Underwater propeller control method based on multi-parameter fusion and related equipment

Abstract

According to the underwater propeller control method based on multi-parameter fusion and the related equipment, the autonomous decision-making ability and adaptability of the underwater propeller in different environments are improved, more intelligent and more efficient control is achieved, and therefore the overall stability of the underwater propeller can be improved. The method comprises the steps that the water flow velocity and regional terrain information of a target region where the underwater propeller is located are collected; a three-dimensional flow field real-time sensing model is established based on the water flow velocity and the regional topographic information, and multi-modal data are acquired according to the three-dimensional flow field real-time sensing model and comprise the water flow velocity, the water flow pressure, the water flow intensity, the three-dimensional topographic coordinates and the topographic height of a target region; carrying out dispersion filtering on the multi-modal data and then fusing the multi-modal data to obtain target multi-modal data; and constructing a multi-modal neural network model through a time sequence attention mechanism based on the target multi-modal data.

Description

Technical Field

[0001] The present application relates to the technical field of underwater thrusters, and particularly to an underwater thruster control method and related devices based on multi-parameter fusion. Background Art

[0002] An underwater thruster is a device that provides power for an underwater vehicle to move forward and turn, etc. It is a core component for underwater operations and navigation, and is widely used in fields such as ocean exploration, underwater operations, and submarine navigation, and is of great significance in aspects such as ocean resource development and environmental monitoring.

[0003] During the actual operation of an underwater thruster, the underwater environment is complex, and the water flow velocity, direction, and density are variable. Therefore, when the underwater thruster is working, it is necessary to continuously collect underwater environment information to continuously adjust the control strategy according to the collected environment information.

[0004] However, existing traditional underwater thrusters generally use single data or simple empirical strategies to control the underwater thruster to work. Due to the complex underwater environment, single data cannot comprehensively reflect the coupled factors such as water flow, pressure, and terrain in the area where the underwater thruster is located. Therefore, the real environment of the area where the underwater thruster is located cannot be comprehensively obtained, resulting in the inability to control the underwater thruster in real time according to the environment data. As a result, the situation where the underwater thruster collides with obstacles due to inaccurate power output or recognition may occur, and the overall stability is relatively low. Summary of the Invention

[0005] The present application provides an underwater thruster control method and related devices based on multi-parameter fusion to solve the above technical problems.

[0006] The first aspect of the present application provides an underwater thruster control method based on multi-parameter fusion, and the method includes: Collect the water flow velocity and regional terrain information of the target area where the underwater thruster is located; Based on the water flow velocity and the regional terrain information, establish a three-dimensional flow field real-time perception model, and obtain multi-modal data according to the three-dimensional flow field real-time perception model. The multi-modal data includes the water flow velocity, water flow pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of the target area; Disperse and filter the multi-modal data and then fuse it to obtain target multi-modal data; Based on the target multi-modal data, construct a multi-modal neural network model through a temporal attention mechanism; Input the real-time multi-modal data into the multi-modal neural network model, and generate a control instruction by using a double-layer reinforcement learning strategy; Generate the power parameters of the underwater thruster according to the control instruction; When it is determined that the power parameter is within the preset parameter range, the underwater thruster is controlled to move according to the power parameter.

[0007] Optionally, after performing decentralized filtering on the multimodal data and then fusing it to obtain target multimodal data, including: Preprocess the multimodal data to obtain preprocessed data; Extract the water flow data and terrain data from the preprocessed data; According to the data characteristics of the water flow data and terrain data, respectively use the Kalman filtering algorithm and the Gaussian filtering algorithm to remove outliers; Perform data registration on the filtered water flow data and terrain data; Use the weighted average algorithm to perform data fusion on the data after data registration; The weighted average algorithm is:

[0008] where S is the weighted average value, n is the data point, x i is the value of the i-th data point, and w i is the weight of the i-th data point; Verify the data after contrast fusion. When it is determined that the data verification passes, then perform data smoothing and output the target multimodal data.

[0009] Optionally, based on the target multimodal data, construct a multimodal neural network model through a temporal attention mechanism, including: Extract the numerical, spatial, and temporal features from the target multimodal data and convert them into a unified dimension vector; Arrange the unified dimension vectors in a time series to obtain a feature matrix; According to the data characteristics of the feature matrix and use the temporal attention mechanism to construct a multimodal neural network model.

[0010] Optionally, input the real-time multimodal data into the multimodal neural network model and use a two-layer reinforcement learning strategy to generate control instructions, including: Obtain real-time multimodal data; Preprocess the real-time multimodal data to form an input feature vector adapted to the multimodal neural network model; Input the input feature vector into the multimodal neural network model to obtain a target feature vector; Build a double - layer reinforcement learning policy network, where the double - layer reinforcement learning policy network includes a high - level decision - making network and a low - level execution network. The high - level decision - making network is used to receive the target feature vector output by the multi - modal neural network model and plan a control strategy, and the low - level execution network is used to refine the control strategy into control parameters; Input the target feature vector into the double - layer reinforcement learning policy network, and use the Actor - Critic algorithm to update the policy network parameters; Generate a control instruction based on the policy network parameters and combined with the current environmental state.

[0011] Optionally, generate the power parameters of the underwater thruster according to the control instruction, including: Calculate the output power of the underwater thruster according to the control instruction; Compare the output power of the underwater thruster with the maximum output power of the underwater thruster, and obtain the comparison result; Generate the power parameters of the underwater thruster according to the comparison result.

[0012] Optionally, when it is determined that the power parameters are within a preset parameter range, control the underwater thruster to move according to the power parameters, including: Parse the power parameters and verify whether the power parameters are complete; If so, compare the parsed power parameters with the preset parameter range; When it is determined that the power parameters are within the preset parameter range, convert the power parameters into control signals executable by the underwater thruster; Control the underwater thruster to move according to the control signal. Optionally, after comparing the parsed power parameters with the preset parameter range, the method further includes: When it is determined that the power parameters are not within the preset parameter range, trigger an alarm instruction; Generate an execution instruction according to the alarm instruction; Retrieve the maximum value parameter within the preset parameter range according to the execution instruction; Control the underwater thruster to move according to the maximum value parameter. The second aspect of this application provides a control device for an underwater thruster based on multi - parameter fusion. The device includes: An acquisition unit, configured to acquire the water flow velocity and regional terrain information of the target area where the underwater thruster is located; A first acquisition unit, configured to establish a real-time three-dimensional flow field perception model based on the water flow velocity and the regional terrain information, and obtain multimodal data according to the real-time three-dimensional flow field perception model, where the multimodal data includes the water flow velocity, water flow pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of a target area; A second acquisition unit, configured to perform decentralized filtering and fusion on the multimodal data to obtain target multimodal data; A construction unit, configured to construct a multimodal neural network model based on the target multimodal data through a temporal attention mechanism; A first generation unit, configured to input the real-time multimodal data of the underwater thruster into the multimodal neural network model, and generate a control instruction by adopting a double-layer reinforcement learning strategy; A second generation unit, configured to generate power parameters of the underwater thruster according to the control instruction; A control unit, configured to control the movement of the underwater thruster according to the power parameters when it is determined that the power parameters are within a preset parameter range.

[0013] A third aspect of the present application provides an underwater thruster control device based on multi-parameter fusion, where the device includes: A processor, a storage, an input / output unit, and a bus; The processor is connected to the storage, the input / output unit, and the bus; The storage stores a program, and the processor calls the program to execute the method of the first aspect and any optional method in the first aspect.

[0014] A fourth aspect of the present application provides a computer-readable storage medium, where a program is stored on the computer-readable storage medium, and when the program is executed on a computer, it executes the method of the first aspect and any optional method in the first aspect.

[0015] As can be seen from the above technical solutions, the present application has the following advantages: The following are the technical effects that the technical solution can achieve: 1. By collecting the water flow velocity and regional terrain information of the target area and establishing a real-time three-dimensional flow field perception model, the present application can more comprehensively and accurately perceive the underwater environment through the real-time three-dimensional flow field perception model, provide a more reliable data basis for subsequent control, and help the underwater thruster better adapt to the complex and changeable underwater environment.

[0016] 2. The multi-modal data obtained from the three-dimensional flow field real-time perception model, including water flow velocity, water flow pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height, cover multiple key parameters of the underwater environment, enriching the data dimension and enabling the control method of this application to comprehensively consider more factors, thereby improving the accuracy and adaptability of control.

[0017] 3. After dispersive filtering and fusion of the multi-modal data in this application, noise and interference can be effectively removed, the quality of the target multi-modal data can be improved, and the subsequent constructed model and generated control instructions can be ensured to be more accurate and reliable, so as to better control the movement of the underwater thruster.

[0018] 4. Based on the target multi-modal data, a multi-modal neural network model is constructed through a temporal attention mechanism, which can better capture the temporal features and important information in the data, improve the adaptability and prediction accuracy of the multi-modal neural network model for the underwater thruster control task, and thus optimize the control effect.

[0019] 5. This application adopts a double-layer reinforcement learning strategy to generate control instructions, enabling the control method to continuously learn and optimize the control strategy according to real-time multi-modal data, improving the autonomous decision-making ability and adaptability of the underwater thruster in different environments, achieving greater intelligence, and generating the power parameters of the underwater thruster according to the control instructions. Only when the determined power parameters are within the preset parameter range, the thruster is controlled to operate, which can effectively avoid thruster failures or damages caused by abnormal parameters and ensure the operation safety and stability of the underwater thruster. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in this application, the following will briefly introduce the drawings required for use in the description of the embodiments. Obviously, the following described drawings are only some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It is a schematic flowchart of an embodiment of the underwater thruster control method based on multi-parameter fusion in this application; Figure 2 It is another schematic diagram of an embodiment of the underwater thruster control method based on multi-parameter fusion in this application; Figure 3 It is another schematic diagram of an embodiment of the underwater thruster control method based on multi-parameter fusion in this application; Figure 4 It is another schematic diagram of an embodiment of the underwater thruster control method based on multi-parameter fusion in this application; Figure 5 It is another schematic diagram of an embodiment of the underwater thruster control method based on multi-parameter fusion in this application; Figure 6 Schematic diagram of another embodiment of the underwater thruster control method based on multi-parameter fusion of the present application; Figure 7 Schematic diagram of another embodiment of the underwater thruster control method based on multi-parameter fusion of the present application; Figure 8 Schematic diagram of an embodiment of the underwater thruster control device based on multi-parameter fusion of the present application; Figure 9 Schematic diagram of another embodiment of the underwater thruster control device based on multi-parameter fusion of the present application. Detailed implementation manners

[0022] The present application provides an underwater thruster control method and related devices based on multi-parameter fusion, which improve the autonomous decision-making ability and adaptability of the underwater thruster in different environments, realize more intelligent and efficient control, and thus contribute to improving the propulsion efficiency of the underwater thruster.

[0023] Please refer to Figure 1 , an embodiment of the underwater thruster control method based on multi-parameter fusion provided by the first aspect of the present application includes: 101. Collect the water flow velocity and regional terrain information of the target area where the underwater thruster is located; 102. Establish a three-dimensional flow field real-time perception model based on the water flow velocity and the regional terrain information, and obtain multi-modal data according to the three-dimensional flow field real-time perception model, where the multi-modal data includes the water flow velocity, water pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of the target area; 103. Perform decentralized filtering and fusion on the multi-modal data to obtain target multi-modal data; 104. Based on the target multi-modal data, construct a multi-modal neural network model through a temporal attention mechanism; 105. Input the real-time multi-modal data of the underwater thruster into the multi-modal neural network model, and generate a control instruction by adopting a double-layer reinforcement learning strategy; 106. Generate the power parameters of the underwater thruster according to the control instruction; 107. When it is determined that the power parameters are within a preset parameter range, control the underwater thruster to move according to the power parameters.

[0024] In the embodiment of the present application, first, the water flow velocity and regional terrain information of the target area where the underwater thruster is located are collected. Then, a real-time perception model of the three-dimensional flow field is established based on the water flow velocity and regional terrain information, and multi-modal data is obtained according to the real-time perception model of the three-dimensional flow field. The multi-modal data includes the water flow velocity, water pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of the target area. After that, the multi-modal data is dispersed filtered and then fused to obtain the target multi-modal data. Then, based on the target multi-modal data, a multi-modal neural network model is constructed through a temporal attention mechanism. Further, the real-time multi-modal data is input into the multi-modal neural network model, and a double-layer reinforcement learning strategy is used to generate control instructions. Then, the dynamic parameters of the underwater thruster are generated according to the control instructions. When it is determined that the dynamic parameters are within the preset parameter range, the underwater thruster is controlled to move according to the dynamic parameters.

[0025] In step 101, the water flow velocity and regional terrain information of the target area where the underwater thruster is located are collected. Specifically, the control accuracy of the underwater thruster highly depends on the accuracy of environmental data. Therefore, it is necessary to collect the environmental data of the area where the underwater thruster is located, and mainly the water flow velocity and regional terrain information are collected.

[0026] Among them, the water flow velocity information can be collected by an ADCP (Acoustic Doppler Current Profiler). The ADCP uses the Doppler effect to emit 4 - 6 beams of sound waves into the water body, and obtains the three-dimensional water flow velocity at different depths and angles by analyzing the frequency offset of the echo reflected by the suspended particles in the water body, providing a real-time water flow velocity reference for the underwater thruster. It can also be collected by the method of electromagnetic current meter and data acquisition frequency, which is not specifically limited here.

[0027] For the collection of regional terrain information, a multi-beam echo sounder is needed. The multi-beam echo sounder measures the water depth data at multiple angles by emitting multiple beam sound waves, thereby constructing a high-precision three-dimensional terrain model. The generated terrain undulation information can be used for the path planning of the thruster to avoid colliding with obstacles. For example, in an area with underwater reefs on the seabed, the multi-beam data can clearly show the position and height of the reefs, and plan a detour route for the thruster. By emitting sound waves to the side and receiving the backscattered signals from the seabed and objects to generate images, it complements the multi-beam data and improves the terrain details. Then, Gaussian filtering is used to remove the noise of the terrain data, and Kriging interpolation is used to encrypt the data to improve the accuracy of the terrain model. The processed terrain data will be an important input for the real-time perception model of the three-dimensional flow field.

[0028] Subsequently, with the help of GPS or high-precision clock devices, a unified time reference is provided for the acquisition devices through the Network Time Protocol to ensure the consistency of water flow velocity and terrain information in the time dimension, so that the established real-time three-dimensional flow field model can truly reflect the environmental state at the same moment. After obtaining the water flow velocity and terrain information of the area, step 102 is executed.

[0029] In step 102, a real-time three-dimensional flow field perception model is established based on the water flow velocity and terrain information of the area, and multimodal data is obtained according to the real-time three-dimensional flow field perception model. Specifically, establishing the real-time three-dimensional flow field perception model is a process of integrating discrete water flow velocity and terrain information into a continuous and dynamic three-dimensional space model.

[0030] First, the collected data needs to be preprocessed, including removing noise, interpolating and filling missing data, etc. For water flow velocity data, there may be some outliers caused by sensor failures or environmental interference, which can be removed by setting thresholds or using statistical methods. For the missing parts in the terrain data, Kriging interpolation or inverse distance weighted interpolation can be used for filling to ensure the integrity of the data.

[0031] After that, mathematical algorithms are used to fuse the processed data into a three-dimensional space. The commonly used one is the computational fluid dynamics algorithm, which simulates the movement and distribution of water flow by solving the Navier-Stokes equations of fluid mechanics. It should be noted that when establishing the model, factors such as boundary conditions and initial conditions of the target area need to be considered. For example, the boundary of the target area is set as a fixed or movable wall surface, the initial water flow velocity and pressure distribution are set according to the actual water flow situation, and then through continuous iterative calculations, a real-time three-dimensional flow field perception model that can reflect the water flow state and terrain characteristics of the target area is obtained.

[0032] Based on the established real-time three-dimensional flow field perception model, multimodal data can be further obtained. In addition to the initially collected water flow velocity, parameters such as water flow pressure and water flow intensity can also be calculated through the model. The water flow pressure can be calculated by fluid mechanics formulas and determined according to the water flow velocity, density and position. The water flow intensity can be measured according to the magnitude and direction changes of the water flow velocity.

[0033] The terrain information of the area, including three-dimensional terrain coordinates and terrain height, is already included in the real-time three-dimensional flow field perception model. The environmental characteristics of the target area can be described through multimodal data. For example, water flow velocity and pressure can reflect the dynamic characteristics of water flow, and terrain coordinates and height can describe the geometric shape of the bottom of the water. These data information provides a more comprehensive basis for subsequent data processing and model construction.

[0034] In step 103, the multi-modal data is dispersed filtered and then fused to obtain the target multi-modal data. Specifically, after the multi-modal data is obtained through the three-dimensional flow field real-time perception model, since it may be affected by various noises and interferences during the acquisition process, directly using these raw data may lead to inaccuracy and instability of the subsequent model. Therefore, it is necessary to perform dispersed filtering on the multi-modal data.

[0035] For the water flow velocity data, the Kalman filter is commonly used for processing. The Kalman filter estimates and updates the system state in a recursive manner by establishing the state equation and observation equation of the system. In the water flow velocity filtering, the water flow velocity is regarded as the state variable of the system, and the sensor measurement value is used as the observation value. By continuously iterating the calculation, the noise in the water flow velocity data can be effectively removed, and the smoothness and accuracy of the data can be improved.

[0036] For the terrain height data, the median filter algorithm can be used for processing. The median filter is a non-linear filtering method. By sorting all the data values within the data window and taking the median value as the filtered output value, this method can effectively remove the outliers and impulse noises in the terrain height data and retain the main features of the data. For data such as water flow pressure and water flow intensity, appropriate filtering algorithms can also be selected according to their characteristics, such as exponential smoothing filtering, low-pass filtering, etc., which are not specifically limited here.

[0037] After the filtering process is completed, the data of different modalities are fused. The purpose of data fusion is to integrate the information of different modalities to obtain a more comprehensive and accurate description of the target area environment. In the data fusion process, the weighted average method or the Bayesian estimation method can be used for processing, which is not specifically limited here. The target multi-modal data obtained after fusion will be used as the basis for constructing the multi-modal neural network model in the subsequent steps. The target multi-modal data combines the advantages of different modality data and can more accurately reflect the environmental state of the target area.

[0038] In step 104, based on the target multi-modal data, a multi-modal neural network model is constructed through the temporal attention mechanism. Specifically, before inputting the target multi-modal data into the neural network model, it is necessary to perform normalization processing on the target multi-modal data to unify the numerical ranges of the data of different modalities in the target multi-modal data into the same interval. For example, the data is normalized to between [0, 1] or [-1, 1]. After processing, it can be avoided that the data of different modalities have too large differences in numerical ranges and affect the model training.

[0039] After that, feature extraction is performed on the target multi-modal data. Among them, for continuous data such as water flow velocity, pressure, and intensity, the Fourier transform method is used to extract its frequency domain features or time-frequency features. For spatial data such as terrain coordinates and height, the principal component analysis method is used to extract its main features. Through feature extraction, the dimension of the data can be reduced, and the training efficiency and generalization ability of the model can be improved.

[0040] Construct a multi-modal neural network model. In the multi-modal neural network model, there are a convolutional neural network (CNN) sub-network and a recurrent neural network (RNN) sub-network. Among them, for data related to water flow such as water flow velocity, pressure, and intensity, a convolutional neural network (CNN) sub-network will be constructed. The CNN has strong feature extraction capabilities and can automatically learn local features and spatial structures in the data. Through multi-layer convolution and pooling operations, the original data can be converted into a high-level feature representation.

[0041] For spatial data such as terrain coordinates and height, a recurrent neural network (RNN) sub-network will be constructed. The RNN can process data with temporal relationships and can capture long-term dependencies in the data. When processing terrain data, the terrain coordinates can be regarded as time series data, and the RNN can be used to learn the changing rules of terrain features.

[0042] In the control scenario of an underwater thruster, the influence degrees of environmental information at different times on the current control decision are different. Therefore, introducing a temporal attention mechanism can automatically focus on important temporal segments in the data, enabling the model to pay more attention to information useful for the current decision. The temporal attention mechanism includes an attention calculation layer and a weight assignment layer. In the attention calculation layer, the attention scores are calculated based on the query vector at the current time and the key vectors at different times. The query vector can be regarded as the focus of the model on information at the current time, and the key vectors contain the feature information of data at different times. By calculating the similarity between the query vector and the key vectors, the attention scores are obtained.

[0043] In the weight assignment layer, the data at different times are weighted and summed according to the attention scores to obtain a weighted feature representation. In this way, when the model processes data, it will automatically give higher weights to data at important times, thereby improving the decision-making ability of the model.

[0044] After that, the processed target multi-modal data is divided into a training set, a validation set, and a test set. The training set is used to train the multi-modal neural network model, and the parameters of the model are continuously adjusted through the backpropagation algorithm to make the output of the model as close as possible to the true value.

[0045] During the training process, a cross-entropy loss function or mean squared error loss function, etc. is used to measure the prediction error of the model. At the same time, a validation set is used to validate the model, and the hyperparameters of the model are adjusted according to the performance metrics on the validation set to prevent the model from overfitting or underfitting. After multiple iterations of training and optimization, a multi-modal neural network model can be obtained, and this multi-modal neural network model can learn the mapping relationship from the target multi-modal data to the control decision.

[0046] In step 105, the real-time multi-modal data of the underwater thruster is input into the multi-modal neural network model, and a double-layer reinforcement learning strategy is used to generate control instructions; specifically, during the actual operation process, the underwater thruster continuously collects real-time multi-modal data, and these data include information such as the current water flow speed, pressure, intensity, and terrain coordinates and altitude. The collected real-time multi-modal data also needs to be preprocessed, including operations such as normalization and feature extraction.

[0047] The preprocessed real-time multi-modal data is input into the trained multi-modal neural network model, and the model will output a preliminary control decision suggestion according to the learned mapping relationship. This decision suggestion contains the probability distribution of different control actions. Then, a double-layer reinforcement learning strategy is used to generate control instructions. Specifically, the double-layer reinforcement learning strategy includes an upper-layer strategy and a lower-layer strategy, and they cooperate with each other to jointly generate the optimal control instructions.

[0048] Among them, the upper-layer strategy is responsible for formulating control objectives and strategies. For example, the upper-layer strategy can determine an overall control direction and priority according to the task requirements of the underwater thruster, such as reaching a specified position, avoiding obstacles, etc. and the current environmental state. If the task requires reaching the target position as soon as possible, and the upper-layer strategy determines that the current water flow direction is beneficial to the forward movement of the thruster, then the upper-layer strategy will tend to select a control direction that makes the thruster accelerate forward.

[0049] The lower-layer strategy then generates specific control instructions according to the guidance of the upper-layer strategy, combined with the current environmental state and the dynamic characteristics of the thruster. The lower-layer strategy adopts a more refined control method. For example, according to the physical model of the thruster and real-time data, specific parameters such as the rotation speed and steering angle of the thruster are calculated.

[0050] During the generation process of the dual-layer reinforcement learning strategy, the terminal will also continuously try different control commands through interaction with the environment, and adjust the strategy according to the rewards or punishments obtained after executing the commands, which can be achieved through a reward function. The reward function can be defined according to the control objectives and task requirements. For example, if the thruster successfully reaches the target position, a large positive reward can be given; if a collision occurs or the target direction is deviated from, a negative reward is given. After multiple iterations of learning and strategy adjustment, the terminal can learn the optimal control strategy, thereby being able to generate more reasonable and efficient control commands.

[0051] In step 106, the power parameters of the underwater thruster are generated according to the control command. Specifically, the control command is a direct signal guiding the movement of the underwater thruster, but these commands are usually represented in an abstract form, such as direction commands like forward, backward, left turn, right turn, etc. and speed magnitude commands. Therefore, it is first necessary to parse the control command and convert it into specific control parameters.

[0052] For example, for the direction command, it can be converted into the steering angle of the thruster; for the speed magnitude command, it can be converted into the target rotational speed of the thruster. During the parsing process, it can be ensured that the converted control parameters can accurately meet the requirements of the control command.

[0053] According to the parsed control parameters, combined with the physical characteristics and dynamic model of the underwater thruster, the corresponding power parameters can be calculated. The power parameters of the thruster mainly include rotational speed and thrust magnitude. There is a certain relationship between the thrust and rotational speed of the thruster, which can be described by the performance curve or mathematical model of the thruster. For example, the thrust of the thruster can be expressed as a quadratic function or a higher-order function of the rotational speed. According to the required thrust magnitude of the control command, the rotational speed that the thruster needs to reach can be calculated by solving this functional relationship.

[0054] At the same time, in practical applications, factors such as water flow resistance affecting the thrust of the thruster need to be considered. The water flow resistance is related to factors such as water flow speed, the shape and size of the thruster. When calculating the power parameters, the water flow resistance should be taken into account to ensure that the thruster can generate sufficient thrust to overcome the resistance during actual operation to achieve the expected motion effect.

[0055] In step 107, when it is determined that the power parameter is within the preset parameter range, the underwater propeller is controlled to move according to the power parameter; specifically, after the power parameter is generated, the power parameter is checked for safety, and the preset parameter range is determined based on the design specifications, performance limitations, and safety requirements of the underwater propeller. For example, the speed of the underwater propeller cannot exceed its maximum allowable speed, otherwise it may cause the propeller to overheat, be damaged, or lose control, and the thrust size cannot exceed the range that its structural strength can withstand, otherwise it may cause damage to the mechanical structure of the propeller.

[0056] Therefore, safety checks can be performed by comparing the generated power parameters with the preset parameter range. If the power parameters exceed the preset parameter range, it means that these parameters are unsafe or infeasible, and the control instructions or power parameter generation process need to be readjusted. For example, if the calculated propeller speed exceeds the maximum allowable speed, the speed requirement in the control instruction can be reduced, or the control strategy of the propeller can be adjusted to generate power parameters that meet safety requirements.

[0057] If it is determined that the generated power parameters are within the preset parameter range, it means that these parameters are safe and feasible. At this time, the power parameters are sent to the actuator of the underwater thruster, such as the motor controller. The motor controller controls the motor operation of the thruster according to the received power parameters and adjusts the speed and direction of the thruster. During the movement of the thruster, its operating status and environmental changes need to be monitored in real time to ensure the safe operation of the underwater thruster.

[0058] Therefore, this application generates control instructions by adopting a double-layer reinforcement learning strategy, so that the control method can continuously learn and optimize the control strategy according to real-time multimodal data, improve the autonomous decision-making ability and adaptability of the underwater propeller in different environments, achieve more intelligent and efficient control, and improve the propulsion efficiency of the underwater propeller. And the power parameters of the underwater propeller are generated according to the control instructions, and the propeller operation is controlled only when it is determined that the power parameters are within the preset parameter range, which can effectively avoid propeller failure or damage caused by abnormal parameters and ensure the safety and stability of the underwater propeller.

[0059] Please refer to Figure 2 According to some embodiments of the present invention, the step 103 of dispersing and filtering the multimodal data and fusing them to obtain the target multimodal data may specifically include, but is not limited to, the following: 201. Preprocess the multimodal data to obtain preprocessed data; 202. Extracting water flow data and terrain data from the preprocessed data; 203. According to the data characteristics of water flow data and terrain data, Kalman filter algorithm and Gaussian filter algorithm are used to eliminate abnormal values ​​respectively; 204. Performing data registration on the filtered water flow data and terrain data; 205. Using weighted average algorithm to fuse the data after data registration; The weighted average algorithm is:

[0060] Where S is the weighted average, n is the data point, x i is the value of the ith data point, w i is the weight of the i-th data point; 206. Verify the compared and fused data. When it is determined that the data verification passes, smooth the data and output the target multimodal data.

[0061] In an embodiment of the present application, after obtaining the multimodal data for controlling the underwater thruster, in order to be able to analyze and process more accurately and efficiently in the future, and then achieve precise control, these multimodal data must first be preprocessed. The preprocessing stage will clean the multimodal data to remove those data points that are obviously wrong or do not conform to the actual situation. For example, some sensors may output abnormally high or abnormally low values ​​due to external interference or their own failures. These abnormal values ​​will affect the accuracy of subsequent data analysis, so they will be eliminated during preprocessing. At the same time, the data will also be formatted to ensure that data from different sources and in different formats can be unified into a standard format, which is convenient for subsequent unified processing and analysis. After the preprocessing operation, the preprocessed data is obtained, and the preprocessed data lays the foundation for subsequent data extraction and analysis.

[0062] Then, water flow data and terrain data are extracted from the preprocessed data. The water flow data contains information such as the water flow speed and direction in the underwater environment, while the terrain data describes the terrain features around the underwater propeller, such as the undulations of the bottom of the water and the location of obstacles. The obtained water flow data and terrain data help plan the movement path of the underwater propeller and avoid collisions with obstacles.

[0063] Among them, since water flow data and terrain data have different data characteristics, different filtering algorithms need to be used to remove outliers. For water flow data, it is characterized by certain dynamics and randomness, and the water flow speed and direction may change with time and space. Therefore, the Kalman filter algorithm is used for filtering. The Kalman filter algorithm can effectively remove noise and outliers in water flow data, and uses the state estimation value of the previous moment and the observed value of the current moment to continuously update the state estimation through recursion, so that the estimated value can more accurately reflect the real water flow state.

[0064] For terrain data, it is characterized by certain spatial correlation and smoothness. Data such as terrain height usually does not have significant mutations between adjacent regions. Therefore, the Gaussian filtering algorithm is used to process terrain data. Gaussian filtering is a linear smoothing filter that removes noise by weighted averaging of data. The Gaussian filtering algorithm uses the Gaussian function as the weight function, making the data points closer to the center point have larger weights and the data points farther from the center point have smaller weights. In this way, during the filtering process, it can effectively remove noise and better retain the overall characteristics and edge information of terrain data. By using the Kalman filtering algorithm and the Gaussian filtering algorithm to process water flow data and terrain data respectively, the quality and reliability of the data can be greatly improved.

[0065] There may still be certain differences between the water flow data and terrain data after filtering. For example, the coordinate systems, time bases, etc. of the data may be inconsistent, which will affect subsequent data fusion and analysis. Therefore, it is necessary to register the filtered water flow data and terrain data. The purpose of data registration is to unify data from different sources and in different formats into a common coordinate system and time base, so that they can be compared and analyzed in the same spatial and time framework.

[0066] When performing data registration, find the corresponding relationship between the water flow data and terrain data. For example, through methods such as feature point matching and spatial transformation, accurately correspond the position information in the water flow data with the position information in the terrain data. At the same time, the time information is also synchronized to ensure that different data are synchronized in time. Through data registration, the water flow data and terrain data can be better fused together, providing an accurate basis for subsequent data fusion.

[0067] Finally, the weighted average algorithm is used to fuse the water flow data and terrain data after data registration. The weighted average algorithm is as follows:

[0068] where S is the weighted average value, n is the data point, x i is the value of the i-th data point, and w i is the weight of the i-th data point; the weighted average algorithm assigns different weights according to the importance and reliability of different data, and then sums the weighted data to obtain the fused data.

[0069] Among them, it should be noted that when determining the weights, the influence degrees of the water flow data and the terrain data on the control decision of the underwater thruster need to be considered. For example, if it is considered that the water flow speed has a greater impact on the movement of the thruster, a higher weight can be assigned to the water flow data; while some detailed features in the terrain data have a smaller impact on the path planning of the thruster, a lower weight is assigned. By reasonably allocating the weights, the weighted average algorithm can combine the advantages of the water flow data and the terrain data to obtain a more comprehensive and accurate description of the environmental information. The fused data will provide a more reliable basis for the control decision of the underwater thruster, enabling the thruster to better adapt to the complex underwater environment and achieve safe and efficient operation.

[0070] Please refer to Figure 3 , according to some embodiments of the present invention, step 104 of constructing a multimodal neural network model based on the target multimodal data and through a temporal attention mechanism may specifically include, but are not limited to, the following: 301. Extract the numerical, spatial, and temporal features in the target multimodal data and convert them into vectors of a unified dimension; 302. Arrange the vectors of the unified dimension in a time series to obtain a feature matrix; 303. Construct a multimodal neural network model according to the data characteristics of the feature matrix and using a temporal attention mechanism.

[0071] In the embodiments of the present application, various features in the target multimodal data are first extracted and converted. Among them, the target multimodal data covers a variety of information about the environment where the underwater thruster is located, including numerical data such as water flow speed, pressure, and intensity, spatial data such as terrain coordinates and height, and temporal data such as the dynamic information of water flow changing over time. For numerical features, such as the water flow speed, the speed values collected at different times and positions are different, and these numerical values directly reflect the changes in the water flow under different conditions. By performing statistical analysis on these numerical values, such as calculating the mean, variance, maximum value, minimum value, etc., the overall characteristics and fluctuation characteristics of the water flow speed can be extracted.

[0072] The spatial features are mainly reflected in the terrain coordinates and height, and these data describe the geometric shape of the environment around the underwater thruster. A spatial interpolation algorithm is used to process the terrain data to obtain more refined terrain features, such as the slope and undulation degree of the terrain.

[0073] In terms of temporal features, since data such as water flow speed and pressure change over time, by calculating the autocorrelation function, power spectral density, etc. of the data, features such as periodicity and trend in the temporal data can be extracted.

[0074] After extracting various types of features, they need to be converted into vectors of a unified dimension. This is because different types of features have different data structures and dimensions, and directly processing them will make it difficult for the model to learn effectively. For numerical features, they can be directly composed into a vector, but for the sake of unifying with other feature dimensions, normalization processing may be required to scale the numerical values to an appropriate range.

[0075] For spatial features, by mapping the terrain coordinates and height information into a specific spatial coordinate system and then converting them into vector form according to grid division. For example, dividing the terrain area into several small grids, each grid corresponding to a feature value, thus forming a high-dimensional vector.

[0076] For time-series features, a time-series encoding method is adopted. For example, slicing the time-series data according to time steps, the features of each time step form a sub-vector, and then these sub-vectors are concatenated into a long vector. It should be noted that during the conversion process, it is necessary to ensure that the positions and weights of various features in the vector are reasonable so that the subsequent model can make full use of these feature information.

[0077] After completing the conversion of the feature vectors, arrange these vectors of unified dimension in time series to obtain a feature matrix. For example: Suppose multiple groups of target multi-modal data are collected over a period of time. Each group of data is obtained as a vector of unified dimension after feature extraction and conversion. Arrange these vectors in chronological order. Each row represents the feature vector at a time point, thus forming a two-dimensional feature matrix. The number of rows of the feature matrix represents the number of time steps, and the number of columns represents the dimension of the feature vector. Through this arrangement method, the feature matrix not only retains the feature information of each time point but also reflects the change relationship of the features over time. In the feature matrix, the difference between adjacent row feature vectors can reflect the change trends of parameters such as water flow velocity and pressure over time, thus helping to understand the dynamic behavior of the water flow and predict future changes.

[0078] Finally, according to the data characteristics of the feature matrix, use the self-attention mechanism to construct a multi-modal neural network model. The feature matrix has the characteristics of high data dimension and strong time-series correlation, and the self-attention mechanism can handle these characteristics well. The self-attention mechanism enables the model to automatically focus on the relevant feature information in other time steps when processing the features of each time step.

[0079] When constructing a neural network model, first input the feature matrix into an embedding layer to convert discrete feature values into continuous vector representations so that the model can process them better. Then, input the embedded vectors into the self-attention layer. The self-attention layer calculates the similarity between the query vector, key vector, and value vector to obtain the attention weights of each time-step feature to other time-step features.

[0080] After that, according to these attention weights, perform a weighted sum on the value vectors to obtain the weighted feature representation. In this way, when the model processes the features at each time step, it can dynamically focus on the important information at different time steps, thereby better capturing the temporal dependence relationship between features.

[0081] To further improve the performance of the model, during the training process of the model, use the backpropagation algorithm and optimizer to continuously adjust the model parameters to make the output of the model as close as possible to the true control decision or target value. At the same time, use the validation set to verify the model, and adjust the hyperparameters of the model according to the performance metrics on the validation set to prevent the model from overfitting or underfitting. After multiple iterations of training and optimization, a multi-modal neural network model that can fully utilize the information of the feature matrix and accurately predict the control instructions of the underwater thruster can be obtained.

[0082] Please refer to Figure 4 , according to some embodiments of the present invention, inputting the real-time multi-modal data into the multi-modal neural network model and generating control instructions using a double-layer reinforcement learning strategy may specifically include, but are not limited to, the following: 401. Obtain real-time multi-modal data; 402. Preprocess the real-time multi-modal data to form an input feature vector adapted to the multi-modal neural network model; 403. Input the input feature vector into the multi-modal neural network model to obtain a target feature vector; 404. Build a double-layer reinforcement learning strategy network, which includes a high-level decision-making network and a low-level execution network. The high-level decision-making network is used to receive the target feature vector output by the multi-modal neural network model and plan the control strategy, and the low-level execution network is used to refine the control strategy into control parameters; 405. Input the target feature vector into the double-layer reinforcement learning strategy network and update the policy network parameters using the Actor-Critic algorithm; 406. Generate control instructions based on the policy network parameters and in combination with the current environmental state.

[0083] In the embodiments of the present application, after obtaining real-time multi-modal data, it is necessary to preprocess it to form an input feature vector adapted to the multi-modal neural network model. Since the data collected by different sensors vary in formats, dimensions, etc., directly using these raw data will lead to difficulties in model training or poor performance. Therefore, it is necessary to normalize the data to unify the numerical ranges of different modal data into a suitable interval, such as normalizing the data to between [0, 1] or [-1, 1], to avoid affecting the model performance due to excessive differences in numerical ranges.

[0084] Next, feature extraction is performed. For continuous data such as water flow velocity and pressure, methods such as Fourier transform and wavelet transform are used to extract their frequency-domain features or time-frequency features, which can better reflect the internal laws of the data; for spatial data such as terrain coordinates and height, methods such as principal component analysis (PCA) and independent component analysis (ICA) are used to extract their main features, retaining key information while reducing the data dimension. Through preprocessing operations such as normalization and feature extraction, the original real-time multi-modal data is converted into an input feature vector that can be understood and processed by the multi-modal neural network model.

[0085] The input feature vector formed after preprocessing is input into the multi-modal neural network model to obtain the target feature vector. Specifically, since the multi-modal neural network model has undergone a large amount of training in the early stage, it has learned the mapping relationship from the input feature vector to the target feature vector. When the input feature vector enters the model, each network layer inside the model will perform a series of complex calculations and transformations on it. For example, the convolutional neural network (CNN) sub-network will perform convolution and pooling operations on the features related to water flow to automatically extract the local features and spatial structures in the data, and the recurrent neural network (RNN) or its variant sub-network will process the features with temporal or spatial dependence relationships such as terrain to capture the long-term dependence relationships in the data. Through the collaborative work of these network layers, the input feature vector is gradually converted into a more representative and discriminative target feature vector, which contains information about the environmental state and the motion requirements of the thruster.

[0086] Building a double - layer reinforcement learning policy network is an important part of the entire control process. This double - layer reinforcement learning policy network includes a high - level decision - making network and a low - level execution network. The main role of the high - level decision - making network is to receive the target feature vectors output by the multi - modal neural network model and plan control strategies based on these feature vectors. It can comprehensively consider the task objectives of the underwater thrusters, such as reaching a specified position, avoiding obstacles, etc., and the current environmental state, and determine the overall control directions and priorities, such as the movement direction and speed range of the thrusters, from a macroscopic level. The low - level execution network is responsible for refining the control strategies planned by the high - level decision - making network into specific control parameters. Combining the physical model of the thruster and real - time data, it calculates precise parameters such as the rotation speed and steering angle of the thruster to ensure that the thruster can move accurately according to the strategies formulated by the high - level decision - making network.

[0087] After inputting the obtained target feature vectors into the double - layer reinforcement learning policy network, the Actor - Critic algorithm is used to update the policy network parameters. The Actor - Critic algorithm is a reinforcement learning algorithm that combines the policy gradient method and the value function estimation method. In the double - layer reinforcement learning policy network, the Actor network is responsible for generating control strategies or control parameters according to the current state, and the Critic network is used to evaluate the quality of the actions selected by the Actor network, that is, to estimate the value function of the state - action pair.

[0088] When the target feature vectors are input into the network, the Actor network generates a control strategy or parameter according to the current state. After the thruster executes this action, the environment will feedback a reward signal. The Critic network updates the estimation of the value of the state - action pair based on this reward signal and subsequent state transition information. At the same time, the Actor network adjusts its own policy parameters according to the value estimation information provided by the Critic network, so that when encountering a similar state in the future, it selects actions with higher values.

[0089] By continuously interacting with the environment, the Actor - Critic algorithm can gradually optimize the parameters of the double - layer reinforcement learning policy network, improving the accuracy and effectiveness of the network in generating control strategies and parameters.

[0090] Finally, control instructions are generated based on the updated policy network parameters and combined with the current environmental state. Specifically, after the parameters of the double - layer reinforcement learning policy network are continuously optimized by the Actor - Critic algorithm, the terminal can generate more reasonable and efficient control instructions according to the input target feature vectors and the current environmental state to ensure the safe operation of the thruster.

[0091] Please refer to Figure 5, according to some embodiments of the present invention, generating the power parameters of the underwater thruster according to the control instruction in step 106 may specifically include, but are not limited to, the following: 501. Calculate the output power of the underwater thruster according to the control instruction; 502. Compare the output power of the underwater thruster with the maximum output power of the underwater thruster, and obtain the comparison result; 503. Generate the power parameters of the underwater thruster according to the comparison result.

[0092] In the embodiments of the present application, after generating the control instruction, it is necessary to calculate the output power of the underwater thruster according to the control instruction. It should be noted that the control instruction is the guiding signal in the entire control process, and the control instruction contains specific requirements for the motion state of the underwater thruster, such as the desired motion speed, direction, and acceleration.

[0093] In order to convert these abstract control requirements into actual operable output power values, the following aspects need to be considered comprehensively: on the one hand, it is necessary to combine the mechanical structure characteristics of the underwater thruster itself, and on the other hand, it is the environmental factors. Based on these factors, an accurate dynamic model is established, and the motion parameters in the control instruction are combined with the mechanical characteristics and environmental parameters of the thruster, and fluid mechanics calculations and dynamic equation solutions are used to calculate the output power required by the underwater thruster under the current control instruction.

[0094] After calculating the output power, compare the output power of the underwater thruster with the maximum output power of the underwater thruster to obtain the comparison result. The maximum output power of the underwater thruster is an important parameter, which is determined by many factors such as the design specifications of the thruster, material strength, and motor performance, and represents the maximum driving force that the thruster can provide within the safe operating range.

[0095] Comparing the calculated current output power with the maximum output power can clearly understand the power demand status of the thruster under the current control requirements. There are three cases for the comparison result: one is that the current output power is less than the maximum output power, which indicates that the thruster can operate safely and stably under the current control instruction without worrying about the problem of power overload; the second is that the current output power is equal to the maximum output power, and at this time the thruster is in a full-load operation state. Although it can meet the current control requirements, it is necessary to pay attention to the operation state of the thruster to avoid power shortage or equipment damage caused by environmental changes or mechanical failures; the third is that the current output power is greater than the maximum output power, which means that operating according to the current control instruction, the thruster will exceed its bearing capacity range, and this situation is very likely to cause serious consequences such as equipment overheating, mechanical component damage, and even out of control. Through this comparison process, a clear basis is provided for the subsequent generation of power parameters.

[0096] After comparison, the comparison result is obtained, and finally, the power parameters of the underwater thruster are generated according to the comparison result. If the comparison result shows that the current output power is less than the maximum output power, then, on the premise of not exceeding the maximum output power, the power parameters can be appropriately adjusted according to the actual requirements and system optimization objectives. For example, in order to improve the operation efficiency or response speed of the thruster, the power output can be appropriately increased within the safe range to generate corresponding power parameters such as rotational speed and thrust.

[0097] If the comparison result is that the current output power is equal to the maximum output power, on the premise of ensuring that the thruster can complete the current task, factors such as equipment heat dissipation and mechanical fatigue should be fully considered, and the power parameters can be finely adjusted, such as reducing the rotational speed or thrust a little, to extend the service life of the thruster.

[0098] When the comparison result is that the current output power is greater than the maximum output power, the control command needs to be re-evaluated and adjusted. One way is to reduce the motion requirements in the control command, such as reducing the desired motion speed or acceleration, so as to recalculate the output power to make it not exceed the maximum output power, and then generate new power parameters. Another way is to optimize the operation strategy of the thruster, such as changing the operation mode of the thruster and adopting intermittent propulsion, etc. While meeting certain task requirements, ensure that the power output is within the safe range. By generating reasonable power parameters according to different comparison results, it can ensure that the underwater thruster can operate safely and efficiently in various complex environments and achieve precise motion control.

[0099] Please refer to Figure 6 , according to some embodiments of the present invention, when it is determined that the power parameters are within the preset parameter range in step 108, controlling the underwater thruster to move according to the power parameters may specifically include, but is not limited to the following: 601. Analyze the power parameters and verify whether the power parameters are complete; 602. If so, compare the analyzed power parameters with the preset parameter range; 603. When it is determined that the power parameters are within the preset parameter range, convert the power parameters into control signals executable by the underwater thruster; 604. Control the underwater thruster to move according to the control signal.

[0100] In the embodiments of the present application, after the generation of the power parameters of the underwater thruster, in order to ensure that these parameters can accurately and effectively control the thruster movement, the power parameters need to be processed and verified.

[0101] First, perform the operation of parsing the power parameters and verifying their integrity. Power parameters usually exist in a specific data format or coding form. The parsing process is to decompose these complex data structures into specific and understandable numerical values or instructions. For example, power parameters may include information such as the rotational speed of the thruster, the magnitude of the thrust, and the steering angle. When parsing, these information need to be extracted from the overall data according to the protocol or rules. At the same time as parsing, the integrity of the power parameters must be verified. Integrity verification includes multiple aspects. On the one hand, it is necessary to check whether the data is complete, that is, whether all necessary parameters are included in the set of power parameters. For example, if the motion control of the thruster requires knowing both the rotational speed and the steering angle at the same time, then during verification, it is necessary to confirm that both of these parameters exist and the data is valid. On the other hand, it is necessary to check whether the data is damaged or incorrect. By using data verification methods such as checksum and cyclic redundancy check (CRC), it is possible to detect whether there are errors in the data. If it is found that the data is incomplete or there are errors, it is necessary to re-obtain or generate the power parameters to ensure the accuracy of the subsequent control process.

[0102] If the power parameters are verified to be complete, next, compare the parsed power parameters with the preset parameter intervals. The preset parameter intervals are determined comprehensively based on factors such as the design specifications, performance limitations, and safe operation requirements of the underwater thruster, and set a reasonable range for each power parameter of the thruster. For example, the rotational speed of the thruster may have a minimum value and a maximum value, and the magnitude of the thrust also has its upper and lower limits.

[0103] Compare the parsed power parameters with these preset intervals one by one to determine whether the current power parameters are within a safe and reasonable range. During the comparison process, consider the specific numerical values of each parameter and their mutual relationships. For example, there may be a certain relationship between the rotational speed and the thrust. When the rotational speed is too high, the thrust may also increase accordingly, but both need to ensure that they are within the preset intervals. If it is found during the comparison process that a certain power parameter exceeds the preset interval, it means that operating the thruster according to the current parameters may pose risks, such as equipment overheating, mechanical damage, or inability to achieve the expected motion effect, etc. At this time, the power parameters need to be adjusted or regenerated until all parameters are within the preset intervals.

[0104] When it is determined that the power parameters are within the preset parameter intervals, the power parameters can be converted into control signals executable by the underwater thruster. Finally, control the underwater thruster to move according to the control signals. The control signals are sent to the actuator of the thruster, such as the motor controller, servo, etc. The motor controller adjusts the rotational speed and steering of the motor according to the received control signals, thereby driving the propeller or other power components of the thruster to generate the corresponding thrust.

[0105] It should be noted that during the movement of the thruster, it is necessary to monitor its operating status in real time, including parameters such as rotational speed, thrust, current, and temperature. These data are collected through sensors and fed back to the control system for comparison with the preset target values. If a deviation is found between the actual operating status and the target value, the control system can timely adjust the control signal to dynamically correct the power parameters of the thruster to ensure that the thruster can move along the expected trajectory and speed. For example, when the water flow suddenly changes and causes the actual speed of the thruster to be lower than the target speed, the control system can increase the control signal to increase the rotational speed of the thruster, thereby increasing the thrust and enabling the thruster to return to the expected motion state. Through this closed-loop control method, precise and stable control of the underwater thruster can be achieved, ensuring its safe and efficient completion of various tasks.

[0106] Please refer to Figure 7 , according to some embodiments of the present invention, after comparing the parsed power parameters with the preset parameter range in step 108, it may specifically include, but is not limited to, the following: 701. When it is determined that the power parameter is not within the preset parameter range, an alarm instruction is triggered; 702. An execution instruction is generated according to the alarm instruction; 703. The maximum value parameter within the preset parameter range is retrieved according to the execution instruction; 704. The underwater thruster is controlled to move according to the maximum value parameter.

[0107] In the embodiment of the present application, when verifying the generated power parameter and determining that it is not within the preset parameter range, the system will quickly trigger an alarm instruction. After the alarm instruction is triggered, the system will immediately generate an execution instruction according to this instruction.

[0108] After generating the execution instruction, the system will retrieve the maximum value parameter within the preset parameter range according to this instruction. The maximum value parameter within the preset parameter range is selected and verified, and it is the maximum driving force or the highest performance index that can be provided on the premise of ensuring the safe operation of the thruster. For example, in the rotational speed parameter of the thruster, the maximum value parameter is determined under the limit conditions such as the heat dissipation capacity of the motor and the load-bearing capacity of the mechanical transmission components. After obtaining the maximum value parameter, the system will control the underwater thruster to move according to these parameters to ensure that the thruster always operates in a safe and efficient state and completes the established underwater tasks.

[0109] Please refer to Figure 8 , the second aspect of the present application provides an underwater thruster control device based on multi-parameter fusion, including: An acquisition unit 801 for acquiring the water flow speed and regional terrain information of the target area where the underwater thruster is located; The first acquisition unit 802 is configured to establish a real-time three-dimensional flow field perception model based on the water flow velocity and the regional terrain information, and acquire multimodal data according to the real-time three-dimensional flow field perception model, where the multimodal data includes the water flow velocity, water pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of the target area; The second acquisition unit 803 is configured to perform decentralized filtering and fusion on the multimodal data to obtain target multimodal data; The construction unit 804 is configured to construct a multimodal neural network model based on the target multimodal data and through a temporal attention mechanism; The first generation unit 805 is configured to input the real-time multimodal data of the underwater thruster into the multimodal neural network model and generate a control instruction by adopting a double-layer reinforcement learning strategy; The second generation unit 806 is configured to generate power parameters of the underwater thruster according to the control instruction; The control unit 807 is configured to control the movement of the underwater thruster according to the power parameters when it is determined that the power parameters are within a preset parameter range.

[0110] Please refer to Figure 9 , this application also provides an underwater thruster control device based on multi-parameter fusion, including: A processor 901, a storage 902, an input / output unit 903, and a bus 904; The processor 901 is connected to the storage 902, the input / output unit 903, and the bus 904; The storage 902 stores a program, and the processor 901 calls the program to execute any of the above methods.

[0111] This application also relates to a computer-readable storage medium, on which a program is stored. When the program runs on a computer, the computer is enabled to execute any of the above methods.

[0112] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be described herein again.

[0113] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0114] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0115] In addition, each functional unit in various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0116] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs and other various media that can store program codes.

Claims

1. An underwater thruster control method based on multi-parameter fusion, characterized in that, The method comprises: Collect water flow velocity and regional terrain information in the target area where the underwater propeller is located; A three-dimensional flow field real-time perception model is established based on the water flow velocity and the regional terrain information, and multimodal data is acquired according to the three-dimensional flow field real-time perception model, wherein the multimodal data includes water flow velocity, water flow pressure, water flow intensity, three-dimensional terrain coordinates and terrain height of the target area; Dispersed filtering and then fusing the multimodal data to obtain target multimodal data; Based on the target multimodal data, a multimodal neural network model is constructed through a temporal attention mechanism; Inputting real-time multimodal data into the multimodal neural network model and generating control instructions using a two-layer reinforcement learning strategy; generating power parameters of the underwater thruster according to the control instruction; When it is determined that the power parameter is within a preset parameter range, the underwater propeller is controlled to move according to the power parameter.

2. The underwater thruster control method based on multi-parameter fusion according to claim 1, wherein The multimodal data is dispersedly filtered and then fused to obtain target multimodal data, including: Preprocessing the multimodal data to obtain preprocessed data; Extracting water flow data and terrain data from the preprocessed data; According to the data characteristics of water flow data and terrain data, Kalman filter algorithm and Gaussian filter algorithm are used to eliminate abnormal values ​​respectively; Perform data registration on the filtered water flow data and terrain data; The weighted average algorithm is used to fuse the data after data registration; The weighted average algorithm is: Where S is the weighted average, n is the data point, x i is the value of the i-th data point, and w i is the weight of the i-th data point; The compared fused data is verified, and when it is determined that the data verification passes, the data is smoothed and then the target multimodal data is output.

3. The underwater thruster control method based on multi-parameter fusion according to claim 1, characterized in that Based on the target multimodal data, a multimodal neural network model is constructed through a temporal attention mechanism, including: Extracting numerical, spatial and temporal features from the target multimodal data and converting them into a vector of uniform dimension; Arranging the uniform dimension vectors in time series to obtain a feature matrix; According to the data characteristics of the feature matrix, a multimodal neural network model is constructed using a temporal attention mechanism.

4. The underwater thruster control method based on multi-parameter fusion according to claim 1, wherein The real-time multimodal data is input into the multimodal neural network model, and a two-layer reinforcement learning strategy is used to generate control instructions, including: Acquire real-time multimodal data; Preprocessing the real-time multimodal data to form an input feature vector adapted to the multimodal neural network model; Inputting the input feature vector into the multimodal neural network model to obtain a target feature vector; Building a two-layer reinforcement learning strategy network, the two-layer reinforcement learning strategy network includes a high-level decision network and a low-level execution network, the high-level decision network is used to receive the target feature vector output by the multimodal neural network model and plan the control strategy, and the low-level execution network is used to refine the control strategy into control parameters; Inputting the target feature vector into the two-layer reinforcement learning strategy network, and updating the strategy network parameters using the Actor-Critic algorithm; A control instruction is generated based on the policy network parameters and in combination with the current environment state.

5. The underwater thruster control method based on multi-parameter fusion according to claim 1, wherein Generating the power parameters of the underwater thruster according to the control instruction includes: Calculating the output power of the underwater thruster according to the control instruction; Compare the output power of the underwater thruster with the maximum output power of the underwater thruster and obtain the comparison result; Generate the power parameter of the underwater thruster according to the comparison result.

6. The underwater thruster control method based on multi-parameter fusion according to claim 1, characterized in that When it is determined that the power parameter is within the preset parameter range, control the underwater thruster to move according to the power parameter, including: Analyze the power parameter and verify whether the power parameter is complete; If so, compare the analyzed power parameter with the preset parameter range; When it is determined that the power parameter is within the preset parameter range, convert the power parameter into a control signal executable by the underwater thruster; Control the underwater thruster to move according to the control signal.

7. The underwater thruster control method based on multi-parameter fusion according to claim 6, characterized in that, After comparing the analyzed power parameter with the preset parameter range, the method further includes: When it is determined that the power parameter is not within the preset parameter range, trigger an alarm instruction; Generate an execution instruction according to the alarm instruction; Retrieve the maximum value parameter within the preset parameter range according to the execution instruction; Control the underwater thruster to move according to the maximum value parameter.

8. An underwater thruster control device based on multi-parameter fusion, characterized in that, The device includes: An acquisition unit for acquiring the water flow velocity and regional terrain information of the target area where the underwater thruster is located; A first acquisition unit for establishing a three-dimensional flow field real-time perception model based on the water flow velocity and the regional terrain information, and acquiring multi-modal data according to the three-dimensional flow field real-time perception model, where the multi-modal data includes the water flow velocity, water pressure, water flow intensity, three-dimensional terrain coordinates, and terrain height of the target area; A second acquisition unit for performing decentralized filtering and fusion on the multi-modal data to obtain target multi-modal data; A construction unit for constructing a multi-modal neural network model based on the target multi-modal data and through a temporal attention mechanism; A first generation unit for inputting the real-time multi-modal data of the underwater thruster into the multi-modal neural network model and generating a control instruction by adopting a double-layer reinforcement learning strategy; A second generation unit for generating the power parameter of the underwater thruster according to the control instruction; A control unit for controlling the underwater thruster to move according to the power parameter when it is determined that the power parameter is within the preset parameter range.

9. An underwater thruster control device based on multi-parameter fusion, characterized in that, The device includes: A processor, a storage, an input / output unit, and a bus; The processor is connected to the storage, the input / output unit, and the bus; The storage stores a program, and the processor calls the program to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A program is stored on the computer-readable storage medium, and when the program is executed on a computer, it executes the method according to any one of claims 1 to 7.

Images