of the deep Q network with the goal of network convergence to obtain a trained deep Q network; the stored experience replay buffer stores the state space, action space, reward function and next state space corresponding to each time step; The trained deep Q network is determined as the path and energy consumption optimization model.

7. The dynamic adaptive amphibious vehicle path and energy consumption optimization method according to claim 6 is characterized in that: The expression of the reward function is: Among them, R(s, a) is the reward function; E is the energy consumption of the amphibious vehicle on the water surface; α, β, and γ are all weight parameters; a i is the action at the i-th time step; δ j is the result of the jth training update adjustment; n is the total number of time steps; m is the total number of training updates; a i+1 is the action at the i+1th time step.

8. The method for dynamically adaptive amphibious vehicle path and energy consumption optimization according to claim 1, characterized in that: The information data obtained specifically includes: Obtain initial information data; Align the initial information data according to the timestamp; A Kalman filter is used to remove noise from the aligned initial information data to obtain processed data; The processed data are fused by using a weighted average method to obtain the information data.

Images