CN-121859763-B - Underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning

CN121859763BCN 121859763 BCN121859763 BCN 121859763BCN-121859763-B

Abstract

The invention provides an underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning, belonging to the technical field of deep reinforcement learning, wherein the method comprises the following steps: determining a first drag reduction scheme according to a preset path, morphological parameters and a deep reinforcement learning model, correcting an actual driving path to obtain a second scheme, arranging an image sensor according to the morphological parameters, acquiring an analysis image to obtain real-time cavitation drag reduction parameters, and generating an optimized scheme by combining multiple parameters and the second scheme. The dynamic self-adaptive optimization of the drag reduction scheme is realized, and the drag reduction efficiency and the sailing stability are improved.

Inventors

GE MINGMING
CHEN BO
ZHANG YUAO
Wei Chifei
Shen Leheng

Assignees

北师香港浸会大学

Dates

Publication Date: 20260508
Application Date: 20260319

Claims (7)

1. An underwater vehicle cavitation drag reduction optimization method based on AI deep reinforcement learning is characterized by comprising the following steps: step 1, determining a first cavitation drag reduction scheme of a target underwater vehicle based on a preset running path of the target underwater vehicle, morphological parameters of the target underwater vehicle and a preset depth reinforcement learning model; Step 2, acquiring an actual running path of the target underwater vehicle in real time, and correcting the first cavitation drag reduction scheme according to the actual running path to obtain a second cavitation drag reduction scheme of the target underwater vehicle; step 3, determining a plurality of image acquisition positions according to morphological parameters of the target underwater vehicle, and installing an image sensor at each image acquisition position; step 4, based on the image sensor, acquiring a real-time running environment image of the target underwater vehicle in real time, and carrying out image analysis on the real-time running environment image to obtain real-time cavitation drag reduction parameters of the target underwater vehicle; step 5, determining a cavitation drag reduction optimization scheme of the target underwater vehicle according to the real-time cavitation drag reduction parameters at each acquisition time, the track coordinates and the navigation attitude parameters of the target underwater vehicle and the second cavitation drag reduction scheme; Wherein, step 5 includes: extracting a drag reduction path and a drag reduction target of each drag reduction direction of the second cavitation drag reduction scheme, performing first association analysis on the drag reduction paths of any two drag reduction directions and performing second association analysis on the drag reduction targets of any two drag reduction directions, and determining a first priority of each drag reduction direction based on the drag reduction paths and a second priority based on the drag reduction targets; First sequencing all drag reduction directions according to the first priority, and carrying out first numbering on each drag reduction direction, and simultaneously, second sequencing all drag reduction directions according to the second priority, and carrying out second numbering on each drag reduction direction; According to the first number and the second number, determining the optimization direction of each drag reduction target according to drag reduction standards, and constructing a basic regulation strategy of the second cavitation drag reduction scheme; Extracting the space-time distribution characteristics of each parameter in the real-time cavitation drag reduction parameters, track coordinates and navigation attitude parameters at the corresponding acquisition time and the coupling correlation characteristics among the parameters, and simultaneously analyzing a navigation section regulation threshold value and a cavitation device configuration reference in the second cavitation drag reduction scheme to construct a cavitation drag reduction full-dimensional coupling characteristic set; Based on the cavitation drag reduction full-dimensional coupling feature set, calculating a quantized deviation value and a state attribution grade of a current cavitation drag reduction state relative to an optimal state by combining pressure and speed distribution parameters of an underwater real-time flow field environment; Constructing a quantized mapping relation between cavitation drag reduction regulation and control requirements and execution parameters according to the state attribution grade and the quantized deviation value, inputting a reference regulation and control parameter, a cavitation drag reduction full-dimensional coupling characteristic set and the quantized deviation value of the second cavitation drag reduction scheme into a deep reinforcement learning model, and constructing a model rewarding function with penalty items in a multi-objective optimization direction by improving drag reduction efficiency, stabilizing navigation posture and correcting track deviation to obtain a multi-dimensional dynamic regulation and control parameter and a self-adaptive fine adjustment parameter of the cavitation device; sequentially carrying out adaptive calculation on cavitation flow fields and removing parameters which are not matched with the current underwater flow field environment and exceed a navigation body power execution threshold value from the multi-dimensional dynamic regulation parameters and the self-adaptive fine tuning parameters by using the target underwater navigation body power to obtain an effective regulation parameter set; and fusing the effective regulation and control parameter set with a basic regulation and control strategy of the second cavitation drag reduction scheme to generate a cavitation drag reduction optimization scheme.
2. The AI deep reinforcement learning-based underwater vehicle cavitation drag reduction optimization method of claim 1, wherein step 1 comprises: determining preset navigation working condition parameters of each navigation section corresponding to a preset travel path of a target underwater navigation body, wherein the preset navigation working condition parameters comprise navigation speed, navigation depth and attack angle range; inputting preset navigation working condition parameters of the target underwater vehicle in each navigation section and morphological parameters of the target underwater vehicle into a preset depth reinforcement learning model to obtain a sub drag reduction scheme of the target underwater vehicle in each navigation section; And combining the sub drag reduction schemes of each leg according to the connection information between each leg and the adjacent leg to obtain a first cavitation drag reduction scheme of the target underwater vehicle.
3. The AI deep reinforcement learning-based underwater vehicle cavitation drag reduction optimization method of claim 1, wherein step 2 comprises: Determining track coordinates and navigation directions of the target underwater vehicle at each acquisition moment based on a position sensor and a navigation module carried by the target underwater vehicle; Integrating the track coordinates and the navigation directions at each acquisition time to obtain the actual travel path of the target underwater vehicle; predicting a future travel path of the target underwater vehicle based on the actual travel path to obtain a predicted travel path; And correcting the first cavitation drag reduction scheme according to the path deviation of the predicted running path and the preset running path to obtain a second cavitation drag reduction scheme of the target underwater vehicle.
4. The AI-deep reinforcement-learning-based underwater vehicle cavitation drag reduction optimization method of claim 3, wherein modifying the first cavitation drag reduction scheme according to a path deviation of the predicted travel path and the preset travel path comprises: Comparing the predicted travel path with the preset travel path time by time to obtain path deviation parameters of the target underwater vehicle at each future time within a preset time length, wherein the path deviation parameters comprise an offset direction, a horizontal offset amount and a vertical offset amount; Normalizing the path deviation parameters, determining the environment deviation parameters corresponding to the future time according to the environment information of the preset running path and the environment information of the predicted running path of the target underwater vehicle at each future time, and determining the weight of each path deviation parameter corresponding to the future time; Calculating the deviation weight of the target underwater vehicle at the corresponding future moment according to the weight of the path deviation parameter corresponding to each future moment and the normalized path deviation parameter; dividing the preset time length of the first future time after the actual driving path according to a preset sliding window, determining the occurrence position of each future time in each divided time window, and determining the deviation degree of the corresponding divided time window by combining the deviation weight of the corresponding future time; Taking the average value of the deviation degrees of all the divided time windows at the first future time as the deviation value corresponding to the first future time, and continuing to analyze the next future time as the first future time to obtain the deviation value of each future time in the preset time length of the first future time after the actual driving path; If the number of the future moments in which the deviation value in the preset time length of the first future moment after the actual running path is located in the first deviation interval is greater than or equal to the first preset number, determining the first cavitation drag reduction scheme of the target underwater vehicle as the second cavitation drag reduction scheme of the target underwater vehicle; And if the number of the future moments in which the deviation value in the preset time length of the first future moment after the actual running path is positioned in the second deviation interval is greater than or equal to the second preset number, inputting the deviation value of each future moment in the corresponding preset time length into a preset first deviation adjustment model by combining the first cavitation drag reduction scheme, so as to obtain a second cavitation drag reduction scheme of the target underwater vehicle.
5. The AI deep reinforcement learning-based underwater vehicle cavitation drag reduction optimization method of claim 1, wherein step 3 comprises: extracting key structural parameters from morphological parameters of a target underwater vehicle, wherein the key structural parameters comprise curved surface curvature distribution, cavitation device layout topology and gradual gradient of a bow-stern profile; Under the preset underwater environment condition, simulating the space influence characteristics of the key structural parameters on the cavitation flow field to obtain cavitation sensitive field space distribution characteristics corresponding to the structural parameters; Dividing a cavitation sensitivity level region of the surface of the target underwater vehicle by combining cavitation flow field pressure and speed distribution data obtained by hydrodynamic simulation based on the cavitation sensitivity field spatial distribution characteristics; Fitting a quantized mapping relation between cavitation sensitivity level and flow field parameters, determining image acquisition resolution, acquisition viewing angle and coverage requirement of each cavitation sensitivity level area, and constructing a multi-dimensional acquisition requirement matrix; Based on the multi-dimensional acquisition demand matrix, the performance parameters of the image sensor and the key morphological feature points of each cavitation sensitive level area, a space resolving model based on curved surface curvature constraint is adopted, three-dimensional space coordinates of each image acquisition position are calculated, and the image sensor is installed.
6. The AI deep reinforcement learning-based underwater vehicle cavitation drag reduction optimization method of claim 1, wherein step 4 comprises: controlling the image sensor to acquire real-time running environment images of the target underwater vehicle according to a preset sampling frequency; performing underwater noise filtering, illumination compensation and contrast enhancement processing on the real-time running environment image to obtain a processed real-time running environment image; Inputting the processed real-time running environment image into an image processing model to obtain real-time cavitation drag reduction parameters of the target underwater vehicle, wherein the real-time cavitation drag reduction parameters comprise cavitation outlines, cavitation distribution characteristics and cavitation wrapping rate of the target underwater vehicle.
7. An underwater vehicle cavitation drag reduction optimizing system based on AI deep reinforcement learning, which is characterized by comprising: The scheme determining module is used for determining a first cavitation drag reduction scheme of the target underwater vehicle based on a preset running path of the target underwater vehicle, morphological parameters of the target underwater vehicle and a preset depth reinforcement learning model; the scheme correction module is used for collecting the actual running path of the target underwater vehicle in real time, correcting the first cavitation drag reduction scheme according to the actual running path, and obtaining a second cavitation drag reduction scheme of the target underwater vehicle; The position determining module is used for determining a plurality of image acquisition positions according to the morphological parameters of the target underwater vehicle and installing an image sensor at each image acquisition position; the image analysis module is used for acquiring the real-time running environment image of the target underwater vehicle in real time based on the image sensor, and carrying out image analysis on the real-time running environment image to obtain the real-time cavitation drag reduction parameters of the target underwater vehicle; the scheme optimizing module is used for determining a cavitation drag reduction optimizing scheme of the target underwater vehicle according to the real-time cavitation drag reduction parameters at each acquisition moment, the track coordinates and the navigation attitude parameters of the target underwater vehicle and the second cavitation drag reduction scheme; wherein, the scheme optimization module is used for: extracting a drag reduction path and a drag reduction target of each drag reduction direction of the second cavitation drag reduction scheme, performing first association analysis on the drag reduction paths of any two drag reduction directions and performing second association analysis on the drag reduction targets of any two drag reduction directions, and determining a first priority of each drag reduction direction based on the drag reduction paths and a second priority based on the drag reduction targets; First sequencing all drag reduction directions according to the first priority, and carrying out first numbering on each drag reduction direction, and simultaneously, second sequencing all drag reduction directions according to the second priority, and carrying out second numbering on each drag reduction direction; According to the first number and the second number, determining the optimization direction of each drag reduction target according to drag reduction standards, and constructing a basic regulation strategy of the second cavitation drag reduction scheme; Extracting the space-time distribution characteristics of each parameter in the real-time cavitation drag reduction parameters, track coordinates and navigation attitude parameters at the corresponding acquisition time and the coupling correlation characteristics among the parameters, and simultaneously analyzing a navigation section regulation threshold value and a cavitation device configuration reference in the second cavitation drag reduction scheme to construct a cavitation drag reduction full-dimensional coupling characteristic set; Based on the cavitation drag reduction full-dimensional coupling feature set, calculating a quantized deviation value and a state attribution grade of a current cavitation drag reduction state relative to an optimal state by combining pressure and speed distribution parameters of an underwater real-time flow field environment; Constructing a quantized mapping relation between cavitation drag reduction regulation and control requirements and execution parameters according to the state attribution grade and the quantized deviation value, inputting a reference regulation and control parameter, a cavitation drag reduction full-dimensional coupling characteristic set and the quantized deviation value of the second cavitation drag reduction scheme into a deep reinforcement learning model, and constructing a model rewarding function with penalty items in a multi-objective optimization direction by improving drag reduction efficiency, stabilizing navigation posture and correcting track deviation to obtain a multi-dimensional dynamic regulation and control parameter and a self-adaptive fine adjustment parameter of the cavitation device; sequentially carrying out adaptive calculation on cavitation flow fields and removing parameters which are not matched with the current underwater flow field environment and exceed a navigation body power execution threshold value from the multi-dimensional dynamic regulation parameters and the self-adaptive fine tuning parameters by using the target underwater navigation body power to obtain an effective regulation parameter set; and fusing the effective regulation and control parameter set with a basic regulation and control strategy of the second cavitation drag reduction scheme to generate a cavitation drag reduction optimization scheme.

Description

Underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning Technical Field The invention relates to the technical field of deep reinforcement learning, in particular to an underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning. Background The speed, cruising ability and resistance characteristics of an underwater vehicle directly determine the operation efficiency, and the frictional resistance during high-speed navigation is a key bottleneck for restricting the performance improvement. Cavitation is used as a high-efficiency drag reduction means, cavitation bubbles are formed in a low-pressure area on the surface of an underwater vehicle, a solid-liquid friction interface is converted into a liquid-gas friction interface, and friction resistance can be greatly reduced, so that the cavitation drag reduction technology has become a core research direction of high-performance design of the underwater vehicle. In the prior art, when cavitation drag reduction of an underwater vehicle is realized, a cavitation device with a fixed shape is designed in advance, constant ventilation flow is set, control parameters are determined by experience or off-line simulation, and a control strategy is kept unchanged in the navigation process. However, the fixed parameter strategy in the prior art does not have online optimization capability, so that the actual running scene of the underwater vehicle is difficult to adapt in actual application, and the drag reduction effect can be influenced. Therefore, the invention provides an underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning. Disclosure of Invention The invention provides an underwater vehicle cavitation drag reduction optimization method and system based on AI deep reinforcement learning, which are used for solving the technical problems. The invention provides an underwater vehicle cavitation drag reduction optimization method based on AI deep reinforcement learning, which comprises the following steps: step 1, determining a first cavitation drag reduction scheme of a target underwater vehicle based on a preset running path of the target underwater vehicle, morphological parameters of the target underwater vehicle and a preset depth reinforcement learning model; Step 2, acquiring an actual running path of the target underwater vehicle in real time, and correcting the first cavitation drag reduction scheme according to the actual running path to obtain a second cavitation drag reduction scheme of the target underwater vehicle; step 3, determining a plurality of image acquisition positions according to morphological parameters of the target underwater vehicle, and installing an image sensor at each image acquisition position; step 4, based on the image sensor, acquiring a real-time running environment image of the target underwater vehicle in real time, and carrying out image analysis on the real-time running environment image to obtain real-time cavitation drag reduction parameters of the target underwater vehicle; and 5, determining a cavitation drag reduction optimization scheme of the target underwater vehicle according to the real-time cavitation drag reduction parameters at each acquisition time, the track coordinates and the navigation attitude parameters of the target underwater vehicle and combining the second cavitation drag reduction scheme. Preferably, step 1 comprises: determining preset navigation working condition parameters of each navigation section corresponding to a preset travel path of a target underwater navigation body, wherein the preset navigation working condition parameters comprise navigation speed, navigation depth and attack angle range; inputting preset navigation working condition parameters of the target underwater vehicle in each navigation section and morphological parameters of the target underwater vehicle into a preset depth reinforcement learning model to obtain a sub drag reduction scheme of the target underwater vehicle in each navigation section; And combining the sub drag reduction schemes of each leg according to the connection information between each leg and the adjacent leg to obtain a first cavitation drag reduction scheme of the target underwater vehicle. Preferably, step 2 includes: Determining track coordinates and navigation directions of the target underwater vehicle at each acquisition moment based on a position sensor and a navigation module carried by the target underwater vehicle; Integrating the track coordinates and the navigation directions at each acquisition time to obtain the actual travel path of the target underwater vehicle; predicting a future travel path of the target underwater vehicle based on the actual travel path to obtain a predicted travel path; And correcting the first cavitation drag reduction scheme according to the