An intelligent control method and system for underwater robots based on deep learning

Abstract

The invention discloses an underwater robot intelligent control method and system based on deep learning, and relates to the technical field of underwater obstacle avoidance of underwater robots. The method comprises steps 1 to 4, wherein a deep learning calibration model is established through a deep learning algorithm, an avoidance rate index BrsL in the underwater robot avoidance process is constructed, the ocean current speed and direction in the area where the underwater robot is located are monitored, and the real-time vertical descent depth, underwater temperature value and diving seawater pressure value of the propeller in the process of encountering obstacle avoidance are monitored, an ocean current impact evaluation index YL and a relative strength index Xd are calculated, and an evaluation is performed. If a multi-level calibration control strategy of a first calibration instruction and a second calibration instruction is triggered, the propeller power and the underwater robot posture are adjusted, and timely adjustments are made according to the change of the vertical descent depth to adapt to the ocean current environment, so as to ensure the stable movement of the underwater robot in the process of encountering obstacle avoidance.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater obstacle avoidance of underwater robots, and in particular to an intelligent control method and system for underwater robots based on deep learning. Background Art

[0002] Underwater robots play a vital role in underwater maintenance of the ocean. However, the high pressure, low temperature, darkness and other characteristics of the seabed environment make it impossible for humans to directly observe and operate. As an effective tool, underwater robots provide us with the possibility of exploring, studying and maintaining the ocean bottom.

[0003] However, underwater robots inevitably face huge pressure challenges when working on the seabed. In addition, when performing tasks, underwater robots also need to avoid objects such as seabed organisms and rocks to protect themselves and the marine environment. However, the avoidance process may be affected by a variety of complex factors, such as changes in ocean currents, temperature and pressure. In traditional technologies, fixed avoidance paths, pre-set speeds and movement directions are usually selected, which have some limitations and fail to respond in time according to real-time ocean current changes. This may cause instability in the underwater robot during the avoidance process and even affect its mission execution efficiency and safety. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides an underwater robot intelligent control method and system based on deep learning to solve the problems mentioned in the background technology.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an underwater robot intelligent control system based on deep learning, comprising:

[0008] The underwater robot navigation acquisition module is used to use laser radar to identify and classify different types of obstacles, and after building a three-dimensional environment model, collect the diameter zj of the environmental obstacle in real time, and set it to three times the diameter of the obstacle diameter zj for emergency avoidance, and generate the first avoidance path;

[0009] The avoidance process acquisition module collects the data of the thruster's speed reduction, vertical up and down movement depth, and lateral movement distance during the process of encountering obstacles and establishes an avoidance operation data set;

[0010] The first analysis module establishes an underwater avoidance three-dimensional model based on the avoidance operation data set and calculates the avoidance rate index BrsL during the avoidance process of the underwater robot;

[0011] The ocean current assessment module monitors the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtains a first data set, uses a deep learning algorithm to establish a deep learning calibration model, and uses fluid dynamics and an avoidance operation data set to construct an ocean current impact index YL. If the ocean current impact index YL exceeds a first load threshold value X, a first calibration instruction is issued;

[0012] The second influencing module introduces a second calibration adjustment mechanism in the deep learning calibration model. When the underwater robot is in working state, the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller during obstacle avoidance are collected in real time. The real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl are used as the second input value. Through the deep learning calibration model, the relative strength index Xd is calculated according to the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl. If the relative strength index Xd causes the reading deviation of the underwater robot propeller speed sensor to exceed the allowable range, the underwater robot is subjected to a second calibration.

[0013] Preferably, the avoidance process acquisition module includes an environment perception unit, an aquatic positioning unit and a first avoidance path generation unit;

[0014] The environmental perception unit is used to use laser radar to identify and classify different types of obstacles, including organisms, rocks and artificial structures;

[0015] Aquatic positioning unit, which is used to integrate seafloor topography and the locations of known obstacles in deep-sea environments, and build a three-dimensional environmental model using the map information system GIS by combining sonar, LiDAR and camera data;

[0016] Generate the first avoidance path unit, which is used to collect diverse data sets in different environments, and use the underwater simulation platform to generate simulated data sets. After algorithm verification and training, collect the environmental obstacle diameter zj in real time, and set it to three times the diameter of the obstacle diameter zj for emergency avoidance, and generate the first avoidance path.

[0017] Preferably, the avoidance process acquisition module includes a data acquisition unit and a speed reduction calculation unit;

[0018] The data acquisition unit is used to pre-process the collected avoidance operation data set, including data cleaning, outlier removal and data alignment;

[0019] The descent speed calculation unit is used to calculate the average descent speed Jsd of the propeller of the underwater robot when it encounters an obstacle. The speed sensor collects the time series data of the speed change of the propeller when it encounters an obstacle. Assume that the collected speed data sequence is , where t is the time, is the speed at the corresponding time point, and extracts the speed data sequence The initial velocity when encountering an obstacle and the final speed at the end of the avoidance process , and the duration of the avoidance process , after dimensionless processing, the average drop speed of the thruster during the avoidance process is calculated by the following formula: :

[0020] .

[0021] Preferably, the avoidance process acquisition module further includes a vertical up and down movement depth calculation unit and a lateral movement distance calculation unit;

[0022] The vertical up and down movement depth calculation unit is used to record the vertical movement depth change data of the propeller when encountering an obstacle, and calculate the average vertical movement depth during the avoidance process by the following formula: :

[0023] ;

[0024] in, is the vertical depth of the kth sampling point, n is the total number of sampling points, and k is an integer from 1 to n, which is used to represent the position of each vertical moving depth data point;

[0025] The lateral movement distance calculation unit is used to record the lateral movement distance change data of the propeller when encountering an obstacle, and calculate the average lateral movement distance by the following formula :

[0026] ;

[0027] In the formula, is the time when the avoidance process starts, is the time when the avoidance process ends, is the duration of the avoidance process, is the lateral displacement of the propeller at time t, Indicates the horizontal movement distance For the rate of change of time t, when time t is continuous, the integral symbol ∫ represents the cumulative summation of the function over a given time period; here dt is a tiny change in time, representing the time increment when taking infinitesimal steps forward on the time axis.

[0028] Preferably, the first analysis module includes a modeling unit and an avoidance rate calculation unit,

[0029] The modeling unit is used to establish an underwater avoidance three-dimensional model based on the pre-processing of the avoidance operation data set and the integration of the three-dimensional environment model, and input the first avoidance path into the underwater avoidance three-dimensional model by drawing path lines in the model or importing path data into a path controller in the model;

[0030] The avoidance rate calculation unit is used to calculate the avoidance rate index BrsL during the avoidance process of the underwater robot according to the avoidance operation data set as input to the underwater avoidance three-dimensional model; the avoidance rate index BrsL is obtained by extracting the average descent speed , average vertical movement depth and the average lateral movement distance , after dimensionless processing, the avoidance rate index BrsL is generated by the following formula:

[0031]

[0032] The meaning of the formula is: the larger the BrsL is, the higher the avoidance rate of the underwater robot when encountering obstacles.

[0033] Preferably, the ocean current assessment module includes an ocean current collection unit and an impact calculation unit;

[0034] The ocean current acquisition unit is used to monitor the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtain a first data set and perform dimensionless processing, and the first data set includes the ocean current speed , underwater robot weight , the angle difference between the ocean current and the underwater robot's movement direction and the surface area of ​​the underwater robot ;

[0035] The impact calculation unit is used to use a deep learning algorithm to establish a deep learning calibration model, and to construct an ocean current impact index YL based on fluid dynamics and the first data set and the avoidance operation data set:

[0036]

[0037] In the formula, The real-time deceleration speed of the underwater robot in the ocean current. Indicates calculating the real-time descent speed of the underwater robot when it is impacted by the ocean current and average descent speed The absolute difference of Indicates the real-time vertical movement depth of the underwater robot, Indicates the calculation of the real-time vertical movement depth of the underwater robot when the ocean current hits it and the average vertical travel depth The absolute difference of Indicates the real-time lateral movement distance of the underwater robot in the ocean current process, Indicates the calculation of the real-time lateral movement distance of the underwater robot when it is hit by the ocean current Average lateral movement distance The absolute difference of is expressed as the average ocean current velocity, where , ,and , w1, w2, w3 and w4 are weight values, and their specific values ​​are adjusted and set by the user;

[0038] The angle difference between the ocean current and the underwater robot's movement direction is calculated by the difference between the real-time angle of the underwater robot and the ocean current angle. is 0, indicating that the underwater robot moves in the same direction as the ocean current; if The larger the value, the more it means that the underwater robot's movement direction and the ocean current's movement direction are perpendicular or opposite to each other. The bigger it is, the greater the impact.

[0039] Preferably, the ocean current assessment module further comprises a first assessment unit,

[0040] The first evaluation unit is used to set a first load threshold value X, and compare and evaluate the ocean current impact index YL with the first load threshold value X to obtain a first evaluation result, including:

[0041] If the ocean current impact index YL> the first load threshold X, it means that the underwater robot is affected by the current ocean current environment and exceeds the first load threshold X, then a first calibration instruction is issued to reduce the power or propulsion of the thruster by 20%, and adjust the posture of the underwater robot, and adjust the angle of the rudder or propeller to be consistent with the direction of the ocean current movement;

[0042] If the ocean current influence index YL ≤ the first load threshold X, it means that the underwater robot is not affected by the current ocean current environment beyond the first load threshold X, and no calibration is performed.

[0043] Preferably, the second influencing module includes a second acquisition unit and a second calculation unit;

[0044] The second acquisition unit is used to acquire the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller in the process of avoiding obstacles when the underwater robot is in working state;

[0045] The second calculation unit is used to input the real-time vertical descent depth sdz, the underwater temperature value wd and the diving seawater pressure value yl as the second input value after dimensionless processing into the deep learning calibration model, and use the regression coefficient corresponding to the measurement condition data in the regression equation as the influencing factor to calculate the relative strength index Xd through the following formula:

[0046]

[0047] in, is the influencing factor of the real-time vertical descent depth sdz, is the influencing factor of the underwater temperature value wd; is the influencing factor of the diving seawater pressure value yl; , and is the weight coefficient, , , C is the preset correction factor coefficient, and its value is 1.125.

[0048] Preferably, the second impact module further includes a third evaluation unit, which is used to preset a deviation threshold P and compare the relative strength index Xd with the deviation threshold P to obtain a second evaluation result, including:

[0049] If the relative strength index Xd>deviation threshold P, it means that the deviation exceeds the allowable range, and the second calibration instruction is triggered to adjust and slow down the power or thrust of the thruster by 3%-5%, and slow down the power or thrust of the thruster by 2% for every 10 meters of the real-time vertical descent depth sdz;

[0050] If the relative strength index Xd ≤ the deviation threshold P, it means that the deviation does not exceed the allowable range and no calibration is performed.

[0051] A method for intelligent control of underwater robots based on deep learning, step 1, using laser radar to identify and classify different types of obstacles and build a three-dimensional environment model; collecting the diameter zjj of environmental obstacles in real time, and setting it to three times the diameter zj of the obstacle for emergency avoidance, and generating a first avoidance path;

[0052] Step 2: Collect the data of the thruster's descent speed, vertical up and down movement depth, and lateral movement distance during obstacle avoidance, establish an avoidance operation data set, build an underwater avoidance three-dimensional model, and calculate the avoidance rate index BrsL during the underwater robot's avoidance process;

[0053] Step 3: Monitor the speed and direction of the ocean current in the area where the underwater robot is located, obtain a first data set, use a deep learning algorithm to establish a deep learning calibration model, construct and evaluate the ocean current impact index YL, and if the ocean current impact index YL exceeds the first load threshold X, issue a first calibration instruction, including: slowing down the power or propulsion of the propeller by 20%, adjusting the posture of the underwater robot, and adjusting the angle of the rudder or propeller to be the same as the direction of the ocean current movement;

[0054] Step 4: collect the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the thruster during obstacle avoidance, use the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl as the second input value, calculate and evaluate the relative strength index Xd, and if the relative strength index Xd exceeds the deviation threshold P, issue a second calibration instruction, including: adjusting and slowing down the power or thrust of the thruster by 3%-5%, and slowing down the power or thrust of the thruster by 2% for every 10 meters decrease in the real-time vertical descent depth sdz.

[0055] (III) Beneficial effects

[0056] The present invention provides an underwater robot intelligent control method and system based on deep learning, which has the following beneficial effects:

[0057] (1) The system uses lidar to perceive the seabed environment in real time, build a three-dimensional environmental model, and make real-time decisions and adjustments based on deep learning algorithms. Specifically, the system can autonomously generate avoidance paths based on real-time collected data, and make intelligent adjustments based on factors such as seabed currents, temperature, and pressure to ensure that the underwater robot can safely and efficiently avoid obstacles when encountering them, thereby ensuring the smooth execution of the mission.

[0058] (2) The issuance of the first calibration instruction is based on the ocean current impact index YL, so that the underwater robot can adjust the propeller power or thrust in time and adjust the posture to adapt to the changes in the ocean current, thereby reducing the interference of the ocean current on the robot's movement and improving the underwater robot's movement stability and control ability. The generation of the first avoidance path and the issuance of the second calibration instruction are based on deep learning analysis and decision-making based on the obstacle diameter, propeller motion data and environmental parameters collected in real time. This intelligent decision-making mechanism enables the underwater robot to quickly and accurately adjust the path and speed when encountering obstacles, effectively avoiding collisions and accidents, and improving the safety and stability of the movement process. The issuance of the first calibration instruction and the implementation of the second calibration instruction, by calibrating the propeller power or thrust, adjusting the posture, and adjusting the robot's movement speed and direction according to environmental parameters, can enable the underwater robot to maintain a more stable movement trajectory during the avoidance process, reduce energy loss, and improve energy utilization efficiency.

[0059] (3) The first calibration instruction and the second calibration instruction, multi-level control steps, reduce the need for manual intervention. Through real-time monitoring and intelligent adjustment, the underwater robot can autonomously complete avoidance and adjustment in a complex seabed environment, reducing the burden on operators and the errors and risks caused by human operation. It promotes the improvement of the motion stability and intelligence level of the underwater robot, enabling it to perform seabed maintenance tasks more efficiently. The stable motion trajectory and intelligent avoidance decision-making mechanism will help improve the efficiency of task execution, reduce task execution time, and thus save resources and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a schematic diagram of a process flow of an underwater robot intelligent control system based on deep learning in the present invention;

[0061] Figure 2 This is a schematic flow chart of the steps of an underwater robot intelligent control method based on deep learning in the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Example 1

[0064] See also Figure 1 The present invention provides an underwater robot intelligent control system based on deep learning, comprising:

[0065] The underwater robot navigation acquisition module is used to use laser radar to identify and classify different types of obstacles, and after building a three-dimensional environment model, collect the diameter zj of the environmental obstacle in real time, and set it to three times the diameter of the obstacle diameter zj for emergency avoidance, and generate the first avoidance path;

[0066] The avoidance process acquisition module collects the data of the thruster's speed reduction, vertical up and down movement depth, and lateral movement distance during the process of encountering obstacles and establishes an avoidance operation data set;

[0067] The first analysis module establishes an underwater avoidance three-dimensional model based on the avoidance operation data set and calculates the avoidance rate index BrsL during the avoidance process of the underwater robot;

[0068] The ocean current assessment module monitors the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtains a first data set, uses a deep learning algorithm to establish a deep learning calibration model, and uses fluid dynamics and an avoidance operation data set to construct an ocean current impact index YL. If the ocean current impact index YL exceeds a first load threshold value X, a first calibration instruction is issued;

[0069] The second influencing module introduces a second calibration adjustment mechanism in the deep learning calibration model. When the underwater robot is in working state, the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller in the process of avoiding obstacles are collected in real time. The real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl are used as the second input value. Through the deep learning calibration model, the relative strength index Xd is calculated according to the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl. If the relative strength index Xd causes the reading deviation of the underwater robot propeller speed sensor to exceed the allowable range, the underwater robot is subjected to a second calibration.

[0070] In this embodiment, the system uses laser radar to perceive the seabed environment in real time, build a three-dimensional environmental model, and make real-time decisions and adjustments based on deep learning algorithms. Specifically, the system can autonomously generate avoidance paths based on real-time collected data, and make intelligent adjustments based on factors such as seabed currents, temperature, and pressure to ensure that the underwater robot can safely and efficiently avoid obstacles when encountering them, thereby ensuring the smooth execution of the mission.

[0071] The issuance of the first calibration instruction is based on the ocean current impact index YL, so that the underwater robot can adjust the propeller power or propulsion force in time, and adjust the posture to adapt to the changes in the ocean current, thereby reducing the interference of the ocean current on the robot's movement and improving the underwater robot's movement stability and control ability. The generation of the first avoidance path and the issuance of the second calibration instruction are based on deep learning analysis and decision-making based on the obstacle diameter, propeller motion data and environmental parameters collected in real time. This intelligent decision-making mechanism enables the underwater robot to quickly and accurately adjust the path and speed when encountering obstacles, effectively avoiding collisions and accidents, and improving the safety and stability of the movement process. The issuance of the first calibration instruction and the implementation of the second calibration instruction, by calibrating the propeller power or propulsion force, adjusting the posture, and adjusting the robot's movement speed and direction according to environmental parameters, can enable the underwater robot to maintain a more stable motion trajectory during the avoidance process, reduce energy loss, and improve energy utilization efficiency.

[0072] This multi-level control step reduces the need for human intervention. Through real-time monitoring and intelligent adjustment, the underwater robot can autonomously avoid and adjust in complex seabed environments, reducing the burden on operators and the errors and risks caused by human operation. It promotes the improvement of the motion stability and intelligence level of the underwater robot, enabling it to perform seabed maintenance tasks more efficiently. Stable motion trajectories and intelligent avoidance decision-making mechanisms will help improve the efficiency of task execution and reduce task execution time, thereby saving resources and costs.

[0073] Example 2: This example is an explanation of Example 1. Figure 1 ,Specifically, the avoidance process acquisition module includes an environment perception unit, an aquatic positioning unit, and a first avoidance path generation unit;

[0074] The environmental perception unit is used to use lidar to identify and classify different types of obstacles, including organisms, rocks and artificial structures; this comprehensive environmental perception capability ensures that the underwater robot has a comprehensive understanding of the surrounding environment when performing tasks, helping to avoid collisions and accidents.

[0075] The aquatic positioning unit is used to integrate seabed topographic maps and the locations of known obstacles in deep-sea environments, and combine sonar, LiDAR and camera data to build a three-dimensional environmental model using the map information system GIS. This precise aquatic positioning capability helps underwater robots accurately identify obstacle locations and plan safe avoidance paths.

[0076] Generate the first avoidance path unit, which is used to collect diverse data sets in different environments, and use the underwater simulation platform to generate simulated data sets. After algorithm verification and training, collect the environmental obstacle diameter zj in real time, and set it to three times the diameter of the obstacle diameter zj for emergency avoidance, and generate the first avoidance path. By generating simulated data sets on the underwater simulation platform and performing algorithm verification and training, it is ensured that the underwater robot can effectively deal with various complex situations in actual tasks, and improve the success rate and safety of task execution.

[0077] In this embodiment, the diameter zj of environmental obstacles is collected in real time, and an emergency avoidance path is generated according to the set rules, so that the underwater robot can respond to emergencies quickly and flexibly to avoid collisions and damage. This emergency avoidance path generation mechanism effectively improves the emergency response capability of the underwater robot and ensures the smooth execution of the task and the safe operation of the underwater robot.

[0078] Example 3, this example is an explanation of Example 1, please refer to Figure 1 Specifically, the avoidance process acquisition module includes a data acquisition unit and a speed reduction calculation unit;

[0079] The data acquisition unit is used to pre-process the collected avoidance operation data set, including data cleaning, outlier removal and data alignment;

[0080] The descent speed calculation unit is used to calculate the average descent speed Jsd of the propeller of the underwater robot when it encounters an obstacle. The speed sensor collects the time series data of the speed change of the propeller when it encounters an obstacle. Assume that the collected speed data sequence is , where t is the time, is the speed at the corresponding time point, and extracts the speed data sequence The initial velocity when encountering an obstacle and the final speed at the end of the avoidance process , and the duration of the avoidance process , after dimensionless processing, the average drop speed of the thruster during the avoidance process is calculated by the following formula: :

[0081] .

[0082] Specifically, the avoidance process acquisition module also includes a vertical up and down movement depth calculation unit and a lateral movement distance calculation unit;

[0083] The vertical up and down movement depth calculation unit is used to record the vertical movement depth change data of the propeller when encountering an obstacle, and calculate the average vertical movement depth during the avoidance process by the following formula: :

[0084] ;

[0085] in, is the vertical depth of the kth sampling point, n is the total number of sampling points, and k is an integer from 1 to n, which is used to represent the position of each vertical moving depth data point;

[0086] The lateral movement distance calculation unit is used to record the lateral movement distance change data of the propeller when encountering an obstacle, and calculate the average lateral movement distance by the following formula :

[0087] ;

[0088] In the formula, is the time when the avoidance process starts, is the time when the avoidance process ends, is the duration of the avoidance process, is the lateral displacement of the propeller at time t, Indicates the horizontal movement distance For the rate of change of time t, when time t is continuous, the integral symbol ∫ represents the cumulative summation of the function over a given time period; here dt is a tiny change in time, representing the time increment when taking infinitesimal steps forward on the time axis.

[0089] In this embodiment, the data acquisition unit is responsible for preprocessing the collected avoidance operation data set, including data cleaning, removing outliers and data alignment. This can ensure that the quality of the data set used is high, effectively reduce the errors and uncertainties caused by data quality problems, and improve the stability and reliability of the system. The descent speed calculation unit can accurately calculate the average descent speed of the underwater robot when it encounters an obstacle. By processing the speed change data collected by the speed sensor, the average descent speed of the propeller during the avoidance process is obtained, which helps the underwater robot to perform safe and effective avoidance operations when encountering obstacles. The vertical up and down movement depth calculation unit and the lateral movement distance calculation unit respectively record the vertical direction and lateral movement depth change data of the underwater robot when encountering obstacles. By processing and analyzing these data, the average vertical movement depth and average lateral movement distance during the avoidance process of the underwater robot can be obtained, which helps to evaluate the avoidance effect and motion stability of the underwater robot. The calculation and analysis of multiple parameters such as descent speed, vertical movement depth and lateral movement distance are combined to fully understand the motion state and behavior characteristics of the underwater robot during the avoidance process. This helps to optimize the avoidance algorithm and control strategy, and improve the avoidance efficiency and motion stability of underwater robots in complex environments.

[0090] Example 4: This example is an explanation of Example 3. Figure 1 Specifically, the first analysis module includes a modeling unit and an avoidance rate calculation unit.

[0091] The modeling unit is used to pre-process the avoidance operation data set, integrate it into the three-dimensional environment model, establish an underwater avoidance three-dimensional model, and input the first avoidance path into the underwater avoidance three-dimensional model by drawing path lines in the model or importing path data into the path controller in the model; by inputting the first avoidance path into the model, the path can be visualized in real time in the model, so as to further analyze and optimize the avoidance path and improve the avoidance efficiency of the underwater robot.

[0092] The avoidance rate calculation unit is used to calculate the avoidance rate index BrsL during the avoidance process of the underwater robot according to the avoidance operation data set as input to the underwater avoidance three-dimensional model; the avoidance rate index BrsL is obtained by extracting the average descent speed , average vertical movement depth and the average lateral movement distance , after dimensionless processing, the avoidance rate index BrsL is generated by the following formula:

[0093]

[0094] The meaning of the formula is: the larger the BrsL is, the higher the avoidance rate of the underwater robot when encountering obstacles.

[0095] In this embodiment, the avoidance rate calculation unit calculates the avoidance rate index BrsL during the avoidance process of the underwater robot according to the avoidance operation data set. This index comprehensively considers factors such as the average descent speed, the average vertical movement depth, and the average lateral movement distance, and can objectively evaluate the avoidance efficiency and speed of the underwater robot when encountering obstacles. Quantify the avoidance efficiency of the underwater robot when encountering obstacles. This helps to timely evaluate the avoidance performance of the underwater robot and provide a reference for optimizing the avoidance strategy and path planning.

[0096] Example 5. This example is an explanation of Example 1. Please refer to Figure 1 ,Specifically, the ocean current assessment module includes an ocean current acquisition unit and an ,impact calculation unit;

[0097] The ocean current acquisition unit is used to monitor the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtain a first data set and perform dimensionless processing, and the first data set includes the ocean current speed , underwater robot weight , the angle difference between the ocean current and the underwater robot's movement direction and the surface area of ​​the underwater robot ; Such real-time monitoring can help underwater robots adapt to the ocean current environment more accurately and improve the stability and efficiency of movement.

[0098] The impact calculation unit is used to use a deep learning algorithm to establish a deep learning calibration model, and to construct an ocean current impact index YL based on fluid dynamics and the first data set and the avoidance operation data set:

[0099]

[0100] In the formula, The real-time deceleration speed of the underwater robot in the ocean current. Indicates calculating the real-time descent speed of the underwater robot when it is impacted by the ocean current and average descent speed The absolute difference of Indicates the real-time vertical movement depth of the underwater robot, Indicates the calculation of the real-time vertical movement depth of the underwater robot when the ocean current hits it and the average vertical travel depth The absolute difference of Indicates the real-time lateral movement distance of the underwater robot in the ocean current process, Indicates the calculation of the real-time lateral movement distance of the underwater robot when it is hit by the ocean current Average lateral movement distance The absolute difference of is expressed as the average ocean current velocity, where , ,and , w1, w2, w3 and w4 are weight values, and their specific values ​​are adjusted and set by the user;

[0101] The angle difference between the ocean current and the underwater robot's movement direction is calculated by the difference between the real-time angle of the underwater robot and the ocean current angle. is 0, indicating that the underwater robot moves in the same direction as the ocean current; if The larger the value, the more it means that the underwater robot's movement direction and the ocean current's movement direction are perpendicular or opposite to each other. The bigger it is, the greater the impact.

[0102] Specifically, the ocean current assessment module further includes a first assessment unit,

[0103] The first evaluation unit is used to set a first load threshold value X, and compare and evaluate the ocean current impact index YL with the first load threshold value X to obtain a first evaluation result, including:

[0104] If the ocean current impact index YL> the first load threshold X, it means that the underwater robot is affected by the current ocean current environment and exceeds the first load threshold X, then a first calibration instruction is issued to reduce the power or propulsion of the thruster by 20%, and adjust the posture of the underwater robot, and adjust the angle of the rudder or propeller to be consistent with the direction of the ocean current movement;

[0105] If the ocean current influence index YL ≤ the first load threshold X, it means that the underwater robot is not affected by the current ocean current environment beyond the first load threshold X, and no calibration is performed.

[0106] In this embodiment, the impact calculation unit uses a deep learning algorithm to establish a deep learning calibration model, and combines fluid dynamics, the first data set and the avoidance operation data set to calculate the ocean current impact index YL. Through the deep learning calibration model, the force of the underwater robot in the ocean current environment can be more accurately evaluated, providing a basis for subsequent calibration and adjustment. Through the deep learning calibration model, the force of the underwater robot in the ocean current environment can be more accurately evaluated, providing a basis for subsequent calibration and adjustment. The first evaluation unit sets a first load threshold X, compares and evaluates the ocean current impact index YL with the first load threshold X, and obtains a first evaluation result. According to the evaluation result, the calibration strategy is dynamically triggered. If the ocean current impact index YL exceeds the threshold X, the first calibration instruction is issued to slow down the propeller power and adjust the posture to cope with the impact of the ocean current, thereby ensuring the stable movement of the underwater robot in a complex ocean current environment. By real-time monitoring of the ocean current environment and dynamically adjusting the motion parameters and posture of the underwater robot according to the ocean current impact index YL, the stability and adaptability of the underwater robot in the ocean current can be effectively improved. This helps to ensure the safety and stability of the underwater robot when performing tasks and improve the efficiency of task execution.

[0107] Example 6. This example is an explanation of Example 1. Please refer to Figure 1 ,Specifically, the second impact module includes a second acquisition unit and a second calculation unit;

[0108] The second acquisition unit is used to acquire the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller in the process of avoiding obstacles when the underwater robot is in working state;

[0109] The second calculation unit is used to input the real-time vertical descent depth sdz, the underwater temperature value wd and the diving seawater pressure value yl as the second input value after dimensionless processing into the deep learning calibration model, and use the regression coefficient corresponding to the measurement condition data in the regression equation as the influencing factor to calculate the relative strength index Xd through the following formula:

[0110]

[0111] in, is the influencing factor of the real-time vertical descent depth sdz, is the influencing factor of the underwater temperature value wd; is the influencing factor of the diving seawater pressure value yl; , and is the weight coefficient, , , C is the preset correction factor coefficient, and its value is 1.125.

[0112] Specifically, the second impact module further includes a third evaluation unit, which is used to preset a deviation threshold P and compare the relative strength index Xd with the deviation threshold P to obtain a second evaluation result, including:

[0113] If the relative strength index Xd>deviation threshold P, it means that the deviation exceeds the allowable range, and the second calibration instruction is triggered to adjust and slow down the power or thrust of the thruster by 3%-5%, and slow down the power or thrust of the thruster by 2% for every 10 meters of the real-time vertical descent depth sdz;

[0114] If the relative strength index Xd ≤ the deviation threshold P, it means that the deviation does not exceed the allowable range and no calibration is performed.

[0115] In this embodiment, the second calculation unit inputs the real-time collected data into the deep learning calibration model as the second input value after dimensionless processing. Through the analysis of the deep learning calibration model, the relative strength index Xd can be obtained, which is used to evaluate the movement of the underwater robot under different environmental conditions. The third evaluation unit presets the deviation threshold P, and compares and evaluates the relative strength index Xd with the deviation threshold P to obtain a second evaluation result. According to the evaluation result, the calibration strategy is dynamically triggered. If the relative strength index Xd exceeds the threshold P, a second calibration instruction is issued to adjust the propeller power and make corresponding adjustments according to the degree of change in the vertical descent depth to ensure the stable movement of the underwater robot. The second calibration instruction can be finely adjusted according to the specific value of the relative strength index Xd, including slowing down the power or propulsion of the propeller, and making appropriate adjustments according to the change in the vertical descent depth to ensure that the underwater robot maintains a stable state of motion when encountering obstacles and avoiding them.

[0116] Example 7, please refer to Figure 2 , an intelligent control method for an underwater robot based on deep learning, comprising the following steps:

[0117] Step 1: Use laser radar to identify and classify different types of obstacles and build a three-dimensional environment model; collect the diameter zj of environmental obstacles in real time, and set it to three times the diameter zj of the obstacle for emergency avoidance, and generate the first avoidance path;

[0118] Step 2: Collect the data of the thruster's descent speed, vertical up and down movement depth, and lateral movement distance during obstacle avoidance, establish an avoidance operation data set, build an underwater avoidance three-dimensional model, and calculate the avoidance rate index BrsL during the underwater robot's avoidance process;

[0119] Step 3: Monitor the speed and direction of the ocean current in the area where the underwater robot is located, obtain a first data set, use a deep learning algorithm to establish a deep learning calibration model, construct and evaluate the ocean current impact index YL, and if the ocean current impact index YL exceeds the first load threshold X, issue a first calibration instruction, including: slowing down the power or propulsion of the propeller by 20%, adjusting the posture of the underwater robot, and adjusting the angle of the rudder or propeller to be the same as the direction of the ocean current movement;

[0120] Step 4: collect the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the thruster during obstacle avoidance, use the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl as the second input value, calculate and evaluate the relative strength index Xd, and if the relative strength index Xd exceeds the deviation threshold P, issue a second calibration instruction, including: adjusting and slowing down the power or thrust of the thruster by 3%-5%, and slowing down the power or thrust of the thruster by 2% for every 10 meters decrease in the real-time vertical descent depth sdz.

[0121] In this embodiment, step 1 uses laser radar to identify and classify different types of obstacles, and collects environmental obstacle diameter data in real time, so that the underwater robot can sense the obstacles ahead in time and generate an emergency avoidance path, thereby avoiding collision or dangerous situations. Step 2 collects the descent speed, vertical up and down movement depth and lateral movement distance data of the propeller in the process of avoiding obstacles, and establishes an avoidance operation data set, which can provide data support for the analysis and optimization of the subsequent avoidance process. Step 3 monitors the speed and direction of the ocean current in the area where the underwater robot is located, establishes a deep learning calibration model through a deep learning algorithm, and evaluates the ocean current impact index YL in real time. If it exceeds the threshold, a first calibration instruction is issued to adjust the propeller power and the underwater robot posture to adapt to the ocean current environment and ensure the stable movement of the underwater robot. Step 4 collects the real-time vertical descent depth, underwater temperature value and diving seawater pressure value of the propeller in the process of avoiding obstacles in real time, and obtains the relative strength index Xd by calculation. If it exceeds the deviation threshold P, a second calibration instruction is issued to dynamically adjust the propeller power and make timely adjustments according to the changes in the vertical descent depth to ensure the stability of the underwater robot in different depths and temperature environments. By implementing a multi-level calibration control strategy from step one to step four, the underwater robot can achieve intelligent, stable and safe movement in the process of avoiding obstacles in a complex seabed environment, thereby improving the efficiency of task execution and ensuring the safety of the underwater robot and its surrounding environment.

[0122] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control system for underwater robots based on deep learning, characterized in that: include, The underwater robot navigation acquisition module is used to use laser radar to identify and classify different types of obstacles, and after building a three-dimensional environment model, collect the diameter zj of the environmental obstacle in real time, and set it to three times the diameter of the obstacle diameter zj for emergency avoidance, and generate the first avoidance path; The avoidance process acquisition module collects the data of the thruster's speed reduction, vertical up and down movement depth, and lateral movement distance during the process of encountering obstacles and establishes an avoidance operation data set; The avoidance process acquisition module includes a data acquisition unit and a speed reduction calculation unit; The data acquisition unit is used to pre-process the collected avoidance operation data set, including data cleaning, outlier removal and data alignment; The descent speed calculation unit is used to calculate the average descent speed Jsd of the propeller of the underwater robot when it encounters an obstacle. The speed sensor collects the time series data of the speed change of the propeller when it encounters an obstacle. Assume that the collected speed data sequence is , where t is the time, is the speed at the corresponding time point, and extracts the speed data sequence The initial velocity when encountering an obstacle and the final speed at the end of the avoidance process , and the duration of the avoidance process , after dimensionless processing, the average drop speed of the thruster during the avoidance process is calculated by the following formula: : ； The avoidance process acquisition module also includes a vertical up and down movement depth calculation unit and a lateral movement distance calculation unit; The vertical up and down movement depth calculation unit is used to record the vertical movement depth change data of the propeller when encountering an obstacle, and calculate the average vertical movement depth during the avoidance process by the following formula: : ； in, is the vertical depth of the kth sampling point, n is the total number of sampling points, and k is an integer from 1 to n, which is used to represent the position of each vertical moving depth data point; The lateral movement distance calculation unit is used to record the lateral movement distance change data of the propeller when encountering an obstacle, and calculate the average lateral movement distance by the following formula : ； In the formula, is the time when the avoidance process starts, is the time when the avoidance process ends, is the duration of the avoidance process, is the lateral displacement of the propeller at time t, Indicates the horizontal movement distance For the rate of change of time t, when time t is continuous, the integral symbol ∫ represents the cumulative summation of the function within a given time period; here dt is a tiny time change, which represents the time increment when taking infinitesimal steps forward on the time axis; The first analysis module establishes an underwater avoidance three-dimensional model based on the avoidance operation data set and calculates the avoidance rate index BrsL during the avoidance process of the underwater robot; The first analysis module includes a modeling unit and an avoidance rate calculation unit, The modeling unit is used to establish an underwater avoidance three-dimensional model based on the pre-processing of the avoidance operation data set and the integration of the three-dimensional environment model, and input the first avoidance path into the underwater avoidance three-dimensional model by drawing path lines in the model or importing path data into a path controller in the model; The avoidance rate calculation unit is used to calculate the avoidance rate index BrsL during the avoidance process of the underwater robot according to the avoidance operation data set as input to the underwater avoidance three-dimensional model; the avoidance rate index BrsL is obtained by extracting the average descent speed , average vertical movement depth and the average lateral movement distance , after dimensionless processing, the avoidance rate index BrsL is generated by the following formula: ； The meaning of the formula is: the larger the BrsL, the higher the avoidance rate of the underwater robot when encountering obstacles; The ocean current assessment module monitors the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtains a first data set, uses a deep learning algorithm to establish a deep learning calibration model, and uses fluid dynamics and an avoidance operation data set to construct an ocean current impact index YL. If the ocean current impact index YL exceeds a first load threshold value X, a first calibration instruction is issued; The second influencing module introduces a second calibration adjustment mechanism in the deep learning calibration model. When the underwater robot is in working state, the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller in the process of avoiding obstacles are collected in real time. The real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl are used as the second input value. Through the deep learning calibration model, the relative strength index Xd is calculated according to the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl. If the relative strength index Xd causes the underwater robot propeller speed sensor reading deviation to exceed the allowable range, the underwater robot is subjected to a second calibration. The second impact module includes a second acquisition unit and a second calculation unit; The second acquisition unit is used to acquire the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the propeller in the process of avoiding obstacles when the underwater robot is in working state; The second calculation unit is used to input the real-time vertical descent depth sdz, the underwater temperature value wd and the diving seawater pressure value yl as the second input value after dimensionless processing into the deep learning calibration model, and use the regression coefficient corresponding to the measurement condition data in the regression equation as the influencing factor to calculate the relative strength index Xd through the following formula: ； in, is the influencing factor of the real-time vertical descent depth sdz, is the influencing factor of the underwater temperature value wd; is the influencing factor of the diving seawater pressure value yl; , and is the weight coefficient, , , , C is the preset correction factor coefficient, and its value is 1.

125.

2. The underwater robot intelligent control system based on deep learning according to claim 1, characterized in that: The avoidance process acquisition module includes an environment perception unit, an aquatic positioning unit and a first avoidance path generation unit; The environmental perception unit is used to use laser radar to identify and classify different types of obstacles, including organisms, rocks and artificial structures; The aquatic positioning unit is used to integrate the seafloor topography and the locations of known obstacles in a deep-sea environment, and to build a three-dimensional environmental model using a map information system (GIS) by combining sonar, LiDAR and camera data; The first avoidance path generating unit is used to collect diverse data sets in different environments, and use an underwater simulation platform to generate a simulated data set. After algorithm verification and training, the environmental obstacle diameter zj is collected in real time, and is set to three times the diameter of the obstacle diameter zj for emergency avoidance, thereby generating a first avoidance path.

3. The underwater robot intelligent control system based on deep learning according to claim 1, characterized in that: The ocean current assessment module includes an ocean current collection unit and an impact calculation unit; The ocean current acquisition unit is used to monitor the ocean current speed and direction in the area where the underwater robot is located when the underwater robot is in a working state, obtain a first data set and perform dimensionless processing, and the first data set includes the ocean current speed , underwater robot weight , the angle difference between the ocean current and the underwater robot's movement direction and the surface area of ​​the underwater robot ; The impact calculation unit is used to use a deep learning algorithm to establish a deep learning calibration model, and to construct an ocean current impact index YL based on fluid dynamics and the first data set and the avoidance operation data set: ； In the formula, The real-time deceleration speed of the underwater robot in the ocean current. Indicates calculating the real-time descent speed of the underwater robot when it is impacted by the ocean current and average descent speed The absolute difference of Indicates the real-time vertical movement depth of the underwater robot, Indicates the calculation of the real-time vertical movement depth of the underwater robot when the ocean current hits it and the average vertical travel depth The absolute difference of Indicates the real-time lateral movement distance of the underwater robot in the ocean current process, Indicates the calculation of the real-time lateral movement distance of the underwater robot when it is hit by the ocean current Average lateral movement distance The absolute difference of is expressed as the average ocean current velocity, where , , , ,and , w1, w2, w3 and w4 are weight values, and their specific values ​​are adjusted and set by the user; The angle difference between the ocean current and the underwater robot's movement direction is calculated by the difference between the real-time angle of the underwater robot and the ocean current angle. is 0, indicating that the underwater robot moves in the same direction as the ocean current; if The larger the value, the more it means that the underwater robot's movement direction and the ocean current's movement direction are perpendicular or opposite to each other. The bigger it is, the greater the impact.

4. The underwater robot intelligent control system based on deep learning according to claim 3, characterized in that: The ocean current assessment module further comprises a first assessment unit; The first evaluation unit is used to set a first load threshold value X, and compare and evaluate the ocean current impact index YL with the first load threshold value X to obtain a first evaluation result, including: If the ocean current impact index YL> the first load threshold X, it means that the underwater robot is affected by the current ocean current environment and exceeds the first load threshold X, then a first calibration instruction is issued to reduce the power or propulsion of the thruster by 20%, and adjust the posture of the underwater robot, and adjust the angle of the rudder or propeller to be consistent with the direction of the ocean current movement; If the ocean current influence index YL ≤ the first load threshold X, it means that the underwater robot is not affected by the current ocean current environment beyond the first load threshold X, and no calibration is performed.

5. According to the deep learning-based underwater robot intelligent control system of claim 1, the second impact module further comprises a third evaluation unit, the third evaluation unit is used to preset a deviation threshold P, and compare the relative strength index Xd with the deviation threshold P to obtain a second evaluation result, including: If the relative strength index Xd>deviation threshold P, it means that the deviation exceeds the allowable range, and the second calibration instruction is triggered to adjust and slow down the power or thrust of the thruster by 3%-5%, and slow down the power or thrust of the thruster by 2% for every 10 meters of the real-time vertical descent depth sdz; If the relative strength index Xd ≤ the deviation threshold P, it means that the deviation does not exceed the allowable range and no calibration is performed.

6. A method for intelligent control of an underwater robot based on deep learning, applied to an intelligent control system of an underwater robot based on deep learning according to any one of claims 1 to 5, characterized in that: Step 1: Use laser radar to identify and classify different types of obstacles and build a three-dimensional environment model; collect the diameter zj of environmental obstacles in real time, and set it to three times the diameter zj of the obstacle for emergency avoidance, and generate the first avoidance path; Step 2: Collect the data of the thruster's descent speed, vertical up and down movement depth, and lateral movement distance during obstacle avoidance, establish an avoidance operation data set, build an underwater avoidance three-dimensional model, and calculate the avoidance rate index BrsL during the underwater robot's avoidance process; Step 3: Monitor the speed and direction of the ocean current in the area where the underwater robot is located, obtain a first data set, use a deep learning algorithm to establish a deep learning calibration model, construct and evaluate the ocean current impact index YL, and if the ocean current impact index YL exceeds the first load threshold X, issue a first calibration instruction, including: slowing down the power or propulsion of the propeller by 20%, adjusting the posture of the underwater robot, and adjusting the angle of the rudder or propeller to be the same as the direction of the ocean current movement; Step 4: collect the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl of the thruster during obstacle avoidance, use the real-time vertical descent depth sdz, underwater temperature value wd and diving seawater pressure value yl as the second input value, calculate and evaluate the relative strength index Xd, and if the relative strength index Xd exceeds the deviation threshold P, issue a second calibration instruction, including: adjusting and slowing down the power or thrust of the thruster by 3%-5%, and slowing down the power or thrust of the thruster by 2% for every 10 meters decrease in the real-time vertical descent depth sdz.

Images