Communication and control methods for underwater unmanned vehicles

By dynamically determining the surfacing timing through real-time sea state assessment and multi-factor comprehensive scoring, combined with semantic reasoning and closed-loop control, the problem of rigid communication surfacing decisions and attitude adaptation for underwater unmanned vehicles has been solved, improving communication success rate and energy utilization efficiency.

CN123074579APending Publication Date: 2026-11-10JIHUA LAB
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611544191.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing communication surfacing decisions and attitude adaptation strategies for underwater unmanned vehicles are rigid, leading to satellite search failures, link interruptions, and ineffective energy consumption in harsh sea conditions, lacking intelligent decision-making and dynamic optimization capabilities.

Method used

By collecting multi-dimensional sensing data and combining the current location with historical communication event data from the local memory, real-time sea state assessment and multi-factor comprehensive scoring are performed to dynamically determine the timing of surfacing for communication. If it is not suitable, underwater acoustic backhaul or cyclic reassessment is performed. After surfacing, semantic reasoning is used to identify the root cause of abnormal satellite communication quality and generate high-level semantic intent commands to coordinate the control of buoyancy, center of gravity, mast, and attitude compensation mechanisms, forming a closed-loop control.

Benefits of technology

It improves the success rate of underwater unmanned vehicles' surfacing communication, environmental adaptability, and energy efficiency, avoids satellite search failures and link interruptions in harsh sea conditions, and enables attitude control to directly serve communication quality.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123074579A_ABST
    Figure CN123074579A_ABST
Patent Text Reader

Abstract

This invention relates to the field of underwater vehicles and discloses a communication control method for underwater unmanned vehicles, comprising: when the communication window arrives, acquiring IMU attitude, depth, and battery level; determining the operating sea area, assessing the sea state level and wave height range, and retrieving historical communication events in the same sea area, sea state, and wave height range from the local memory; obtaining the task priority, and scoring it based on sea state, battery level, priority, and historical data; if the target is met, surfacing is initiated; otherwise, the process is delayed and underwater acoustic echo or cyclic reassessment is triggered; after surfacing and establishing a link, acquiring satellite communication quality, mast status, and IMU update data, and semantically reasoning to output fault categories; outputting high-level intents based on fault categories and priorities, transferring them to low-level instructions, driving the execution controller to optimize using a communication quality evaluation function, and coordinating the control of at least one of buoyancy, center of gravity, communication folding mast, and attitude compensation propulsion mechanism to achieve communication quality-oriented attitude adaptation; and storing all process data in the local memory when the communication window ends.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of underwater vehicles, and more particularly to a communication control method for underwater unmanned vehicles. Background Technology

[0002] When underwater unmanned vehicles (UAVs) operate for extended periods, they typically need to surface periodically to communicate with the command center via a satellite communication link on the folding mast. Currently, existing technologies fall into four main categories: First, fixed-period surfacing schemes employ a hard-coded trigger mechanism with fixed timing and range; surfacing is forced upon reaching preset conditions, lacking intelligent decision-making capabilities. Second, conventional attitude closed-loop schemes rely on IMUs for roll and yaw attitude closed-loop control after surfacing, using only zeroing the attitude angle as the control target, operating independently of the communication decision-making system, and not incorporating satellite communication quality into the feedback. Third, simple threshold judgment schemes only introduce thresholds such as sea state and battery level for extreme fault emergencies; these remain fixed-threshold open-loop single-judgment schemes, lacking dynamic optimization capabilities. Fourth, intelligent algorithm-assisted schemes mostly only achieve fitting mapping from sensor data to control quantities, typically lacking semantic reasoning, experience memory, and high-level decision-making capabilities.

[0003] The above-mentioned scheme has the following problems: the fixed-period ascent strategy is rigid, and the surface-up maneuver is still performed in severe sea conditions, which can easily lead to satellite search failure, link interruption, and ineffective consumption of limited energy storage; attitude control and communication quality are completely decoupled, resulting in situations where the attitude angle is qualified but the satellite communication quality is poor; communication anomalies can only be determined by whether the connection is open or closed, and cannot identify the root causes such as wave disturbance, mast jamming, radio frequency hardware failure, or sea state exceeding limits, resulting in a crude handling; the overall control architecture is an open-loop architecture without local memory, result evaluation, and strategy iteration links, and similar problems occur repeatedly in the same sea area.

[0004] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0005] The purpose of this invention is to provide a communication control method for underwater unmanned vehicles, which aims to solve the technical problems of underwater vehicle communication surfacing decision and attitude adaptation in the prior art.

[0006] To achieve the above objectives, the solution provided by the present invention is as follows: A communication control method for an underwater unmanned vehicle includes: when the vehicle is in an underwater navigation state, upon reaching a preset periodic communication time window, collecting multi-dimensional sensing data, including IMU attitude data, navigation depth data, and remaining battery power data; obtaining the vehicle's current position to determine the current operating sea area; performing real-time sea state assessment based on the IMU attitude data and the navigation depth data to obtain the sea state level and wave height range; and retrieving historical communication event data of the same sea state level and wave height range within the current operating sea area from a preset local memory database using the sea state level and wave height range as retrieval conditions; obtaining the priority information of the currently executed task; performing a multi-factor comprehensive score based on the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data; and outputting a surfacing communication intention when the comprehensive score reaches a surfacing determination threshold, and outputting a delayed surfacing intention when the comprehensive score does not reach the surfacing determination threshold, and triggering underwater acoustic echo or entering the next communication window for reassessment; The ascent communication intent triggers the vehicle to ascend to the surface communication depth, deploy the communication mast, and activate the satellite communication module to establish a link. It then collects satellite communication quality data, communication mast folding status data, and IMU attitude update data. Based on these data, semantic reasoning is performed to output standardized fault semantic categories. According to the standardized fault semantic categories and the priority information of the current task, a high-level semantic intent command is output and converted into a low-level control command. This command drives the vehicle's low-level execution controller to coordinate and regulate at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication mast folding mechanism, and attitude compensation propulsion mechanism, using the satellite communication link quality evaluation function as the optimization target, to achieve communication quality-oriented adaptive attitude control. After the current communication window ends, the entire process data of the current ascent communication event is acquired and stored in the local memory.

[0007] Preferably, when the vehicle is in underwater navigation mode, upon reaching a preset periodic communication time window, the collection of multi-dimensional perception data includes: when the vehicle is in underwater navigation mode, upon reaching the preset periodic communication time window, and when the vehicle has sailed to a preset communication depth, collecting multi-dimensional perception data through a perception module, wherein the multi-dimensional perception data includes IMU attitude data, navigation depth data, and remaining battery power data; and inputting the multi-dimensional perception data into an AI agent module for communication control, wherein the AI ​​agent module includes a planning and reflection sub-agent, a communication perception sub-agent, and an attitude and intention sub-agent.

[0008] Preferably, the step of obtaining the current position of the vehicle to determine the current operating sea area, performing real-time sea state assessment based on the IMU attitude data and the navigation depth data to obtain the sea state level and wave height range, and retrieving historical communication event data of the same sea state level and wave height range within the current operating sea area from a preset local memory using the sea state level and wave height range as search criteria, includes: performing frequency domain transformation on the IMU attitude data through a planning and reflection sub-agent, taking the frequency corresponding to the maximum spectral amplitude as the dominant frequency; calculating the peak value of the attitude amplitude of the IMU attitude data, and calculating the depth fluctuation variance of the navigation depth data; based on The pre-defined correspondence between ship motion and wave height is used to convert the peak value of attitude amplitude, the variance of depth fluctuation, and the dominant frequency into sea surface wave height values. Based on the sea surface wave height values, a pre-defined mapping table of wave height and sea state levels in a local memory is consulted to obtain the sea state level. The wave height range is then determined based on the sea surface wave height values. The sea state levels include level 1 light waves, level 2 moderate waves, level 3 large waves, and level 4 giant waves. The current position of the vehicle is obtained to determine the current operating sea area. Using the sea state level and the wave height range as search criteria, historical communication event data within the same sea state level and wave height range in the current operating sea area are retrieved from the pre-defined local memory.

[0009] Preferably, the step of obtaining the priority information of the currently executed task, performing a multi-factor comprehensive score by combining the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data, and outputting an intention to ascend when the comprehensive score reaches the ascent determination threshold, and outputting an intention to postpone ascent when the comprehensive score does not reach the ascent determination threshold, and triggering underwater acoustic echo or entering the next communication window for re-evaluation, includes: the planning and reflection sub-agent reading the type identifier of the currently executed task from the vehicle's task scheduling system; determining the priority information of the currently executed task according to a preset task type and priority mapping relationship, wherein the task type includes emergency tasks, routine tasks, and inspection tasks; performing a multi-factor comprehensive score by combining the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data; outputting an intention to ascend when the comprehensive score reaches the ascent determination threshold, and outputting an intention to postpone ascent when the comprehensive score does not reach the ascent determination threshold, and triggering underwater acoustic echo or entering the next communication window for re-evaluation.

[0010] Preferably, the multi-factor comprehensive scoring method, which combines the sea state level, the remaining power data, the priority information of the currently executed task, and the historical communication event data, includes: performing reverse normalization on the sea state level according to its level to obtain a sea state factor; performing linear normalization on the remaining power data according to the percentage of remaining power to rated capacity to obtain a power factor; discretely assigning values ​​to the priority information of the currently executed task according to a preset task level to obtain a task priority factor; extracting the historical surfacing communication success rate under the same sea area and sea state level from the historical communication event data as a historical success rate factor; and performing a weighted summation of the sea state factor, the power factor, the task priority factor, and the historical success rate factor according to preset weights to obtain a multi-factor comprehensive score.

[0011] Preferably, after the surfacing communication intention triggers the vehicle to surface to the communication depth, deploys the communication mast, and activates the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data, and IMU attitude update data are collected. Semantic reasoning is then performed based on the satellite communication quality data, the communication folding mast operation status data, and the IMU attitude update data to output standardized fault semantic categories, including: after the surfacing communication intention triggers the vehicle to surface to the communication depth, deploys the communication mast, and activates the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data, and IMU attitude update data are collected; and the communication sensing sub-intelligence... The system extracts the mean signal-to-noise ratio, fluctuation amplitude, and duration of anomalies from the satellite communication quality data; extracts the amplitude of sudden changes in driving current, hysteresis of folding angle, and rate of angle change from the communication folding mast operation status data; and extracts the dominant frequency, peak amplitude, and frequency band distribution of disturbances from the IMU attitude update data through time-frequency transformation. The mean signal-to-noise ratio, fluctuation amplitude, duration of anomalies, amplitude of sudden changes in driving current, hysteresis of folding angle, rate of angle change, dominant frequency of disturbance, peak amplitude, and frequency band distribution are input into a pre-constructed time-series feature extraction network to obtain standardized fault semantic categories. These standardized fault semantic categories include at least one of wave disturbance, mast mechanism jamming, radio frequency hardware failure, and sea state exceeding limits.

[0012] Preferably, the step of outputting high-level semantic intent commands based on the standardized fault semantic categories and the priority information of the currently executing task, and converting the high-level semantic intent commands into low-level control commands, and driving the low-level execution controller of the vehicle to coordinate and regulate at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism with the satellite communication link quality evaluation function as the optimization target, to complete the attitude adaptive control guided by communication quality, includes: the planning and reflection sub-agent querying the preset fault-intent mapping relationship based on the standardized fault semantic categories to determine candidate high-level semantic intents, the candidate high-level semantic... The intents include communication optimization intents, attitude control intents, and hazard avoidance termination intents. Combining the priority information of the current task, the candidate high-level semantic intents are arbitrated in descending order of priority, and high-level semantic intent instructions are output. The attitude intent sub-agent uses a semantic intent-execution strategy mapping mechanism to convert the high-level semantic intent instructions into low-level control instructions. These low-level control instructions drive the vehicle's low-level execution controller to coordinate and control at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, using the satellite communication link quality evaluation function as the optimization target, to complete communication quality-oriented adaptive attitude control.

[0013] Preferably, the attitude intention sub-agent uses a semantic intention-to-execution-policy mapping mechanism to convert the high-level semantic intention instruction into low-level control instructions, including: when the high-level semantic intention instruction is a communication optimization intention, the attitude intention sub-agent generates low-level control instructions for controlling the full-range coordinated control of the buoyancy adjustment mechanism, the center of gravity adjustment mechanism, the communication folding mast mechanism, and the attitude compensation propulsion mechanism; when the high-level semantic intention instruction is an attitude control intention, the attitude intention sub-agent generates low-level control instructions for prioritizing the control of the buoyancy adjustment mechanism and the center of gravity adjustment mechanism to complete attitude correction; when the high-level semantic intention instruction is a risk avoidance termination intention, the attitude intention sub-agent generates low-level control instructions only for controlling the retraction of the communication folding mast mechanism and the diving of the buoyancy adjustment mechanism.

[0014] Preferably, the satellite communication link quality evaluation function is expressed as:

[0015] In the formula, This represents the quality evaluation value of the satellite communication link. The signal-to-noise ratio (SNR) extracted from the satellite communication quality data is the normalized value. This is the normalized value of the link packet loss rate extracted from the satellite communication quality data. and These are the weighting coefficients. .

[0016] Preferably, after the current communication window ends, the full-process data of the current ascent communication event is acquired and stored in the local memory, including: after the current communication window ends, the planning and reflection sub-agent summarizes the decision basis, sea state parameters, execution actions, and fault semantic categories of the current ascent communication event; the satellite communication module provides the communication results; and the perception module provides energy consumption data. The decision basis, sea state parameters, execution actions, fault semantics, communication results, and energy consumption data of the current ascent communication event are integrated into the full-process data of the current ascent communication event in a structured case format. The structured case format includes environmental feature fields, task constraint fields, decision action fields, and effect result fields. The environmental feature fields include sea area grid code, sea state level, and wave height range. The task constraint fields include task type, remaining power percentage, and priority level. The decision action fields include final decision type, delay duration, and control intention mode. The effect result fields include communication success rate, average signal-to-noise ratio, fault semantic category, and task completion degree. The integrated full-process data is stored in the local memory.

[0017] The underwater unmanned vehicle communication control method provided by this invention collects multi-dimensional perception data such as IMU attitude, navigation depth, and remaining battery power when the underwater communication window is triggered. It combines the current location with historical communication event data from the same sea area, sea state level, and wave height range in the local memory to perform real-time sea state assessment and multi-factor comprehensive scoring. This allows for dynamic judgment of the timing of surfacing for communication, avoiding satellite search failures, link interruptions, and unnecessary energy consumption caused by blindly surfacing in adverse sea conditions. When surfacing is not suitable, it improves decision-making flexibility and safety by using underwater acoustic feedback or entering the next communication window for cyclical reassessment. After establishing a link upon surfacing, it performs semantic reasoning on satellite communication quality data, communication folding mast operation status data, and IMU attitude update data to output standardized fault semantic categories, thereby distinguishing between wave disturbances, mast jamming, and radio frequency hardware failures. Anomalies such as exceeding sea state limits provide a basis for differentiated handling. Furthermore, standardized fault semantic categories are combined with task priorities to generate high-level semantic intent commands, which are then converted into low-level control commands. Using the satellite communication link quality evaluation function as the optimization target, the buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism are coordinated and controlled. This ensures that attitude control directly serves communication quality rather than merely pursuing zero attitude angles, effectively solving the problem of acceptable attitude but poor communication quality. Finally, after the communication window ends, all process data is saved to the local memory, forming a closed loop of perception, decision-making, execution, evaluation, and review. This allows for continuous accumulation of experience and autonomous optimization of subsequent decisions, thereby improving the overall success rate of underwater unmanned vehicle surfacing communication, environmental adaptability, energy efficiency, and long-term operational autonomy. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the underwater unmanned vehicle communication control method provided in an embodiment of the present invention. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 An underwater unmanned vehicle communication control method according to an embodiment of the present invention includes: S101. When the vehicle is in underwater navigation mode, when the preset periodic communication time window is reached, multi-dimensional sensing data is collected. The multi-dimensional sensing data includes IMU attitude data, navigation depth data and remaining power data. S102. Obtain the current position of the vehicle to determine the current operating sea area, perform real-time sea state assessment based on the IMU attitude data and the navigation depth data to obtain the sea state level and wave height range, and retrieve historical communication event data with the same sea state level and wave height range in the current operating sea area from the preset local memory bank using the sea state level and wave height range as search conditions. S103. Obtain the priority information of the currently executed task, combine the sea state level, the remaining power data, the priority information of the currently executed task and the historical communication event data to perform a multi-factor comprehensive score, and output the intention to ascend when the comprehensive score reaches the ascent judgment threshold, and output the intention to postpone ascent when the comprehensive score does not reach the ascent judgment threshold, and trigger underwater acoustic feedback or enter the next communication window for loop re-evaluation. S104. After the surfacing communication intention triggers the vehicle to surface to the communication depth and deploy the communication mast, and activate the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data and IMU attitude update data are collected. Based on the satellite communication quality data, the communication folding mast operation status data and the IMU attitude update data, semantic reasoning is performed to output a standardized fault semantic category. S105. Output high-level semantic intent instructions based on the standardized fault semantic category and the priority information of the current task being executed, and convert the high-level semantic intent instructions into low-level control instructions. Drive the low-level execution controller of the vehicle to coordinate and regulate at least one of the following: buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, with the satellite communication link quality evaluation function as the optimization target, to complete the attitude adaptive control guided by communication quality. S106. After the current communication window ends, obtain the full process data of the current floating communication event and save it in the local memory.

[0022] In this embodiment, step 101, when the vehicle is in underwater navigation mode and reaches a preset periodic communication time window, collects multi-dimensional perception data, including: when the vehicle is in underwater navigation mode, reaches the preset periodic communication time window, and the vehicle has sailed to a preset communication depth, collects multi-dimensional perception data through a perception module, the multi-dimensional perception data including IMU attitude data, navigation depth data, and remaining battery power data; inputs the multi-dimensional perception data into an AI agent module for communication control, the AI ​​agent module including a planning and reflection sub-agent, a communication perception sub-agent, and an attitude and intention sub-agent.

[0023] When an underwater unmanned vehicle (UAV) routinely performs preset detection, inspection, or mission navigation underwater, the system pre-sets periodic communication time windows. Upon reaching this communication time window, the UAV can navigate to a preset communication depth and collect multi-dimensional sensing data through its sensing modules. The preset communication depth can be set according to the UAV model and mission requirements, for example, 5-15 meters underwater, preferably 10 meters.

[0024] In this embodiment, the sensing module includes an IMU (Inertial Measurement Unit), a depth sensor, a battery status detection unit, a satellite communication module, and a communication folding mast sensing unit. The IMU is used to measure the vehicle's attitude data, specifically the roll and pitch angles. The sampling frequency of the IMU attitude data is 10-20 Hz. During underwater navigation, the IMU attitude data is used for real-time sea state assessment, extracting the dominant frequency and statistically analyzing the peak attitude amplitude through frequency domain transformation, and then converting it into the sea surface wave height. After surfacing and establishing a link, the IMU attitude update data is used to extract the dominant frequency, peak amplitude, and frequency band distribution of disturbances through time-frequency transformation, serving as one of the inputs for the communication sensing sub-agent to perform fault semantic reasoning.

[0025] The depth sensor is used to measure the navigation depth data of the vehicle, that is, the underwater depth at which the vehicle is located. The sampling frequency of the navigation depth data is 1~3Hz. During the underwater navigation phase, the navigation depth data is used for real-time sea state assessment. By statistically analyzing the depth fluctuation variance, combined with the peak value of the attitude amplitude and the dominant frequency, the sea surface wave height is calculated. At the same time, the depth sensor is also used to determine whether the vehicle has sailed to the preset communication depth, so as to trigger the collection of multi-dimensional sensing data.

[0026] The battery status detection unit measures the remaining battery power of the aircraft, specifically the percentage of the battery's rated capacity. This remaining battery power data serves as the power factor input in a multi-factor comprehensive scoring system, determining whether the current battery power meets the energy consumption requirements for ascent communication and attitude control. Furthermore, it provides a basis for postponing ascent decisions if the comprehensive score fails to reach the ascent threshold.

[0027] The satellite communication module is used to measure satellite communication quality data, including signal-to-noise ratio (SNR) and link packet loss rate, after surfacing to the communication depth and deploying the communication mast. This satellite communication quality data is used to extract the mean SNR, fluctuation amplitude, and duration of continuous anomalies, serving as one of the inputs for the communication sensing sub-agent to perform fault semantic reasoning. Simultaneously, the SNR and link packet loss rate are also used to construct a satellite communication link quality evaluation function, serving as an optimization target for attitude adaptive control.

[0028] The communication folding mast sensing unit is used to measure the operational status data of the communication folding mast, specifically including mast drive current, folding angle, and working status. This operational status data is used to extract the amplitude of sudden changes in drive current, hysteresis of the folding angle, and the rate of angle change, serving as one of the inputs for the communication sensing sub-agent to perform fault semantic reasoning, thereby distinguishing standardized fault semantic categories such as wave disturbance, mast mechanism jamming, radio frequency hardware failure, and exceeding sea state limits.

[0029] In this embodiment, the AI ​​agent module is a lightweight, offline deployment architecture that does not rely on the real-time computing power of large cloud models and can operate completely independently in an underwater disconnected environment. The AI ​​agent module includes mutually cooperating planning and reflection sub-agents, communication and perception sub-agents, and attitude and intent sub-agents, as well as a local memory bank and a device task scheduling interface.

[0030] The planning and reflection sub-agent integrates real-time sea state assessment results, remaining vehicle battery power, current task priority, and historical communication effects and debriefing experiences stored in its local memory for comprehensive reasoning. It autonomously determines the timing of surfacing communication and outputs corresponding high-level semantic intent commands. Simultaneously, it performs operational result debriefing, extracting the entire process data into structured cases and updating the database after each communication event, thus achieving continuous autonomous optimization of the control strategy. For example, the planning and reflection sub-agent is responsible for sea state assessment, multi-factor comprehensive scoring, surfacing or postponement decisions, fault and intent mapping, candidate high-level semantic intent arbitration, and debriefing database entry.

[0031] The communication sensing sub-agent receives raw sensor data and communication data collected by the sensing module. It then uses a semantic reasoning model to identify the root causes of communication quality anomalies and outputs standardized fault semantic categories. These categories include four fault scenarios: wave disturbance, mast mechanism jamming, radio frequency hardware failure, and sea state exceeding limits. This achieves accurate classification of communication anomalies, replacing the traditional single-function link connectivity judgment logic. For example, the communication sensing sub-agent is responsible for feature extraction, temporal feature extraction, network reasoning, fault classification, and evidence conflict arbitration.

[0032] The attitude intention sub-agent receives high-level semantic intention instructions output by the planning reflection sub-agent. Combined with the real-time satellite communication link quality, it completes the conversion of semantic intentions into low-level executable control requirements, transforming abstract intentions such as communication optimization, attitude control, and risk avoidance termination into specific control instructions such as buoyancy adjustment, center of gravity adjustment, mast fine-tuning, and propulsion compensation.

[0033] The local memory bank is deployed using UUV local storage media to persistently store data from all ascent communication events, including the operating area, sea state parameters, ascent decision results, communication success or failure status, root causes of failures, power consumption data, and attitude control records. It supports local real-time retrieval and experience reuse. After connecting to the satellite link upon ascent, the memory data can be selectively synchronized to shore-based equipment for cloud backup.

[0034] The equipment task scheduling interface is pre-configured with a set of hardware operations that the UUV can execute independently, including basic operations such as surfacing, diving, deploying and folding the communication mast, and deploying emergency communication buoys. The equipment task scheduling interface is used to transform the high-level semantic decision intentions output by the planning and reflection sub-agent into standardized and directly executable equipment start-stop scheduling instructions.

[0035] In this embodiment, step S102, which involves obtaining the current position of the vehicle to determine the current operating sea area, performing real-time sea state assessment based on the IMU attitude data and the navigation depth data to obtain the sea state level and wave height range, and retrieving historical communication event data of the same sea state level and wave height range from a preset local memory database using the sea state level and wave height range as search criteria, includes: performing frequency domain transformation on the IMU attitude data through a planning and reflection sub-agent, taking the frequency corresponding to the maximum spectral amplitude as the dominant frequency; and statistically analyzing the peak value of the attitude amplitude of the IMU attitude data. The depth fluctuation variance of the navigation depth data is calculated; based on the pre-calibrated correspondence between hull motion and wave height, the peak value of attitude amplitude, the depth fluctuation variance, and the dominant frequency are converted into sea surface wave height values; the sea surface wave height is queried from the preset local memory database to obtain the sea state level, and the wave height range is determined based on the sea surface wave height value; the current position of the vehicle is obtained to determine the current operating sea area, and historical communication event data with the same sea state level and the same wave height range are retrieved from the preset local memory database using the sea state level and the wave height range as search conditions.

[0036] Specifically, real-time sea state assessment is performed by a planning and reflection sub-agent. During the assessment phase, the vehicle maintains a preset communication depth and collects IMU attitude data (i.e., IMU roll and pitch angles) and depth data within a 30-60 second long time window. The peak attitude amplitude, depth fluctuation variance, and dominant frequency within the window are statistically analyzed. The relationship between hull motion and wave height can be calibrated through tank tests or sea trials, eliminating the need for additional wave sensors and reducing hardware costs.

[0037] In this embodiment, sea state levels include Level 1 light waves, Level 2 moderate waves, Level 3 large waves, and Level 4 giant waves. For example, Level 1 light waves correspond to a wave height of 0-0.5m and an absolute roll amplitude of ≤3°; Level 2 moderate waves correspond to a wave height of 0.5-1.5m and an absolute roll amplitude of 3°-8°; Level 3 large waves correspond to a wave height of 1.5-3m and an absolute roll amplitude of 8°-15°; and Level 4 giant waves correspond to a wave height >3m and an absolute roll amplitude >15°. Levels 1 and 2 are suitable for surfacing, Level 3 is a restricted sea state, and Level 4 is a hard threshold prohibiting surfacing.

[0038] In this embodiment, the local memory retrieval mechanism employs a lightweight retrieval logic combining gridded sea area matching and feature-based sea state matching. This eliminates the need for complex vector calculations and is suitable for embedded low-computing-power environments. During sea area matching, the operating sea area is uniformly encoded using a grid of equal latitude and longitude. The retrieval prioritizes matching historical cases within the same grid; if cases are insufficient, matching is degraded to adjacent grids. The matching logic is a simple encoding comparison. For sea state matching, sea state level and wave height range are used as discrete feature vectors. A matching degree scoring system is employed, awarding 1 point for each matched dimension. A total score reaching a set threshold indicates a similar sea state scenario. When outputting results, the Top-K historical successful cases are sorted by matching degree, with attenuation weights assigned based on time proximity. The average communication success rate, recommended delay duration, and historically optimal control parameters for the same scenario are output.

[0039] Specifically, K is a preset positive integer, which is typically 3 to 5.

[0040] In this embodiment, step S103 involves obtaining the priority information of the currently executed task, performing a multi-factor comprehensive score by combining the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data, and outputting an intention to ascend when the comprehensive score reaches the ascent determination threshold, and outputting an intention to postpone ascent when the comprehensive score does not reach the ascent determination threshold, and triggering underwater acoustic echo or entering the next communication window for cyclic reassessment. This includes: the planning and reflection sub-agent reading the type identifier of the currently executed task from the vehicle's task scheduling system; determining the priority information of the currently executed task according to a preset task type and priority mapping relationship, wherein the task type includes emergency tasks, routine tasks, and inspection tasks; performing a multi-factor comprehensive score by combining the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data; outputting an intention to ascend when the comprehensive score reaches the ascent determination threshold, and outputting an intention to postpone ascent when the comprehensive score does not reach the ascent determination threshold, and triggering underwater acoustic echo or entering the next communication window for cyclic reassessment.

[0041] Furthermore, a multi-factor comprehensive score is performed by combining the sea state level, the remaining power data, the priority information of the currently executed task, and the historical communication event data. This includes: reverse normalizing the sea state level according to its level to obtain a sea state factor; linearly normalizing the remaining power data according to the percentage of remaining power to rated capacity to obtain a power factor; discretely assigning values ​​to the priority information of the currently executed task according to a preset task level to obtain a task priority factor; extracting the historical surfacing communication success rate under the same sea area and sea state level from the historical communication event data as a historical success rate factor; and weighting and summing the sea state factor, the power factor, the task priority factor, and the historical success rate factor according to preset weights to obtain a multi-factor comprehensive score.

[0042] For example, emergency tasks are assigned a value of 1.0, routine tasks are assigned a value of 0.7, and inspection tasks are assigned a value of 0.4; sea state level 1 corresponds to 1.0, and level 4 corresponds to 0; battery power 0~100% corresponds to 0~1; and the historical success rate ranges from 0 to 1.

[0043] The surfacing determination threshold is exemplarily 0.6. When sea state reaches level 4, a delayed surfacing is triggered directly, without entering the soft scoring stage, to ensure safety.

[0044] For delayed ascents, the delay duration can be dynamically set according to the current sea state level. For example, a delay of 10-20 minutes is set for sea state 2, 30-60 minutes for sea state 3, and 60-120 minutes for sea state 4; the duration can also be adjusted by referring to historical durations of the same sea state in the local memory. After the delay period ends, the perception module re-collects IMU attitude data and navigation depth data, and the planning and reflection sub-agent re-executes the sea state assessment and multi-factor decision-making. If the ascent conditions are still not met after a preset limit of consecutive delays (e.g., 3 times), low-rate simplified underwater acoustic data transmission is initiated. The transmitted data is categorized by priority from high to low into emergency alarm types, core status types, and decision description types, using a fixed-length simplified frame format. Transmission is triggered only when there are consecutive delayed ascents or when an emergency alarm occurs; underwater acoustic communication is not activated under normal circumstances to reduce energy consumption.

[0045] In this embodiment, in step S104, after the surfacing communication intention triggers the vehicle to surface to the communication depth and deploy the communication mast, and activates the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data, and IMU attitude update data are collected. Semantic reasoning is then performed based on the satellite communication quality data, the communication folding mast operation status data, and the IMU attitude update data to output a standardized fault semantic category. This includes: after the surfacing communication intention triggers the vehicle to surface to the communication depth and deploy the communication mast, and activates the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data, and IMU attitude update data are collected; based on the communication... The signal-to-noise ratio (SNR) mean, fluctuation amplitude, and duration of anomalies are extracted from the satellite communication quality data. The driving current mutation amplitude, folding angle hysteresis, and angle change rate are extracted from the communication folding mast operation status data. The disturbance main frequency, peak amplitude, and frequency band distribution are extracted from the IMU attitude update data through time-frequency transformation. The SNR mean, fluctuation amplitude, duration of anomalies, driving current mutation amplitude, folding angle hysteresis, angle change rate, disturbance main frequency, peak amplitude, and frequency band distribution are input into a pre-constructed time-series feature extraction network to obtain standardized fault semantic categories. The standardized fault semantic categories include at least one of wave disturbance, mast mechanism jamming, radio frequency hardware failure, and sea state exceeding limits.

[0046] Specifically, the core network of the communication sensing sub-agent adopts a lightweight temporal feature extraction architecture, with four independent classification output heads at the end, supporting simultaneous identification of single and compound faults. The model parameters are in the hundreds of KB range, supporting integer quantization deployment, and can run on ordinary ARM architecture embedded controllers. A supporting rule-based inference engine covers extreme operating conditions and sensor failure scenarios, automatically switching to rule-based judgment when the network output confidence is insufficient, ensuring the reliability of basic diagnostic capabilities. A sliding time window temporal sampling mechanism is adopted, with a window duration of 5-10 seconds and a step size of 1 second for continuous sliding. The sampling frequency is 1-2Hz for satellite communication, 3-10Hz for mast status, 10-20Hz for IMU attitude, and 1-3Hz for depth information.

[0047] The core judgment logic for the four types of faults is as follows: Wave disturbance manifests as periodic and regular oscillations in the IMU attitude, synchronous periodic fluctuations in the satellite signal-to-noise ratio, and no abnormal offset or hysteresis in the mast drive current and folding angle; Mast mechanism jamming manifests as a sudden and continuous increase in the mast drive current, a slow change in the folding angle, a monotonically decreasing trend in the signal-to-noise ratio, and no corresponding large disturbances in the hull attitude; Radio frequency hardware faults manifest as a sudden drop in the signal-to-noise ratio with no recovery trend, a sharp increase in the link packet loss rate, and the mast operating parameters and hull attitude parameters are all within the normal operating range; Exceeding sea state limits manifests as a sustained excessive IMU roll amplitude, significant depth fluctuations, a sustained signal-to-noise ratio below the receiving threshold, and no typical characteristics of a single fault.

[0048] For conflicts of evidence from multiple sources, a confidence-weighted fusion mechanism can be adopted, which assigns corresponding weights according to the priority of satellite communication data, mast status data, IMU attitude data, and depth data, and calculates the final confidence of various types of faults by weighting. When there are conflicts of evidence, a hierarchical strategy of extending the observation window for smoothing, giving priority to hardware status, and using rule-based reasoning as a fallback is adopted to ensure the reliability of the diagnosis.

[0049] In this embodiment, in step S105, a high-level semantic intent instruction is output based on the standardized fault semantic category and the priority information of the currently executed task. This high-level semantic intent instruction is then converted into a low-level control instruction. The low-level control instruction drives the vehicle's low-level execution controller to coordinate and regulate at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, using the satellite communication link quality evaluation function as the optimization target. This completes communication quality-oriented attitude adaptive control, including: the planning and reflection sub-agent queries a preset fault-intent mapping relationship based on the standardized fault semantic category to determine candidate high-level semantic intents. The high-level semantic intent includes communication optimization intent, attitude control intent, and hazard avoidance termination intent. Combining the priority information of the current task, the candidate high-level semantic intents are arbitrated in descending order of priority, and a high-level semantic intent instruction is output. The attitude intent sub-agent uses a semantic intent-execution strategy mapping mechanism to convert the high-level semantic intent instruction into a low-level control instruction. The low-level control instruction drives the vehicle's low-level execution controller to coordinate and control at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, using the satellite communication link quality evaluation function as the optimization target, to complete communication quality-oriented adaptive attitude control.

[0050] Specifically, the priorities from highest to lowest are: risk avoidance termination intent > attitude control intent > communication optimization intent; when a high-priority intent is triggered, low-priority instructions are automatically blocked, and when risk avoidance termination is triggered, attitude optimization instructions are immediately reset to avoid action conflicts.

[0051] In this embodiment, the attitude intention sub-agent employs a semantic intent-to-execution strategy mapping mechanism to convert the high-level semantic intent instruction into low-level control instructions. This includes: when the high-level semantic intent instruction is a communication optimization intent, the attitude intention sub-agent generates low-level control instructions for controlling the full-range coordinated control of the buoyancy adjustment mechanism, the center of gravity adjustment mechanism, the communication folding mast mechanism, and the attitude compensation propulsion mechanism; when the high-level semantic intent instruction is an attitude control intent, the attitude intention sub-agent generates low-level control instructions for prioritizing the control of the buoyancy adjustment mechanism and the center of gravity adjustment mechanism to complete attitude correction; and when the high-level semantic intent instruction is a hazard avoidance termination intent, the attitude intention sub-agent generates low-level control instructions solely for controlling the retraction of the communication folding mast mechanism and the descent of the buoyancy adjustment mechanism.

[0052] The satellite communication link quality evaluation function is expressed as follows:

[0053] In the formula, This represents the quality evaluation value of the satellite communication link. The signal-to-noise ratio (SNR) extracted from the satellite communication quality data is the normalized value. This is the normalized value of the link packet loss rate extracted from the satellite communication quality data. and These are the weighting coefficients. .

[0054] For example, It is 0.6. The value is 0.4, which can be dynamically adjusted according to task requirements. An example adjustment range is... Take a value of 0.5 to 0.8. Take a value of 0.2~0.5, and The control cycle is aligned with the front-end sampling window, preferably iterating once every 1 second. The convergence criterion is within three consecutive cycles. If the value fluctuation is ≤5% and higher than the qualified threshold, it is determined that the steady-state optimum has been reached.

[0055] In this embodiment, the four types of actuators work collaboratively in a coarse-to-fine adjustment hierarchy. The buoyancy adjustment mechanism is responsible for low-frequency, large-amplitude attitude correction, used to balance the overall static tilt angle and long-period sea state tilt, with the largest adjustment range and slowest response. The center of gravity adjustment mechanism is responsible for medium-to-low frequency, medium-amplitude attitude adjustment, used to compensate for the average tilt of medium-period waves, with moderate response speed and adjustment range. The communication folding mast mechanism is responsible for medium-to-high frequency, small-amplitude, high-precision pointing adjustment, using a composite control of attitude feedforward and signal closed loop. It first compensates for tilt deviation based on the ship's attitude angle, and then tracks and locks the optimal signal pointing with real-time communication quality as the target, making it the core actuator for communication quality optimization. The attitude compensation propulsion mechanism is responsible for high-frequency, small-amplitude disturbance compensation, used to compensate for the instantaneous disturbance of high-frequency waves, with the fastest response and smallest adjustment amount. The total attitude correction amount is decomposed by frequency band and distributed to the corresponding mechanisms, and the final outputs are linearly superimposed, working together to affect the ship's attitude and antenna pointing.

[0056] In this embodiment, a coupling gain modeling scheme for attitude and communication quality can also be adopted, which is mainly based on pre-calibration and supplemented by online interpolation correction. In advance, the signal-to-noise ratio gain of satellite antenna under different hull attitude angles and mast angles is calibrated through pool tests or sea trials to form a reference mapping relationship. During real-time operation, a small-range linear interpolation correction is made on the reference relationship based on the current measured data to compensate for individual hardware differences and environmental offsets, and to ensure closed-loop stability.

[0057] In this embodiment, the underlying execution controller adopts an embedded architecture of a main control processing module and multiple drive channels. The main control processing module uniformly receives the control requirements after the attitude intention sub-agent is converted. After completing motion calculation and instruction allocation, it drives the buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism and attitude compensation propulsion mechanism to coordinate their actions through the corresponding drive channels.

[0058] In this embodiment, step S106, which involves acquiring the full-process data of the current ascent communication event after the current communication window ends and storing it in the local memory, includes: after the current communication window ends, the planning and reflection sub-agent summarizes the decision basis, sea state parameters, execution actions, and fault semantic categories of the current ascent communication event; the satellite communication module provides the communication results; and the perception module provides the energy consumption data. The decision basis, sea state parameters, execution actions, fault semantics, communication results, and energy consumption data of the current ascent communication event are then integrated into the full-process data of the current ascent communication event in a structured case format. The integrated full-process data is then stored in the local memory.

[0059] In this embodiment, the structured case format includes environmental feature fields, task constraint fields, decision action fields, and effect result fields. The environmental feature fields include sea area grid code, sea state level, and wave height range; the task constraint fields include task type, remaining battery percentage, and priority level; the decision action fields include final decision type, delay duration, and control intent mode, where the control intent mode includes at least one of communication optimization intent, attitude control intent, and risk avoidance termination intent; the effect result fields include communication success rate, average signal-to-noise ratio, fault semantic category, and task completion degree.

[0060] In this embodiment, the local memory adopts a two-layer structure with a structured case library at its core and an event-level pipeline as backup. The structured case library, based on a single surface communication event, is extracted and archived offline by the planning and reflection sub-agent, and stored in a structured manner according to fixed fields for real-time retrieval and experience recall. The event-level pipeline stores the original sensor data and execution logs of the entire link in real time according to timestamps, and is only used for offline updates of the case library and parameter calibration correction, and does not participate in real-time retrieval, in order to control storage and computing overhead. After surfacing and connecting to the satellite link, the memory data can be selectively synchronized to shore-based equipment to complete cloud backup.

[0061] In this embodiment, historical experience is embedded in the multi-factor comprehensive scoring model of the planning and reflection sub-agent in three lightweight forms to achieve positive feedback from experience to decision-making. First, there is a decision weighting factor, which uses the historical communication success rate in the same scenario as a quantitative input for the historical success rate dimension. The weighting coefficients for sea state, power consumption, and task dimensions can be fine-tuned within a preset range based on the historical effect distribution to adapt to specific sea area characteristics. Second, there is a decision threshold correction term, which dynamically fine-tunes the surfacing or delaying scoring threshold based on historical case results. When the historical success rate is low, the threshold is increased, and the strategy becomes more conservative; conversely, the threshold is decreased, and the strategy becomes more aggressive. The threshold correction is limited to a preset safety range. Third, there is an auxiliary parameter reference, where the average delay time and optimal attitude control baseline parameters from historical cases can be directly used as the initial recommended values ​​for this decision, reducing iterative computation and accelerating convergence.

[0062] In this embodiment, the satellite communication link quality is continuously monitored in a closed loop. A tiered handling strategy is implemented based on different operational scenarios, including: if communication quality recovers to the required standard after attitude compensation and equipment fine-tuning, the communication task is successfully completed; if satellite communication quality still fails to reach the communication threshold after multiple attitude compensations and equipment fine-tunings, it is determined that the current sea state exceeds the equipment's attitude control capability, and the current communication window is proactively terminated; if the communication sensing sub-agent identifies irreversible hardware anomalies such as severe mast jamming or radio frequency hardware failure, tiered handling is implemented based on task priority. The decision-making basis, sea state conditions, actions performed, communication results, root causes of failures, and energy consumption data for the current ascent window are all written into the local memory, completing the accumulation of experience from this operation; subsequent tasks in the same sea area and sea state will directly reuse this experience, enabling autonomous iterative optimization of decision-making strategies and forming a complete closed loop.

[0063] The underwater unmanned vehicle communication control method in this embodiment collects multi-dimensional sensing data such as IMU attitude, navigation depth, and remaining battery power when the underwater communication window is triggered. It combines the current location with historical communication event data from the same sea area, sea state level, and wave height range in the local memory to perform real-time sea state assessment and multi-factor comprehensive scoring. This allows for dynamic judgment of the timing of surfacing for communication, avoiding satellite search failures, link interruptions, and unnecessary energy consumption caused by blindly surfacing in adverse sea conditions. When surfacing is not suitable, it improves decision-making flexibility and safety by using underwater acoustic feedback or entering the next communication window for cyclical reassessment. After establishing a link upon surfacing, it performs semantic reasoning on satellite communication quality data, communication folding mast operation status data, and IMU attitude update data to output standardized fault semantic categories, thereby distinguishing between wave disturbances, mast jamming, radio frequency hardware failures, or other issues. The system identifies abnormal root causes such as exceeding sea state limits, providing a basis for differentiated handling. Furthermore, it combines standardized fault semantic categories with task priorities to generate high-level semantic intent commands, which are then converted into low-level control commands. Using the satellite communication link quality evaluation function as the optimization target, it coordinates and controls the buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism. This ensures that attitude control directly serves communication quality rather than merely pursuing zero attitude angles, effectively solving the problem of acceptable attitude but poor communication quality. Finally, after the communication window ends, the entire process data is saved to the local memory, forming a closed loop of perception, decision-making, execution, evaluation, and review. This allows for continuous accumulation of experience and autonomous optimization of subsequent decisions, thereby comprehensively improving the success rate of underwater unmanned vehicle surfacing communication, environmental adaptability, energy efficiency, and long-term operational autonomy.

[0064] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A communication control method for an underwater unmanned vehicle, characterized in that, include: When the vehicle is in underwater navigation mode, when it reaches the preset periodic communication time window, it collects multi-dimensional sensing data, including IMU attitude data, navigation depth data and remaining power data. The current position of the vehicle is obtained to determine the current sea area. Based on the IMU attitude data and the navigation depth data, the sea state is evaluated in real time to obtain the sea state level and wave height range. The historical communication event data with the same sea state level and wave height range in the current sea area are retrieved from the preset local memory database using the sea state level and wave height range as search conditions. Obtain the priority information of the currently executing task, combine the sea state level, the remaining power data, the priority information of the currently executing task, and the historical communication event data to perform a multi-factor comprehensive score, and output the intention to ascend when the comprehensive score reaches the ascending judgment threshold, and output the intention to postpone ascending when the comprehensive score does not reach the ascending judgment threshold, and trigger underwater acoustic feedback or enter the next communication window for loop reassessment. After the surfacing communication intention triggers the vehicle to surface to the communication depth and deploy the communication mast, and activate the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data and IMU attitude update data are collected. Based on the satellite communication quality data, the communication folding mast operation status data and the IMU attitude update data, semantic reasoning is performed to output a standardized fault semantic category. Based on the standardized fault semantic category and the priority information of the current task, a high-level semantic intent instruction is output, and the high-level semantic intent instruction is converted into a low-level control instruction. The low-level control instruction drives the low-level execution controller of the vehicle to coordinate and regulate at least one of the following: the buoyancy adjustment mechanism, the center of gravity adjustment mechanism, the communication folding mast mechanism, and the attitude compensation propulsion mechanism, with the satellite communication link quality evaluation function as the optimization target, to complete the attitude adaptive control guided by communication quality. After the current communication window ends, acquire the full process data of the current floating communication event and save it in the local memory.

2. The underwater unmanned vehicle communication control method according to claim 1, characterized in that, When the vehicle is in underwater navigation mode, upon reaching a preset periodic communication time window, multi-dimensional sensing data is collected, including: When the vehicle is in underwater navigation mode, when it reaches the preset periodic communication time window and the vehicle has sailed to the preset communication depth, the sensing module collects multi-dimensional sensing data, which includes IMU attitude data, navigation depth data and remaining power data. The multi-dimensional perception data is input into the AI ​​agent module for communication control. The AI ​​agent module includes a planning and reflection sub-agent, a communication and perception sub-agent, and a posture and intention sub-agent.

3. The underwater unmanned vehicle communication control method according to claim 2, characterized in that, The process involves acquiring the vehicle's current position to determine the current operational sea area, performing real-time sea state assessment based on the IMU attitude data and the navigation depth data to obtain the sea state level and wave height range, and retrieving historical communication event data within the current operational sea area of ​​the same sea state level and wave height range from a preset local memory database using the sea state level and wave height range as search criteria. This includes: The planning and reflection sub-agent performs frequency domain transformation on the IMU attitude data and takes the frequency corresponding to the maximum frequency of the spectrum amplitude as the main oscillation frequency. Statistically analyze the peak value of attitude amplitude in the IMU attitude data and the variance of depth fluctuation in the navigation depth data; Based on the pre-calibrated relationship between hull motion and wave height, the peak value of the attitude amplitude, the variance of the depth fluctuation, and the dominant frequency are converted into the wave height value. The sea state level is obtained by querying the preset local memory database mapping table of wave height and sea state level according to the sea surface wave height value, and the wave height range is determined according to the sea surface wave height value. The sea state level includes level 1 light waves, level 2 moderate waves, level 3 large waves and level 4 giant waves. The current position of the vehicle is obtained to determine the current operating sea area. Using the sea state level and the wave height range as search conditions, historical communication event data with the same sea state level and the same wave height range in the current operating sea area are retrieved from the preset local memory.

4. The underwater unmanned vehicle communication control method according to claim 2, characterized in that, The process of obtaining the priority information of the currently executing task, combining the sea state level, the remaining battery power data, the priority information of the currently executing task, and the historical communication event data to perform a multi-factor comprehensive score, and outputting an surfacing communication intention when the comprehensive score reaches the surfacing determination threshold, and outputting a delayed surfacing intention when the comprehensive score does not reach the surfacing determination threshold, and triggering underwater acoustic echo or entering the next communication window for re-evaluation, includes: The planning and reflection sub-agent reads the type identifier of the currently executing task from the vehicle's task scheduling system; According to the preset task type and priority mapping relationship, the priority information of the currently executing task is determined. The task types include emergency tasks, routine tasks and inspection tasks. A multi-factor comprehensive score is performed by combining the sea state level, the remaining battery power data, the priority information of the currently executed task, and the historical communication event data; When the comprehensive score reaches the upward judgment threshold, the upward communication intention is output. When the comprehensive score does not reach the upward judgment threshold, the upward intention is delayed, and underwater acoustic feedback is triggered or the next communication window is entered for re-evaluation.

5. The underwater unmanned vehicle communication control method according to claim 4, characterized in that, The multi-factor comprehensive scoring, which combines the sea state level, remaining battery power data, priority information of the currently executing task, and historical communication event data, includes: The sea state levels are reverse normalized according to their order of importance to obtain the sea state factor. The remaining power data is linearly normalized according to the percentage of remaining power to rated capacity to obtain the power factor; The priority information of the currently executing task is discretely assigned according to the preset task level to obtain the task priority factor; The historical surface communication success rate under the same sea area and sea state level is extracted from the historical communication event data and used as a historical success rate factor. The sea state factor, the power factor, the task priority factor, and the historical success rate factor are weighted and summed according to preset weights to obtain a multi-factor comprehensive score.

6. The underwater unmanned vehicle communication control method according to claim 2, characterized in that, After the surfacing communication intention triggers the vehicle to surface to the communication depth, deploys the communication mast, and activates the satellite communication module to establish a link, satellite communication quality data, communication mast folding status data, and IMU attitude update data are collected. Semantic reasoning is then performed based on the satellite communication quality data, the communication mast folding status data, and the IMU attitude update data to output standardized fault semantic categories, including: After the surfacing communication intention triggers the vehicle to surface to the communication depth and deploy the communication mast, activate the satellite communication module to establish a link, satellite communication quality data, communication folding mast operation status data, and IMU attitude update data are collected. The communication sensing sub-agent extracts the mean signal-to-noise ratio, fluctuation amplitude, and duration of continuous anomalies from the satellite communication quality data; extracts the amplitude of sudden changes in driving current, hysteresis of folding angle, and rate of angle change from the communication folding mast operation status data; and extracts the main frequency of disturbance, peak amplitude, and frequency band distribution from the IMU attitude update data through time-frequency transformation. The mean signal-to-noise ratio, fluctuation amplitude, duration of continuous abnormality, amplitude of sudden change in driving current, hysteresis of folding angle, rate of change of angle, main frequency of disturbance, peak amplitude and frequency band distribution are input into a pre-constructed time-series feature extraction network to obtain standardized fault semantic categories. The standardized fault semantic categories include at least one of wave disturbance, mast mechanism jamming, radio frequency hardware failure and sea state exceeding limits.

7. The underwater unmanned vehicle communication control method according to claim 2, characterized in that, The process involves outputting high-level semantic intent commands based on the standardized fault semantic categories and the priority information of the currently executing task, converting these commands into low-level control commands, and using these commands to drive the vehicle's low-level execution controller to coordinate and regulate at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, with the satellite communication link quality evaluation function as the optimization target. This achieves communication quality-oriented adaptive attitude control, including: The planning reflection sub-agent queries the preset fault and intent mapping relationship based on the standardized fault semantic category to determine candidate high-level semantic intents, which include communication optimization intents, attitude control intents, and risk avoidance termination intents. Based on the priority information of the currently executing task, the candidate high-level semantic intents are arbitrated in descending order of priority, and a high-level semantic intent instruction is output. The attitude intention sub-agent uses a semantic intention-to-execution-policy mapping mechanism to convert the high-level semantic intention instructions into low-level control instructions; The underlying control commands drive the vehicle's underlying execution controller to coordinate and regulate at least one of the vehicle's buoyancy adjustment mechanism, center of gravity adjustment mechanism, communication folding mast mechanism, and attitude compensation propulsion mechanism, with the satellite communication link quality evaluation function as the optimization target, to complete the communication quality-oriented adaptive attitude control.

8. The underwater unmanned vehicle communication control method according to claim 7, characterized in that, The gesture intent sub-agent uses a semantic intent-to-execution-policy mapping mechanism to convert the high-level semantic intent instructions into low-level control instructions, including: When the high-level semantic intent instruction is a communication optimization intent, the attitude intent sub-agent generates a low-level control instruction for the full-range coordinated control of the buoyancy adjustment mechanism, the center of gravity adjustment mechanism, the communication folding mast mechanism, and the attitude compensation propulsion mechanism. When the high-level semantic intent instruction is an attitude control intent, the attitude intent sub-agent generates a low-level control instruction for prioritizing the control of the buoyancy adjustment mechanism and the center of gravity adjustment mechanism to complete attitude correction. When the high-level semantic intent instruction is a risk avoidance termination intent, the attitude intent sub-agent generates low-level control instructions that are only used to control the retraction of the communication folding mast mechanism and the diving of the buoyancy adjustment mechanism.

9. The underwater unmanned vehicle communication control method according to claim 8, characterized in that, The satellite communication link quality evaluation function is expressed as follows: In the formula, This represents the quality evaluation value of the satellite communication link. The signal-to-noise ratio (SNR) extracted from the satellite communication quality data is the normalized value. This is the normalized value of the link packet loss rate extracted from the satellite communication quality data. and These are the weighting coefficients. .

10. The underwater unmanned vehicle communication control method according to claim 7, characterized in that, After the current communication window ends, acquire the complete process data of the current floating communication event and save it in the local memory, including: After the current communication window ends, the planning reflection sub-agent summarizes the decision basis, sea state parameters, execution actions and fault semantic categories of the current surfacing communication event, the satellite communication module provides the communication results, and the perception module provides energy consumption data. The decision-making basis, sea state parameters, execution actions, fault semantics, communication results, and energy consumption data of the current ascent communication event are integrated into the full-process data of the current ascent communication event in a structured case format. The structured case format includes environmental feature fields, task constraint fields, decision action fields, and effect result fields. The environmental feature fields include sea area grid code, sea state level, and wave height range. The task constraint fields include task type, remaining power percentage, and priority level. The decision action fields include final decision type, delay duration, and control intention mode. The effect result fields include communication success rate, average signal-to-noise ratio, fault semantic category, and task completion degree. The integrated end-to-end data is stored in the local memory.