An underwater mechanical arm adaptive neural network control method

Abstract

The application provides an underwater mechanical arm adaptive neural network control method, belongs to the technical field of underwater intelligent robot mechanical arm control, acquires historical data and real-time data, simulates the operation state of the underwater mechanical arm through an RBF neural network, carries out prediction through model training based on the historical data, compares the real-time data with the model prediction data, judges the accuracy of the model prediction, corrects the underwater mechanical arm according to the error condition, and further generates an adaptive control command, so that the real-time operation state of the underwater mechanical arm is accurately controlled, the underwater mechanical arm can work according to a predetermined track in different water environments, the underwater mechanical arm can automatically adapt to different water environments, and the end operation precision, the adaptive capacity and the working efficiency of the underwater mechanical arm are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of underwater intelligent robot robotic arm control technology, specifically, it relates to an adaptive neural network control method for underwater robotic arms. Background Technology

[0002] The ocean covers a vast area of ​​the Earth, approximately 71% of the Earth's surface, far exceeding the land area. Marine resources are a crucial spatial and resource guarantee for a country's sustainable development. The ocean contains extremely rich resources. Currently, human exploitation of marine resources is still at a low level. As human demand for resources approaches the limits of land-based resources, the broader development of marine resources has significant and far-reaching implications for promoting sustainable development. Underwater robots, also known as remotely operated vehicles (ROVs), are diving devices that can replace humans in deep-sea operations. Robotic arms, as the preferred equipment for operational robots, are widely used in deep-sea environments.

[0003] Underwater robotic arms are mechatronic devices involving numerous disciplines such as robotics, mechanical design, electrical engineering, theoretical mechanics, mechanics of materials, and control theory. They are widely used in fields such as seabed fishing, marine environmental surveying, and underwater engineering. The level of research into underwater robotic arms largely reflects a country's scientific and technological level. Unlike land-based robotic arms, underwater robotic arms must consider the effects of gravity, buoyancy, water viscosity, and the impact of seabed currents during operation. Therefore, the design of underwater robotic arms needs to account for errors in the dynamic model caused by the unknown environment of deep water, and they need to have a certain degree of adaptability to meet the operational requirements of unknown environments. Summary of the Invention

[0004] Based on the above background, this invention conducts research on adaptive control of deep-sea robotic arms. To address the highly uncertain control problem of robotic arms facing different working environments and tasks in unknown marine environments, an adaptive neural network control method for underwater robotic arms is proposed, which improves the control accuracy and adaptability of underwater robotic arms.

[0005] This invention is achieved through the following technical solution:

[0006] An adaptive neural network control method for an underwater robotic arm: The method specifically includes the following steps:

[0007] Step 1: Acquire real-time and historical data from the underwater robotic arm and perform preprocessing:

[0008] Step 2: Construct an adaptive neural network to simulate the dynamics model of the underwater manipulator. Use the preprocessed data from Step 1 as the input to the adaptive neural network of the underwater manipulator. The underwater manipulator model simulated by the adaptive neural network will process the input data to obtain the predicted data of the underwater manipulator's operation under different states.

[0009] Step 3: Utilize historical data to establish an underwater robotic arm operational status assessment model. Compare the real-time data status assessment results of the underwater robotic arm with the operational prediction data from Step 2 to obtain the operational error data of the underwater robotic arm.

[0010] Step 4: Combine the operational error data with the historical operational data of the underwater robotic arm to form a data set to be optimized. Then, use the data set under different environments as input to the adaptive neural network to optimize the underwater robotic arm's operational status evaluation model.

[0011] Step 5: Classify the historical operation data of the underwater robot arm according to different scenarios to obtain the historical underwater robot arm operation scenario types. Process the underwater robot arm operation data according to the obtained scenario types to obtain the corrected underwater robot arm target data. Generate new underwater robot arm adaptive control commands based on the corrected underwater robot arm target data.

[0012] Further, step 1 includes:

[0013] Step 1.1: Retrieve historical data. Retrieve historical operation data and historical target operation data of the underwater robotic arm from the underwater robotic arm database.

[0014] The historical operating data of the underwater robotic arm includes: historical operating position data of the underwater robotic arm, historical operating speed data of the underwater robotic arm, and historical operating acceleration data of the underwater robotic arm.

[0015] The historical target operation data of the underwater robotic arm includes: historical target position data, historical target velocity data, historical target acceleration data, and external environmental parameters during the historical operation of the underwater robotic arm.

[0016] Step 1.2: Obtain real-time working data and real-time target data of the underwater robotic arm;

[0017] The real-time working data of the underwater robotic arm includes the position of the underwater robotic arm's end effector, the speed of the underwater robotic arm's end effector, the acceleration of the underwater robotic arm's end effector, and the disturbance torque of the water on the robotic arm.

[0018] The real-time target data of the underwater robotic arm includes the position of the underwater robotic arm's end effector and the target torque;

[0019] Step 1.3, data preprocessing: preprocess the real-time working data, real-time target data, historical operation data and historical target data of the underwater robotic arm to obtain the underwater robotic arm running time, the running angular velocity of each joint and the driving torque of each joint under different environments.

[0020] Furthermore, in step 2,

[0021] Set up approximation terms for the inherent uncertainty of the robotic arm and hydrodynamic terms, approximation weights for the inherent uncertainty of the underwater robotic arm and hydrodynamic terms, and approximation weight errors for the inherent uncertainty of the underwater robotic arm and hydrodynamic terms. Use these data to obtain the inherent uncertainty error and hydrodynamic error of the underwater robotic arm.

[0022] The dynamic model of an underwater robotic arm is simulated using the hidden layer of a neural network. Error approximation is performed on the internal modeling error of the underwater robotic arm, the joint friction during operation, and the water disturbance torque experienced by the underwater robotic arm in the flow field environment. This yields predicted data for the operation of the underwater robotic arm under different conditions.

[0023] Further, step 3 includes:

[0024] Step 3.1: Retrieve historical operating data and historical target data of the underwater robotic arm to establish an underwater robotic arm operating status evaluation model;

[0025] Step 3.2: Obtain real-time underwater robotic arm operation data and underwater robotic arm target operation data, use the evaluation model to evaluate the state of the real-time data, and obtain the real-time data state evaluation results;

[0026] Step 3.3: Compare the real-time data status assessment results of the underwater robotic arm with the operational prediction data of the underwater robotic arm under different states to obtain the operational error data of the underwater robotic arm.

[0027] Furthermore, in step 3.1,

[0028] Set the adaptive neural network weight control rate, obtain the underwater manipulator position error based on the underwater manipulator position data and underwater manipulator target data in step 2, and input the underwater manipulator position error into the underwater manipulator evaluation model. The underwater manipulator evaluation model can assign values ​​to the underwater manipulator's own uncertainty term error weight and hydrodynamic term error weight in the hidden layer of the adaptive neural network according to the underwater manipulator operation data.

[0029] The control law of the neural network is set according to the inherent uncertainty error of the underwater manipulator and the hydrodynamic term error. At the same time, a robust term is set to compensate for the approximation error of the two neural networks and external interference during the operation of the underwater manipulator. The underwater manipulator's operation data is input into the adaptive neural network, and the predicted operation data of the underwater manipulator is obtained by using the underwater manipulator dynamic model simulated by the neural network.

[0030] Furthermore, in step 4,

[0031] Based on the data of underwater robotic arm operation time, angular velocity of each joint, and driving torque of each joint under different environments, the environmental characteristics of the underwater robotic arm during operation can be analyzed. The operation error of the underwater robotic arm can be obtained by comparing the operation data of the underwater robotic arm under different environments with the target operation data of the underwater robotic arm. The operation time, angular velocity of each joint, driving torque of each joint data and historical data in the underwater robotic arm database are combined.

[0032] The combined data is grouped according to the underwater robotic arm's operating environment to obtain combinations of underwater robotic arm operating data, target data, error data, and historical operating data under different external environments, resulting in a data group to be optimized. The data groups under different environments are used as input to an adaptive neural network, and the underwater robotic arm dynamics model is optimized through the hidden layer of the neural network to obtain the optimized simulated underwater robotic arm dynamics model and the optimized underwater robotic arm operating status evaluation model.

[0033] Environmental characteristics are analyzed based on the underwater robotic arm's running time, angular velocity of each joint, and driving torque data under different environments. The running data and target data are compared to obtain the running error. The relevant data are combined and grouped according to the running environment to obtain data groups under different environments. These data groups are used as inputs to an adaptive neural network for data processing to obtain an adaptive optimization model for the underwater robotic arm under different environments. The different adaptive control models of the underwater robotic arm are combined with the corresponding input data to construct an adaptive neural network control model for the underwater robotic arm based on different operating states.

[0034] Furthermore, in step 5,

[0035] Combining different adaptive control models for underwater robotic arms with their corresponding input data can yield adaptive neural network control models for underwater robotic arms under different operating conditions.

[0036] By matching and combining underwater robotic arm operation data of different scenario types, underwater robotic arm dynamics model simulated by optimized adaptive neural network, and underwater robotic arm operation status evaluation model, we can obtain underwater robotic arm application scenario data set.

[0037] By matching real-time underwater robotic arm operation data and real-time underwater robotic arm task data with underwater robotic arm usage scenario data sets, test data sets for various scenarios of the real-time underwater robotic arm are obtained.

[0038] The data within each scenario of the real-time underwater robotic arm are analyzed using corresponding adaptive neural network simulation models of the robotic arm's dynamics and operational status. This yields predicted motion data and operational status data for each scenario of the real-time underwater robotic arm.

[0039] The real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario are compared with the real-time underwater robotic arm operation data. If the error of the comparison result is greater than the preset error range, the underwater robotic arm task data is corrected using the real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario, and the corrected underwater robotic arm task data is obtained. Based on the corrected underwater robotic arm task data, the underwater robotic arm adaptive control command is generated.

[0040] An adaptive neural network control system for an underwater robotic arm:

[0041] The control system includes: a data acquisition and preprocessing module, an operation prediction module, an operation status evaluation module, a model optimization module, and a control application module;

[0042] The data acquisition and preprocessing module acquires real-time and historical data from the underwater robotic arm and performs preprocessing:

[0043] The operation prediction module constructs an adaptive neural network to simulate the dynamic model of the underwater manipulator. The preprocessed data is used as the input of the adaptive neural network of the underwater manipulator. The underwater manipulator model simulated by the adaptive neural network will perform calculations on the input data to obtain the operation prediction data of the underwater manipulator under different states.

[0044] The operational status assessment module uses historical data to establish an operational status assessment model for the underwater robotic arm. It compares the real-time operational status assessment results with the operational prediction data to obtain the operational error data of the underwater robotic arm.

[0045] The model optimization module combines the operational error data with the historical operational data of the underwater robotic arm to form a data set to be optimized. Then, it uses the data set under different environments as input to the adaptive neural network to optimize the underwater robotic arm's operational status evaluation model.

[0046] The control application module classifies the historical operation data of the underwater robotic arm according to different scenarios to obtain the historical underwater robotic arm operation scenario types. Based on the obtained scenario types, the module processes the underwater robotic arm operation data to obtain corrected underwater robotic arm target data. Based on the corrected underwater robotic arm target data, the module generates new underwater robotic arm adaptive control commands.

[0047] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0048] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0049] Beneficial effects of the invention

[0050] The underwater robotic arm adaptive neural network control method of the present invention improves the control accuracy of the underwater robotic arm by approximating the uncertainty terms and hydrodynamic terms of the underwater robotic arm itself through the approximation function of the radial basis function neural network. By setting an underwater robotic arm evaluation model, the underwater robotic arm adaptive neural network can automatically adjust the error weights of the underwater robotic arm's own uncertainty terms and hydrodynamic term errors in the hidden layer, thereby improving the adaptive capability of the underwater robotic arm. This is of great significance for the development and utilization of marine resources. Attached Figure Description

[0051] Figure 1 This is a flowchart of the underwater robotic arm drag control method of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Unless otherwise specified, the experimental methods used in the following examples are conventional methods. Unless otherwise specified, the materials, reagents, methods, and instruments used are all conventional materials, reagents, methods, and instruments in the art, and can be obtained commercially by those skilled in the art.

[0054] First, we construct the dynamic formula for the underwater robotic arm: According to the Lagrange equation, we can obtain the dynamic equation for the four-joint robotic arm, which is formula (1):

[0055]

[0056] In the formula q, Let M(q) be the joint position, velocity, and acceleration matrix of the robotic arm; M(q) is the positive definite inertia matrix of the robotic arm. G(q) represents the centrifugal force and Coriolis force of the robotic arm; G(q) represents the gravity of the robotic arm. denoted as F(q), where F is the friction term of the robotic arm; D is the modeling error and hydrodynamic term of the robotic arm; B(q) is the buoyancy term of the robotic arm; and H is the force exerted on the robotic arm by the flow field.

[0057] For underwater robotic arms, controlling the operation solely by the deviation between the actual and desired trajectories of the joints is insufficient to guarantee good dynamic and static performance. Underwater robotic arms need to track the continuously changing motion trajectory of the moving object during operation, requiring model-based control to achieve better dynamic and static performance and meet the requirements for underwater docking.

[0058] When a robotic arm operates in an underwater environment, it is necessary to consider unstable factors such as ocean currents and waves. The effects of these factors bring uncertainty to the underwater robotic arm system. In addition, it is usually difficult to obtain complete robotic arm model information in practice. However, neural networks have the characteristic of nonlinear approximation, and a neural network adaptive controller can be established on the basis of the nominal dynamic model to identify the internal uncertainties and external unknown disturbances in the underwater robotic arm system, thereby achieving effective control of the underwater robotic arm system. The RBF neural network is a feedforward neural network based on radial basis functions. In the adaptive control based on the RBF neural network, the adaptive law of reasonably setting the weights can be used to ensure the stable convergence of the control system by adjusting the weights. The applied algorithm is expressed as formula (2):

[0059]

[0060] F = W T h+ε (2)

[0061] The nonlinear uncertainties and hydrodynamic terms in the dynamic model of the robotic arm are approximated by RBF neural networks as formula (3):

[0062]

[0063] In the formula The uncertainties are those of the arm itself and the hydrodynamic factors; ε1 and ε2 are the approximation weights for the neural network; ε1 and ε2 are the errors in the network approximation values.

[0064] The network approximation weight error is given by formula (4):

[0065]

[0066] Combining formulas (2), (3), and (4), we obtain formula (5):

[0067]

[0068] in x2 = q T

[0069] The adaptive neural network controller also needs to set the neural network control law, which includes a robust term to compensate for the approximation error of the two neural networks and external disturbances. The robust term formula is (6):

[0070] τ r =(ε N +b d )sgn(r)(6)

[0071] The adaptive neural network control rate of the underwater robotic arm is given by formula (7):

[0072]

[0073] The adaptive neural network weight control rate of the underwater robotic arm is given by formula (8):

[0074]

[0075] Where λ1 and λ2 are symmetric matrices and are positive definite.

[0076] This invention presents an adaptive neural network controller that utilizes the nonlinear mapping capability of an RBF neural network to approximate the unknown dynamics of a robotic arm, as well as the modeling errors within the underwater robotic arm and external disturbances caused by the flow environment. This allows for the design of the control law and stability analysis. The controller employs a neural network f(x) to approximate the uncertainties of the robotic arm itself and a neural network H(x) to approximate the additional hydrodynamic forces acting on the robotic arm.

[0077] After the controller design is completed, its stability needs to be analyzed. This paper uses the Lyapunov function to prove the stability based on the designed underwater robotic arm control law.

[0078] Substituting formula (7) into formula (1) yields formula (9):

[0079]

[0080] Verifying controller stability requires defining a stability function, and the Lyapunov function is defined as formula (10):

[0081]

[0082] Differentiating formula (10) yields formula (11):

[0083]

[0084] Substituting formula (9) into formula (11) yields formula (12):

[0085]

[0086] The derivative of the ideal weights in a neural network is 0. According to formula (8), the error derivative of the approximating weights can be expressed as formula (13):

[0087]

[0088] Substituting formula (13) into formula (12) yields formula (14):

[0089]

[0090] Considering the robust term τ r Represented as formula (15):

[0091] r T (τ d +ε1+ε2)-||r||(ε N +b d )sgn(r)≤0 (15)

[0092] According to formula (15), we can obtain that Since L≥0, Therefore r, There are boundaries, when When r = 0, the robotic arm system is asymptotically stable.

[0093] Based on the above analysis, this invention proposes an adaptive neural network control method for underwater robotic arms; the steps are as follows: Figure 1 As shown,

[0094] Step 1: Acquire real-time and historical data from the underwater robotic arm and perform preprocessing:

[0095] Step 1.1: Retrieve historical data. Retrieve historical operation data and historical target operation data of the underwater robotic arm from the underwater robotic arm database.

[0096] The historical operating data of the underwater robotic arm includes: historical operating position data of the underwater robotic arm, historical operating speed data of the underwater robotic arm, and historical operating acceleration data of the underwater robotic arm.

[0097] The historical target operation data of the underwater robotic arm includes: historical target position data, historical target velocity data, historical target acceleration data, and external environmental parameters during the historical operation of the underwater robotic arm.

[0098] The data is initialized, and the historical operation data and historical task data of the underwater robotic arm are identified and classified according to different scenarios to obtain the historical underwater robotic arm operation scenario types. The underwater robotic arm data is then classified according to different scenarios to form an underwater robotic arm dataset for subsequent model training.

[0099] Step 1.2: Acquire real-time operational data and real-time target data of the underwater robotic arm. The real-time operational data of the underwater robotic arm includes the end effector position, end effector velocity, end effector acceleration, and water disturbance torque on the robotic arm; the real-time target data of the underwater robotic arm includes the end effector position and target torque.

[0100] Real-time data is used to compare with the data predicted by the model to determine the accuracy of the model prediction. Based on the error, the underwater mechanical parts are corrected, and adaptive control commands are generated to achieve precise control of the real-time operating status of the underwater robotic arm.

[0101] Step 1.3, data preprocessing: preprocess the real-time working data, real-time target data, historical operation data and historical target data of the underwater robotic arm to obtain the underwater robotic arm running time, the running angular velocity of each joint and the driving torque of each joint under different environments.

[0102] Step 2, construct the adaptive neural network model:

[0103] The data preprocessed in step 1 is used as input to the underwater robotic arm's adaptive neural network. The underwater robotic arm model simulated by the adaptive neural network will perform calculations on the input data.

[0104] To input the underwater robotic arm dataset, including the underwater robotic arm end position, underwater robotic arm end velocity, and underwater robotic arm end acceleration, we input the data into formulas (5), (7), and (8). After calculation, we obtain the control rate and control rate error of each part, which are then input into formula (1) to obtain the underwater robotic arm's operating position data, operating speed position, and operating acceleration position under different environments. These data are used as the underwater robotic arm's adaptive neural network evaluation data.

[0105] We set up approximation terms for the inherent uncertainty of the underwater robot arm and hydrodynamic terms, approximation weights for the inherent uncertainty of the underwater robot arm and hydrodynamic terms, and approximation weight errors for the inherent uncertainty of the underwater robot arm and hydrodynamic terms. We then obtain the inherent uncertainty error and hydrodynamic error of the underwater robot arm from these data.

[0106] The dynamic model of an underwater robotic arm is simulated using the hidden layer of a neural network. Error approximation is performed on the internal modeling error of the underwater robotic arm, the joint friction during operation, and the water disturbance torque experienced by the underwater robotic arm in the flow field environment. This yields predicted data for the operation of the underwater robotic arm under different conditions.

[0107] Step 3: Establish an evaluation model and conduct error analysis.

[0108] Establish an evaluation model: retrieve historical operation data and historical target data of the underwater robotic arm to establish an evaluation model for the operating status of the underwater robotic arm.

[0109] Set the adaptive neural network weight control rate, obtain the underwater manipulator position error based on the underwater manipulator position data and underwater manipulator target data from step 2, and input the underwater manipulator position error into the underwater manipulator evaluation model. The underwater manipulator evaluation model can assign values ​​to the underwater manipulator's own uncertainty term error weight and hydrodynamic term error weight in the hidden layer of the adaptive neural network according to the underwater manipulator operation data.

[0110] The control law of the neural network is set based on the inherent uncertainty error of the underwater manipulator and the hydrodynamic error. A robust term is also included to compensate for the approximation errors of the two neural networks and external disturbances during the operation of the underwater manipulator. The underwater manipulator's operating data is input into the adaptive neural network, and the predicted operating data of the underwater manipulator is obtained using the dynamic model of the underwater manipulator simulated by the neural network.

[0111] Acquire and evaluate real-time data: Acquire real-time underwater robotic arm operation data and underwater robotic arm target operation data, use an evaluation model to evaluate the state of the real-time data, and obtain the real-time data state evaluation results.

[0112] Error analysis: The real-time data status assessment results of the underwater robotic arm are compared with the predicted operation data of the underwater robotic arm under different states to obtain the operation error data of the underwater robotic arm.

[0113] Step 4: Optimize the underwater robotic arm's operational status evaluation model:

[0114] Based on the obtained data on the underwater robotic arm's operating time, angular velocity of each joint, and driving torque of each joint under different environments, the environmental characteristics of the underwater robotic arm during operation can be analyzed. The operating error of the underwater robotic arm can be obtained by comparing the operating data of the underwater robotic arm under different environments with the target operating data. The operating time, angular velocity of each joint, and driving torque data of each joint are then combined with historical data from the underwater robotic arm database.

[0115] The combined data is grouped according to the underwater robotic arm's operating environment, resulting in combinations of underwater robotic arm operating data, target data, error data, and historical operating data under different external environments, thus obtaining the data set to be optimized. This data set from different environments is then used as input to an adaptive neural network. The network's hidden layers simulate the underwater robotic arm's dynamics model for optimization, yielding an optimized simulated underwater robotic arm dynamics model and an optimized underwater robotic arm operating status evaluation model.

[0116] Environmental characteristic analysis was conducted based on the underwater robotic arm's operating time, angular velocity of each joint, and driving torque data under different environments. The operating error was obtained by comparing the operating data with the target data. Relevant data were combined and grouped according to the operating environment, resulting in data groups for different environments. These data groups were then used as input to an adaptive neural network for data processing, yielding adaptive optimization models for the underwater robotic arm under different environments. These different adaptive control models were then combined with their corresponding input data to construct an adaptive neural network control model for the underwater robotic arm based on different operating states.

[0117] Step 5, Model Application and Control Command Generation:

[0118] By combining different adaptive control models for underwater robotic arms with their corresponding input data, we can obtain adaptive neural network control models for underwater robotic arms based on different operating conditions.

[0119] Identify the historical operation data and historical task data of the underwater robotic arm to obtain the types of historical underwater robotic arm operation scenarios;

[0120] By matching and combining underwater robotic arm operation data of different scenario types, underwater robotic arm dynamics model simulated by optimized adaptive neural network, and underwater robotic arm operation status evaluation model, we can obtain underwater robotic arm application scenario data set.

[0121] By matching real-time underwater robotic arm operation data and real-time underwater robotic arm task data with underwater robotic arm usage scenario data sets, test data sets for each scenario of the real-time underwater robotic arm are obtained.

[0122] The data within each scenario of the real-time underwater robotic arm are analyzed using corresponding adaptive neural network simulation models of the robotic arm's dynamics and operational status. This yields predicted motion data and operational status data for each scenario of the real-time underwater robotic arm.

[0123] The real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario are compared with the real-time underwater robotic arm operation data. If the error of the comparison result is greater than the preset error range, the underwater robotic arm task data is corrected using the real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario, and the corrected underwater robotic arm task data is obtained. Based on the corrected underwater robotic arm task data, the underwater robotic arm adaptive control command is generated.

[0124] An adaptive neural network control system for an underwater robotic arm:

[0125] The control system includes: a data acquisition and preprocessing module, an operation prediction module, an operation status evaluation module, a model optimization module, and a control application module;

[0126] The data acquisition and preprocessing module acquires real-time and historical data from the underwater robotic arm and performs preprocessing:

[0127] The operation prediction module constructs an adaptive neural network to simulate the dynamic model of the underwater manipulator. The preprocessed data is used as the input of the adaptive neural network of the underwater manipulator. The underwater manipulator model simulated by the adaptive neural network will perform calculations on the input data to obtain the operation prediction data of the underwater manipulator under different states.

[0128] The operational status assessment module uses historical data to establish an operational status assessment model for the underwater robotic arm. It compares the real-time operational status assessment results with the operational prediction data to obtain the operational error data of the underwater robotic arm.

[0129] The model optimization module combines the operational error data with the historical operational data of the underwater robotic arm to form a data set to be optimized. Then, it uses the data set under different environments as input to the adaptive neural network to optimize the underwater robotic arm's operational status evaluation model.

[0130] The control application module classifies the historical operation data of the underwater robotic arm according to different scenarios to obtain the historical underwater robotic arm operation scenario types. Based on the obtained scenario types, the module processes the underwater robotic arm operation data to obtain corrected underwater robotic arm target data. Based on the corrected underwater robotic arm target data, the module generates new underwater robotic arm adaptive control commands.

[0131] An electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above method.

[0132] A computer-readable storage medium for storing computer instructions that, when executed by a processor, implement the steps of the above-described method.

[0133] The memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory of the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0134] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means such as coaxial cable, optical fiber, digital subscriber line, DSL, or wireless means such as infrared, wireless, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium such as a floppy disk, hard disk, magnetic tape; an optical medium such as a high-density digital video disc, DVD; or a semiconductor medium such as a solid-state disk, SSD, etc.

[0135] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0136] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as execution by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0137] The above provides a detailed description of the adaptive neural network control method for underwater robotic arms proposed in this invention, and elucidates the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An adaptive neural network control method for an underwater robotic arm, characterized in that: The method specifically includes the following steps: Step 1: Acquire real-time and historical data from the underwater robotic arm and perform preprocessing: Step 2: Construct an adaptive neural network to simulate the dynamics model of the underwater manipulator. Use the preprocessed data from Step 1 as the input to the adaptive neural network of the underwater manipulator. The underwater manipulator model simulated by the adaptive neural network will process the input data to obtain the predicted data of the underwater manipulator's operation under different states. In step 2, Set up approximation terms for the inherent uncertainty of the robotic arm and hydrodynamic terms, approximation weights for the inherent uncertainty of the underwater robotic arm and hydrodynamic terms, and approximation weight errors for the inherent uncertainty of the underwater robotic arm and hydrodynamic terms. Use these data to obtain the inherent uncertainty error and hydrodynamic error of the underwater robotic arm. The dynamic model of an underwater robotic arm is simulated using the hidden layer of a neural network. Error approximation is performed on the internal modeling error of the underwater robotic arm, the joint friction during operation, and the water disturbance torque experienced by the underwater robotic arm in the flow field environment. The operation prediction data of the underwater robotic arm under different states are obtained. Step 3: Utilize historical data to establish an underwater robotic arm operational status assessment model. Compare the real-time data status assessment results of the underwater robotic arm with the operational prediction data from Step 2 to obtain the operational error data of the underwater robotic arm. Step 3 includes: Step 3.1: Retrieve historical operating data and historical target data of the underwater robotic arm to establish an underwater robotic arm operating status evaluation model; Step 3.2: Obtain real-time underwater robotic arm operation data and underwater robotic arm target operation data, use the evaluation model to evaluate the state of the real-time data, and obtain the real-time data state evaluation results; Step 3.3: Compare the real-time data status assessment results of the underwater robotic arm with the operation prediction data of the underwater robotic arm under different states to obtain the operation error data of the underwater robotic arm. In step 3.1, Set the adaptive neural network weight control rate, obtain the underwater manipulator position error based on the underwater manipulator position data and underwater manipulator target data in step 2, substitute the underwater manipulator position error into the underwater manipulator evaluation model, and the underwater manipulator evaluation model can assign values ​​to the underwater manipulator's own uncertainty term error weight and hydrodynamic term error weight in the hidden layer of the adaptive neural network according to the underwater manipulator operation data. The control law of the neural network is set according to the uncertainty error of the underwater manipulator itself and the hydrodynamic term error. At the same time, a robust term is set to compensate for the approximation error of the two neural networks and external interference when the underwater manipulator is working. The underwater manipulator's operation data is input into the adaptive neural network, and the predicted operation data of the underwater manipulator is obtained by using the underwater manipulator dynamic model simulated by the neural network. Step 4: Combine the operational error data with the historical operational data of the underwater robotic arm to form a data set to be optimized. Then, use the data set under different environments as input to the adaptive neural network to optimize the underwater robotic arm's operational status evaluation model. Step 5: Classify the historical operation data of the underwater robot arm according to different scenarios to obtain the historical underwater robot arm operation scenario types. Process the underwater robot arm operation data according to the obtained scenario types to obtain the corrected underwater robot arm target data. Generate new underwater robot arm adaptive control commands based on the corrected underwater robot arm target data.

2. The control method according to claim 1, characterized in that: Step 1 includes: Step 1.1: Retrieve historical data. Retrieve historical operation data and historical target operation data of the underwater robotic arm from the underwater robotic arm database. The historical operating data of the underwater robotic arm includes: historical operating position data of the underwater robotic arm, historical operating speed data of the underwater robotic arm, and historical operating acceleration data of the underwater robotic arm. The historical target operation data of the underwater robotic arm includes: historical target position data, historical target velocity data, historical target acceleration data, and external environmental parameters during the historical operation of the underwater robotic arm. Step 1.2: Obtain real-time working data and real-time target data of the underwater robotic arm; The real-time working data of the underwater robotic arm includes the position of the underwater robotic arm's end effector, the speed of the underwater robotic arm's end effector, the acceleration of the underwater robotic arm's end effector, and the disturbance torque of the water on the robotic arm. The real-time target data of the underwater robotic arm includes the position of the underwater robotic arm's end effector and the target torque; Step 1.3, data preprocessing: preprocess the real-time working data, real-time target data, historical operation data and historical target data of the underwater robotic arm to obtain the underwater robotic arm running time, the running angular velocity of each joint and the driving torque of each joint under different environments.

3. The control method according to claim 2, characterized in that: In step 4, Based on the data of underwater robotic arm operation time, angular velocity of each joint, and driving torque of each joint under different environments, the environmental characteristics of the underwater robotic arm during operation can be analyzed. The operation error of the underwater robotic arm can be obtained by comparing the operation data of the underwater robotic arm under different environments with the target operation data of the underwater robotic arm. The operation time, angular velocity of each joint, driving torque of each joint data and historical data in the underwater robotic arm database are combined. The combined data is grouped according to the underwater robotic arm's operating environment to obtain combinations of underwater robotic arm operating data, target data, error data, and historical operating data under different external environments, resulting in a data group to be optimized. The data groups under different environments are used as input to an adaptive neural network, and the underwater robotic arm dynamics model is optimized through the hidden layer of the neural network to obtain the optimized simulated underwater robotic arm dynamics model and the optimized underwater robotic arm operating status evaluation model. Environmental characteristics are analyzed based on the underwater robotic arm's running time, angular velocity of each joint, and driving torque data under different environments. The running data and target data are compared to obtain the running error. The relevant data are combined and grouped according to the running environment to obtain data groups under different environments. These data groups are used as inputs to an adaptive neural network for data processing to obtain an adaptive optimization model for the underwater robotic arm under different environments. The different adaptive control models of the underwater robotic arm are combined with the corresponding input data to construct an adaptive neural network control model for the underwater robotic arm based on different operating states.

4. The control method according to claim 3, characterized in that: In step 5, Combining different adaptive control models for underwater robotic arms with their corresponding input data can yield adaptive neural network control models for underwater robotic arms under different operating conditions. By matching and combining underwater robotic arm operation data of different scenario types, underwater robotic arm dynamics model simulated by optimized adaptive neural network, and underwater robotic arm operation status evaluation model, we can obtain underwater robotic arm application scenario data set. By matching real-time underwater robotic arm operation data and real-time underwater robotic arm task data with underwater robotic arm usage scenario data sets, test data sets for various scenarios of the real-time underwater robotic arm are obtained. The data within each scenario of the real-time underwater robotic arm are analyzed using corresponding adaptive neural network simulation models of the robotic arm's dynamics and operational status. This yields predicted motion data and operational status data for each scenario of the real-time underwater robotic arm. The real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario are compared with the real-time underwater robotic arm operation data. If the error of the comparison result is greater than the preset error range, the underwater robotic arm task data is corrected using the real-time underwater robotic arm motion prediction data and real-time underwater robotic arm operation status prediction data for each scenario, and the corrected underwater robotic arm task data is obtained. Based on the corrected underwater robotic arm task data, the underwater robotic arm adaptive control command is generated.

5. A control system for executing the adaptive neural network control method for an underwater robotic arm according to any one of claims 1 to 4, characterized in that: The control system includes: a data acquisition and preprocessing module, an operation prediction module, an operation status evaluation module, a model optimization module, and a control application module; The data acquisition and preprocessing module acquires real-time and historical data from the underwater robotic arm and performs preprocessing: The operation prediction module constructs an adaptive neural network to simulate the dynamic model of the underwater manipulator. The preprocessed data is used as the input of the adaptive neural network of the underwater manipulator. The underwater manipulator model simulated by the adaptive neural network will perform calculations on the input data to obtain the operation prediction data of the underwater manipulator under different states. The operational status assessment module uses historical data to establish an operational status assessment model for the underwater robotic arm. It compares the real-time operational status assessment results with the operational prediction data to obtain the operational error data of the underwater robotic arm. The model optimization module combines the operational error data with the historical operational data of the underwater robotic arm to form a data set to be optimized. Then, it uses the data set under different environments as input to the adaptive neural network to optimize the underwater robotic arm's operational status evaluation model. The control application module classifies the historical operation data of the underwater robotic arm according to different scenarios to obtain the historical underwater robotic arm operation scenario types. Based on the obtained scenario types, the module processes the underwater robotic arm operation data to obtain corrected underwater robotic arm target data. Based on the corrected underwater robotic arm target data, the module generates new underwater robotic arm adaptive control commands.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 4.

Images