Intelligent thermal management energy-saving method for underwater robot

Abstract

The invention provides an underwater robot intelligent thermal management energy-saving method, which comprises the following steps: acquiring temperature data and an operation load peak value of a processor in real time through a heat-sensitive sensor arranged in an underwater robot cluster, generating a thermal state vector group, and storing the thermal state vector group in a distributed data node to obtain an initial thermal state record set; analyzing task queue congestion conditions of other underwater robots based on the task capability limited list through a task transfer mechanism under the support of a cluster communication bandwidth, redistributing affected tasks, generating a task takeover distribution table, and determining a target takeover underwater robot combination; and according to the task takeover distribution table, the navigation trajectory of the target takeover robot is recalculated in combination with the water flow resistance change of the underwater environment and the terrain shielding coefficient, path overlapping conflicts are avoided, and an adjusted navigation trajectory group is obtained.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an intelligent thermal management and energy-saving method for underwater robots. Background Technology

[0002] Underwater robot swarm collaboration is a crucial technological means for marine exploration and resource development, playing an indispensable role in deep-sea exploration, environmental monitoring, and engineering operations. Its research is of key significance for enhancing technological capabilities in the marine field. However, current solutions for swarm collaboration often overlook the dynamic nature of thermal management and the multi-robot interaction in complex underwater environments. Especially in extreme environments, changes in the thermal state of a single robot can have a cascading impact on the operational efficiency of the entire swarm, and existing methods lack systematic strategies to address this issue. In this field, the core challenge lies in the real-time sharing and dynamic adjustment of thermal state information among robots in the swarm. Due to the high pressure and low temperature characteristics of the underwater environment, the heat generated during robot operation is difficult to dissipate effectively. Once a robot faces the risk of thermal overload, its operational capability will decline in a stepwise manner, affecting not only individual performance but also interfering with the overall task allocation and path planning of the swarm. A deeper problem is that when the capability of one robot is limited, other robots need to take over the task promptly. The dynamic adjustment of this takeover strategy involves complex coordination among multiple robots; failure to adapt quickly to changes will lead to delays or a decline in overall operational progress and quality. These two factors are closely related; the sharing of thermal status information directly determines the response speed and rationality of task takeover. Therefore, how to construct an effective distributed thermal management mechanism in underwater robot swarm collaboration, ensure the real-time sharing of thermal status information, and guarantee the overall operational progress and quality of the swarm through dynamic path replanning and adjustment of task takeover strategies, has become a key issue that this research urgently needs to address. Summary of the Invention

[0003] This invention provides an intelligent thermal management and energy-saving method for underwater robots, mainly comprising:

[0004] The underwater robot cluster uses built-in thermal sensors to collect processor temperature data and peak operating load in real time, generate thermal state vector sets, store them in distributed data nodes, and obtain an initial thermal state record set.

[0005] Based on the initial thermal state record set, broadcast the thermal state vector group of each underwater robot. If the temperature value of an underwater robot exceeds the preset processor overheat threshold or the duration of thermal overload exceeds the safe range, it is marked as a thermal risk unit, and a list of thermal risk units is obtained.

[0006] For the thermal risk unit list, analyze the task interruption probability of each affected robot, combine the thermal state fluctuation frequency and the peak operating load, calculate the ratio of the current task load to the efficiency of the heat dissipation module, determine the decline in task execution capability, and obtain the task capability limitation list.

[0007] By using the task transfer mechanism supported by cluster communication bandwidth, the task queue congestion of other underwater robots is analyzed based on the task capability-limited list, the affected tasks are reallocated, a task takeover allocation table is generated, and the target takeover underwater robot combination is determined.

[0008] Based on the task takeover assignment table, and taking into account the changes in water flow resistance and terrain obstruction coefficient in the underwater environment, the navigation trajectory of the target takeover robot is recalculated to avoid path overlap and conflict, resulting in an adjusted navigation trajectory set.

[0009] The task takeover assignment table and the adjusted navigation trajectory group are sent to the target takeover robot group. The task execution sequence of each robot is updated synchronously. The thermal status data delay and temperature change of each robot during task execution are monitored in real time. If the temperature value of a robot approaches the processor overheating threshold again or the thermal status fluctuation frequency is abnormal, the thermal load control process is triggered and a thermal load control instruction is generated.

[0010] The peak operating load of the affected robots is dynamically adjusted according to the thermal load control command, the processor frequency is reduced or secondary tasks are paused, and the adjusted operating state combination is obtained. The overall task priority deviation and takeover delay of the cluster are analyzed. By comparing the current task progress with the preset target, the task progress correction command is generated.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses an intelligent thermal management and energy-saving method for underwater robots. It generates and broadcasts thermal state vector groups by real-time acquisition of processor temperature and load data, identifying thermally risky units. For robots at thermal risk, it analyzes the probability of task interruption and the extent of performance degradation, generating a list of tasks with limited capabilities. Based on this list, affected tasks are reallocated, a task takeover allocation table is generated, and the navigation trajectory of the target takeover robot is calculated. A multi-machine collaborative response protocol is employed to distribute task and trajectory information and monitor thermal status in real time. When the overheating threshold is approached again, thermal load regulation is triggered, dynamically adjusting the operating load, reducing processor frequency, or pausing secondary tasks. Finally, a task progress correction instruction is generated, determining the execution blueprint for cluster collaborative operations, realizing thermal management and dynamic task scheduling of the underwater robot cluster, ensuring system stability and task continuity. Attached Figure Description

[0013] Figure 1This is a flowchart of an intelligent thermal management and energy-saving method for underwater robots according to the present invention. Detailed Implementation

[0014] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0015] like Figure 1 This embodiment of an intelligent thermal management energy-saving method for underwater robots may specifically include:

[0016] S101. Using the thermal sensors built into the underwater robot cluster, processor temperature data and peak operating load are collected in real time, generating a thermal state vector set, which is stored in the distributed data nodes to obtain the initial thermal state record set.

[0017] In an underwater robot swarm, each robot's thermal sensor acquires processor core temperature data at a preset sampling frequency. Simultaneously, it reads the current load percentage from the processor status register. Ambient water temperature parameters are collected via a water temperature sensor integrated into the robot's shell. The difference between the processor temperature and the ambient water temperature is calculated and divided by the difference between the processor's rated maximum temperature and the ambient water temperature to obtain the heat dissipation efficiency coefficient. This yields a raw data sequence containing timestamps, temperature values, load values, and the heat dissipation efficiency coefficient. A fixed-size sliding window is set for the raw data sequence. The temperature difference between two adjacent sampling points within the window is divided by the sampling time interval to obtain the temperature change rate, and the load difference between two adjacent sampling points is divided by the sampling time interval to obtain the load change rate. A Fast Fourier Transform is performed on the temperature data within the window, and the frequency component with the largest amplitude is taken as the dominant frequency. If the dominant frequency exceeds a preset threshold, it is marked as an abnormal fluctuation state. A four-dimensional thermal state vector is constructed based on the current temperature value, load value, heat dissipation efficiency coefficient, and dominant frequency. Each robot node broadcasts its own four-dimensional thermal state vector to neighboring nodes through a built-in acoustic modem. After receiving the thermal state vector of a neighboring node, it determines the location of the storage node. Each storage node stores the thermal state vector as a value in its local memory using a timestamp as the primary key, forming a distributed thermal state record set arranged in chronological order.

[0018] In one possible implementation, the underwater robot's thermal sensor uses a platinum resistance temperature probe, whose resistance changes linearly with temperature. This resistance change is converted into a voltage signal via a Wheatstone bridge circuit. The processor status register stores information such as the number of currently running tasks and CPU utilization. By reading these registers, a load percentage value between 0 and 100 can be obtained. The heat dissipation efficiency coefficient is calculated based on the principle of heat conduction. When the processor temperature is 85 degrees Celsius and the ambient water temperature is 15 degrees Celsius, the actual temperature difference is 70 degrees Celsius. The theoretical maximum temperature difference between the processor's rated maximum temperature of 95 degrees Celsius and the water temperature is 80 degrees Celsius. Therefore, the heat dissipation efficiency coefficient is 70 divided by 80, resulting in 0.875. This coefficient reflects the heat transfer efficiency under the current cooling conditions.

[0019] Specifically, sliding window analysis uses a data window of fixed time length, such as a window containing 128 sampling points, and slides forward one sampling point at a time. The rate of temperature change is calculated by the difference between adjacent sampling points. If the temperature at the nth sampling point is 85.2 degrees Celsius, the temperature at the (n+1)th sampling point is 85.5 degrees Celsius, and the sampling interval is 0.1 seconds, then the rate of temperature change is 3 degrees Celsius per second. The Fast Fourier Transform (FFT) converts the time-domain temperature sequence to the frequency domain, identifying the periodic characteristics of temperature fluctuations. When a dominant frequency component above 0.5 Hz is detected, it indicates the presence of abnormal temperature oscillations, which may be caused by sudden changes in processor load or cooling system failure.

[0020] For example, the application of consistent hashing algorithms in distributed storage ensures uniform data distribution and high availability. Each robot node is mapped to a virtual hash ring, and the hot state vector is used to calculate a hash value through a hash function. Then, the nearest node is searched clockwise as the storage location. When a node fails, the data it was responsible for is automatically migrated to the next node, ensuring the system's fault tolerance. Acoustic modems utilize underwater sound waves to transmit data. Digital signals are modulated onto an acoustic carrier wave using frequency-shift keying or phase-shift keying. The receiving end demodulates to recover the original data. Although this communication method has limited bandwidth, it offers long-distance transmission and is stable and reliable in underwater environments.

[0021] S102. Based on the initial thermal state record set, broadcast the thermal state vector group of each underwater robot. If the temperature value of an underwater robot exceeds the preset processor overheat threshold or the duration of thermal overload exceeds the safe range, it is marked as a thermal risk unit, and a list of thermal risk units is obtained.

[0022] The thermal state vector group data of each robot node is read from the initial thermal state record set. Using a low-latency communication protocol within the cluster, the identifier, current temperature value, load value, and timestamp information of each robot are encapsulated into communication data frames. These data frames are sent to all nodes in the cluster at fixed time intervals. Each receiving node parses the data frame content and stores it in its local memory, resulting in a real-time synchronized cluster thermal state dataset. For this dataset, the temperature value of each robot is compared with a preset processor overheating threshold. If the temperature value exceeds the threshold, the robot identifier and overheating start time are recorded. The duration of thermal overload is calculated by subtracting the overheating start time from the current time. If the duration exceeds a preset safe duration, a thermal risk flag is set to true for that robot record. All robot records with a true thermal risk flag are iterated through, and their identifier, current temperature value, and thermal overload duration information are extracted. A risk value is calculated based on the product of the temperature exceeding the threshold and the duration. The risk values ​​are sorted from largest to smallest to generate a thermal risk unit list containing the robot identifier, temperature value, duration, and risk value.

[0023] For example, in one possible implementation, the low-latency communication protocol within the cluster employs a time-division multiple access (TDMA) mechanism, where each robot node is allocated a fixed time slot for data transmission. The encapsulation process of the communication data frame includes three parts: a frame header, a data field, and a checksum. The frame header contains an 8-bit robot identifier and a 16-bit sequence number. The data field stores a 32-bit floating-point temperature value, an 8-bit integer load percentage value, and a 32-bit timestamp. Finally, a 16-bit cyclic redundancy check (CRC) code is appended. This structural design ensures the integrity of data transmission while keeping the length of each frame within 128 bits, thus helping to reduce transmission latency.

[0024] Specifically, the setting of the fixed time interval needs to comprehensively consider the rate of thermal state change and communication bandwidth limitations. When the sampling period is 100 milliseconds, the broadcast interval is set to 500 milliseconds. This not only reflects the temperature change trend in a timely manner but also avoids network congestion caused by frequent communication. When parsing data frames, each receiving node first verifies the correctness of the checksum, then extracts each field data and stores it in a circular buffer in its local memory. The buffer size is set to the most recent 1000 records, and it is updated using a first-in, first-out (FIFO) method, thus forming complete historical thermal state data within the sliding time window.

[0025] For example, the processor overheat threshold is set based on the device's thermal design power parameters, typically set to 85% of the highest junction temperature. When a robot's processor temperature reaches 92 degrees Celsius, and the overheat threshold is set to 90 degrees Celsius, the overheating start time is immediately recorded. The duration of thermal overload is calculated using a real-time clock and updated every second. The safe duration range is determined based on the processor's thermal capacity and heat dissipation capability, generally set between 30 and 180 seconds. When the duration exceeds the lower limit of 30 seconds, the thermal risk flag is set to a true value of 1, indicating that the robot has entered a thermal risk state. The risk value calculation comprehensively considers both the degree of temperature exceeding the threshold and the duration. The magnitude of temperature exceeding the threshold directly reflects the severity of thermal stress, while the duration reflects the cumulative effect of thermal damage. Specifically, if the temperature exceeds the threshold by 5 degrees Celsius and the duration is 60 seconds, the risk value is 5 multiplied by 60, which equals 300. This calculation method allows for reasonable risk assessments for both short-term high temperatures and long-term medium temperatures. The generated list of hot risk units is arranged in descending order of risk value, which enables the cluster management module to prioritize high-risk nodes and realize hierarchical management and precise intervention of hot risks.

[0026] S103. For the thermal risk unit list, analyze the task interruption probability of each affected robot, combine the thermal state fluctuation frequency and the peak operating load, calculate the ratio of the current task load to the efficiency of the heat dissipation module, determine the decline in task execution capability, and obtain the task capability limitation list.

[0027] The system reads the temperature value, duration, and identifier data of each risk robot from the thermal risk unit list. It then counts the number of times the robot's tasks were paused due to temperature exceeding limits within a preset time window, divides this number by the total number of tasks executed during the same period, and calculates the task interruption probability. Simultaneously, it extracts the temperature fluctuation frequency data and peak operating load data of the corresponding robot from the initial thermal state record set, resulting in a comprehensive status data table containing robot identifier, interruption probability, fluctuation frequency, and peak load. For each record in the comprehensive status data table, it reads the robot's current task load percentage value. Based on the temperature difference between the processor and the ambient water temperature and the preset thermal conductivity coefficient, it calculates the current heat dissipation rate. The current task load percentage value is divided by the ratio of the heat dissipation rate to the rated heat dissipation rate to obtain the load efficiency ratio. If the ratio exceeds a preset threshold, the robot is determined to have a risk of limited task execution capability. The percentage decrease in task execution capability is calculated by multiplying the load efficiency ratio and the task interruption probability. Combined with the fluctuation frequency, the capability limitation level is determined. The robots are then sorted from highest to lowest based on the decrease in capability, generating a task capability limitation list containing robot identifier, percentage decrease, and capability limitation level.

[0028] In one possible implementation, the probability of task interruption is calculated based on a statistical method using a sliding time window. The preset time window is the most recent 24 hours, a duration that reflects recent operational status while accumulating sufficient sample data. If a robot performs 120 tasks in the past 24 hours, and 18 of those tasks are forced to pause due to processor temperature exceeding a safety threshold, then the probability of task interruption is 18 divided by 120, which equals 0.15, or a 15% interruption rate. This statistical method based on actual operational data more accurately reflects the robot's real-world working state than theoretical models.

[0029] Specifically, the frequency of temperature fluctuations reflects the stability of the processor's temperature. Spectral analysis of the temperature time series identifies the main fluctuation periods. A fluctuation frequency of 0.2 Hz means a temperature peak occurs every 5 seconds; this periodic temperature change is usually related to task scheduling patterns. Peak load records the highest load percentage within the statistical period; for example, if a robot's peak load reaches 95%, it indicates that its processor is running at almost full capacity, under immense heat dissipation pressure. The thermal conductivity coefficient is determined based on Newton's law of cooling, reflecting the heat exchange efficiency between the processor and the cooling medium. In an underwater environment, the thermal conductivity coefficient is affected by water flow velocity, water temperature, and the robot's shell material. The heat dissipation rate is calculated by multiplying the processor power consumption by the thermal conductivity coefficient and then by the temperature difference.

[0030] For example, when the processor power consumption is 50 watts, the temperature difference is 30 degrees Celsius, and the thermal conductivity is 0.8, the actual heat dissipation rate is 1200 watts per degree Celsius. The rated heat dissipation rate is the maximum heat dissipation capacity measured under standard test conditions, typically 1500 watts per degree Celsius.

[0031] For example, the calculation process of the load efficiency ratio reflects the balance between task requirements and heat dissipation capacity. If the current task load percentage is 80%, and the ratio of the actual heat dissipation rate to the rated heat dissipation rate is 0.8, then the load efficiency ratio is 80 divided by 0.8, which equals 100. When this ratio exceeds a preset threshold of 85, it indicates that the heat dissipation capacity can no longer meet the current task load requirements, and the robot faces the risk of overheating. The determination of the capability limitation level comprehensively considers two factors: the magnitude of the performance drop and the frequency of fluctuation. The magnitude of the performance drop reflects the degree of performance loss, while the frequency of fluctuation reflects the instability of the system. When the magnitude of the performance drop exceeds 30% and the frequency of fluctuation is higher than 0.3 Hz, it is classified as severely limited; a magnitude of 15% to 30% is moderately limited; and a magnitude of less than 15% is slightly limited. This hierarchical management approach enables the cluster scheduling system to adopt corresponding task redistribution strategies for different degrees of severity.

[0032] S104. Through the task transfer mechanism supported by cluster communication bandwidth, analyze the congestion status of other underwater robots' task queues based on the task capability-limited list, reallocate affected tasks, generate a task takeover allocation table, and determine the target takeover underwater robot combination.

[0033] Obtain a list of restricted task capabilities for each robot in the cluster, recording the currently unsupported task types and corresponding hardware limitations for each robot. Based on the limitations in the restricted capability list and real-time monitoring data of cluster communication bandwidth, calculate the effective communication capacity between robot nodes to obtain an available bandwidth matrix for task transfer between nodes. Combine the available bandwidth matrix with the real-time task queue length of each robot to calculate the congestion level for each robot. The calculation process is as follows: sum the product of the estimated execution time of each task in the task queue and the priority value of that task, then divide by the total available bandwidth from the corresponding robot to other nodes, generating a robot load status table containing congestion level values ​​and task type distribution. Identify high-load robots exceeding a preset congestion threshold based on the congestion level values ​​in the load status table. Extract tasks affected by capability limitations from their task queues. Combine the task type distribution information in the load status table to match candidate takeover robots with corresponding task execution capabilities and low loads. Construct a task-candidate robot allocation cost matrix, and use the bipartite graph maximum matching method to solve for the optimal allocation scheme, outputting a task takeover allocation table. If the required capacity of a task in the task takeover allocation table exceeds the processing range of a single robot, then the task is decomposed into multiple sub-tasks according to the functional module division of the task. Robots with complementary capabilities are selected from the candidate takeover robots to form a takeover combination. Sub-tasks are allocated according to the dependencies between sub-tasks and the remaining processing capacity of each robot, and the target takeover robot combination and its specific sub-tasks are determined.

[0034] In one possible implementation, a task capability limitation list records the types of tasks that each robot cannot perform due to differences in hardware configuration. For example, some robots may be unable to perform precision assembly tasks due to a lack of high-precision vision sensors, or may be unable to handle complex path planning tasks due to limitations in computing unit performance. This capability difference directly affects task allocation efficiency. When a robot receives a task that is beyond its capabilities, it must transfer the task to other robots with the corresponding capabilities through cluster communication.

[0035] Specifically, the calculation of the available bandwidth matrix for task transfer involves a comprehensive consideration of multiple factors. First, the physical communication bandwidth between each node is obtained. Then, the actual available bandwidth is calculated based on the current network load. If the physical bandwidth from node A to node B is 100Mbps, and 60Mbps is currently used for state synchronization and coordination control, then the remaining 40Mbps can be used for task transfer. This available bandwidth value directly determines the transmission efficiency of large task data packets.

[0036] It should be noted that the congestion level value is calculated by fully considering the time cost and priority weight of task execution. High-priority tasks, even with shorter execution times, have a greater impact on the system. By multiplying the task execution time by the priority value and summing the results, the actual load pressure on the robot can be accurately reflected. Dividing this by the robot's total available bandwidth yields a congestion level value that reflects both the task backlog and the communication capacity limitations for task transfer.

[0037] In one embodiment, the preset congestion threshold is set based on the overall performance requirements of the cluster and task response time constraints. When the congestion level of a robot exceeds the threshold of 0.8, it indicates that its task queue is severely backlogged, and continuing to receive new tasks will cause system response delays. At this time, tasks to be transferred are extracted from the robot's task queue, prioritizing those tasks that cannot be executed efficiently due to capacity limitations.

[0038] For example, when constructing the allocation cost matrix, each element represents the total cost of assigning a specific task to a specific robot. This cost includes not only communication transmission time but also the target robot's current load state and task execution efficiency. The bipartite graph maximum matching method ensures that each task is assigned to the most suitable robot by finding the allocation scheme with the minimum total cost.

[0039] Preferably, when encountering complex tasks requiring multi-robot collaboration, task decomposition follows the principle of functional modularization. A complex object recognition and grasping task can be decomposed into sub-tasks such as image acquisition, feature extraction, pose calculation, and robotic arm control. Robots with high-performance vision systems are responsible for image processing-related sub-tasks, while robots equipped with precision robotic arms undertake the grasping and execution sub-tasks.

[0040] S105. Based on the task takeover assignment table, and taking into account the changes in water flow resistance and terrain obstruction coefficient in the underwater environment, recalculate the navigation trajectory of the target takeover robot to avoid path overlap and conflict, and obtain the adjusted navigation trajectory group.

[0041] Based on the robot number and target position in the task takeover assignment table, real-time water flow velocity field data collected by underwater environmental sensors is obtained. The cosine of the angle between the water flow vector at each depth level and the robot's preset motion direction is calculated, and multiplied by the water flow velocity modulus to obtain the water flow resistance coefficient. Simultaneously, the obstacle distribution of the target area is extracted from the underwater terrain database. Based on the obstacle height and robot sensor position, the occlusion angle range is calculated to determine the terrain occlusion coefficient, resulting in an environmental constraint matrix containing the water flow resistance coefficient and the terrain occlusion coefficient. The starting position and target position of each robot are extracted from the task takeover assignment table to generate an initial navigation path with straight connections. The initial navigation path is corrected using the water flow resistance coefficient in the environmental constraint matrix. If the resistance coefficient of a certain path segment exceeds a preset threshold, the direction of that path segment is adjusted to reduce the angle between it and the water flow direction. The communication blind zone range is determined based on the terrain occlusion coefficient. The position with the smallest occlusion coefficient at the boundary of the blind zone is selected as the relay point, generating a set of candidate navigation trajectories considering environmental constraints. Each trajectory in the candidate navigation trajectory set is discretized into a spatiotemporal coordinate sequence according to the robot's velocity planning. The spatial coordinate distance at the same time in different robot trajectory sequences is detected. If the distance is less than the preset safe distance threshold, it is marked as a potential conflict point. By adjusting the movement speed of the conflicting robot before the conflict point or inserting a waiting time node, the passage time of each robot at that spatial position is staggered. The detection and adjustment are repeated until the minimum distance between all trajectories meets the safety requirements, and the adjusted navigation trajectory group without conflict is determined.

[0042] In one possible implementation, the calculation of the water flow drag coefficient is based on the drag principle in fluid mechanics. When an underwater robot moves at a specific depth, the water flow drag it experiences is closely related to the water flow velocity, the robot's direction of motion, and the angle between the two. The cosine value of the angle reflects the degree of influence of the water flow on the robot's motion. When the robot moves against the current, the angle is 180 degrees, the cosine value is -1, and the drag reaches its maximum; when moving with the current, the angle is 0 degrees, the cosine value is 1, and the water flow provides thrust. By multiplying the cosine value of the angle by the magnitude of the water flow velocity, the resulting water flow drag coefficient can accurately quantify the energy consumption differences when moving in different directions.

[0043] Specifically, determining the terrain occlusion coefficient involves the physical characteristics of sonar and optical sensors. Underwater obstacles such as reefs and shipwrecks can obstruct sensor signals. The calculation of the occlusion angle range takes into account the actual height of the obstacle, the installation position of the robot's sensors, and the relative distance between them. When a robot approaches a reef 10 meters high, if the sensor is 2 meters from the seabed, the occlusion angle at different horizontal distances can be calculated using geometric relationships. This occlusion coefficient directly affects the robot's perception capabilities and communication quality.

[0044] It should be noted that the environmental constraint matrix integrates these two types of key parameters to form a comprehensive description of the underwater environment. Each element in the matrix corresponds to the environmental characteristics of a specific spatial location, enabling subsequent path planning to fully consider the impact of environmental factors.

[0045] In one embodiment, the initial navigation path is generated using the simplest straight-line connection method because a baseline path is needed as a starting point for optimization in complex underwater environments. When the water flow resistance coefficient of a certain path segment exceeds the threshold of 0.7, it indicates that the energy consumption of that path segment is too high. At this time, by adjusting the path direction to reduce the angle between the robot's movement direction and the water flow direction from 90 degrees to 30 degrees, the power consumption of the thrusters can be significantly reduced, extending the robot's operating time.

[0046] For example, handling communication blind spots presents unique challenges for underwater operations. Due to severe attenuation of electromagnetic waves underwater, robots primarily rely on acoustic communication. When the terrain occlusion coefficient indicates the existence of a communication blind spot in a certain area, a relay point is set up at the location with the lowest occlusion coefficient at the boundary of the blind spot. The selection of the relay point not only considers the degree of occlusion but also ensures that the location itself has good communication conditions, enabling it to maintain communication with robots both inside and outside the blind spot simultaneously.

[0047] Preferably, the process of generating the spatiotemporal coordinate sequence discretizes the continuous trajectory into a series of time-stamped spatial points. Each robot, according to its speed plan, records its position coordinates at fixed time intervals. When the spatial distance between the trajectory sequences of two robots is less than a safety threshold of 5 meters at a certain moment, the system identifies a potential collision risk. By having one robot reduce its speed or pause for 2 seconds before the point of conflict, the time taken for both robots to pass through that location is staggered, thereby avoiding a collision. This spatiotemporal coordination mechanism ensures the safe collaborative operation of multiple robots in complex underwater environments.

[0048] S106. Distribute the task takeover allocation table and the adjusted navigation trajectory group to the target takeover robot group, synchronously update the task execution sequence of each robot, and monitor the thermal status data delay and temperature change of each robot when performing tasks in real time. If the temperature value of a robot approaches the processor overheating threshold again or the thermal status fluctuation frequency is abnormal, the thermal load control process is triggered to generate thermal load control instructions.

[0049] For the adjusted navigation trajectory group and task takeover allocation table, a preset multicast communication protocol is used to send data packets to the target takeover robot group. The response speed is optimized through the priority sorting mechanism in the protocol. The data packets contain the new task number, execution order, and corresponding navigation trajectory coordinate sequence of each robot. Each receiving robot extracts relevant information based on its own number and updates its local task execution sequence, completing the synchronous update process of task allocation. During the execution of the updated task sequence by each robot, the processor core temperature value is obtained at preset sampling intervals through the temperature sensor built into the processor. At the same time, the time interval from sending the task command to receiving the execution feedback is recorded as a communication delay parameter. The difference between the current temperature value and the temperature value at the previous sampling time is calculated, and the number of times the temperature difference exceeds the preset change threshold within a unit time window is counted to obtain the temperature fluctuation frequency value and the real-time temperature monitoring dataset. Based on the temperature monitoring dataset, determine whether the processor temperature of a robot exceeds the first proportion of the preset overheating threshold, or whether the temperature fluctuation frequency exceeds the preset frequency upper limit threshold. If either condition is met, construct a heat load control instruction set based on the current task priority and remaining task queue length of the robot. This set includes instructions to reduce the processor operating frequency, pause the execution of low-priority tasks, and start the auxiliary cooling device. Send the control instruction set to the corresponding robot for execution through the same multicast communication protocol.

[0050] In one possible implementation, the application of a multicast communication protocol solves the information synchronization problem of underwater robot swarms. This protocol allows a sending node to transmit the same data packets to multiple receiving nodes simultaneously, avoiding the overhead of redundant transmissions in point-to-point communication. A priority sorting mechanism assigns different transmission priorities to data packets based on the urgency and importance of the tasks. When high-priority navigation trajectory data and low-priority status reports are waiting to be sent simultaneously, the protocol prioritizes the transmission of navigation data, ensuring the timely delivery of critical task information.

[0051] Specifically, the synchronous update process of the task execution sequence involves complex data parsing and storage operations. Upon receiving a data packet, each robot first verifies the list of target robot IDs in the packet header to confirm whether it is the receiving target. It then extracts the task segment corresponding to its own ID, including the task type code, estimated execution time, and dependencies on preceding tasks. This information is parsed and stored in the local task queue, replacing the original task schedule and enabling dynamic adjustment of cluster task allocation.

[0052] It's important to note that the sampling period for temperature monitoring directly impacts the accuracy of thermal state assessment. A sampling period that's too short increases processor load, while one that's too long may miss sudden temperature changes. In practice, it's typically set between 500 milliseconds and 2 seconds. The temperature difference is calculated by subtracting the previous sample value from the current one; a positive value indicates a temperature increase, and a negative value indicates a temperature decrease. The frequency of temperature fluctuations is determined statistically within a sliding time window. For example, if the temperature difference exceeds 2 degrees Celsius more than 6 times within the last 10 sampling periods, the fluctuation frequency is considered abnormal.

[0053] In one embodiment, obtaining the communication delay parameter provides crucial information for subsequent thermal load analysis. When the robot processor is overloaded, not only does the temperature rise, but the response speed to external commands also slows down. The communication delay can be obtained by recording the timestamp of sending the task command and the timestamp of receiving the execution confirmation signal, and calculating the difference between the two. This delay value indirectly reflects the processor's workload, forming a complementary monitoring indicator with temperature data.

[0054] For example, the thermal load control instruction set employs a tiered response mechanism. When the temperature reaches 80% of the overheat threshold, a frequency reduction instruction is first executed, lowering the processor's clock speed by 20%. If the temperature continues to rise to 90% of the threshold, tasks with a priority lower than 3 in the task queue are suspended. In the most severe cases, when the temperature approaches the threshold or the frequency fluctuation is severely abnormal, the water cooling system is immediately activated or the cooling fan speed is increased. This gradual control avoids task interruptions caused by overreaction.

[0055] Preferably, the design of sending control commands via the same multicast communication protocol ensures the reliability and real-time performance of command transmission. Because the same communication mechanism is used, control commands can reuse established communication links without requiring an additional handshake process. Upon receiving the control command, the robot immediately executes the corresponding operation and feeds back the execution result to the cluster management node, forming a closed-loop control system that ensures the thermal load remains within a safe range.

[0056] S107. Dynamically adjust the peak operating load of the affected robot according to the heat load control command, reduce the processor frequency or suspend secondary tasks to obtain the adjusted operating state combination, analyze the overall task priority deviation and takeover delay of the cluster, and generate task progress correction command by comparing the current task progress with the preset target.

[0057] Based on the frequency reduction ratio and task pause list in the heat load control command, the current processor operating frequency and task queue status are read through the local thermal management module of each robot. Real-time water temperature values ​​measured by the ambient water temperature sensor are obtained. The difference between the water temperature value and the preset standard heat dissipation temperature is used to calculate the heat dissipation efficiency reduction coefficient. This coefficient is multiplied by the original frequency reduction ratio to obtain the actual frequency reduction magnitude. Frequency adjustment is performed, and specified low-priority tasks are paused. A combination of operating statuses is output, including the adjusted frequency values ​​of each robot and the task numbers to be executed. Based on the new frequency values ​​of each robot in the operating status combination, the percentage decrease in its task processing capacity is calculated. The difference between the number of tasks with a priority higher than the preset level in the current cluster and the number before adjustment is calculated. This difference is divided by the number before adjustment to obtain the task priority deviation percentage. Simultaneously, the communication time and startup time consumed by transferring each task from the original executing robot to the takeover robot are accumulated to obtain a set of quantitative parameters including the deviation percentage and cumulative delay time. The estimated completion time of each task is adjusted by quantifying the delay duration in the parameter set. The completed workload of each task is read from the task management database and divided by the total workload to obtain the completion percentage. This percentage is compared with the target percentage that should be reached at the current time according to the original plan. If the difference exceeds a preset deviation threshold, the required time is recalculated based on the remaining workload and currently available computing resources, generating a progress correction instruction set that includes a new task execution order and completion time nodes. A task-robot allocation matrix is ​​constructed based on the task execution order in the progress correction instruction set. An execution timeline is drawn according to the new completion time nodes of each task. The load occupancy of each robot in each time period is statistically analyzed to form a resource utilization data table. By integrating the above matrix, timeline, and data table, a cluster collaborative operation execution blueprint containing a complete task allocation scheme and timing arrangement is determined.

[0058] In one possible implementation, the thermal management module, as a key component of the robot, integrates temperature monitoring, frequency control, and task scheduling functions. This module collects real-time temperature data from the processor core, memory controller, and graphics processing unit via a built-in temperature sensor array. When it receives a frequency reduction command, the module first reads the current dynamic frequency value; for example, if the processor is running at 2.4GHz, and the frequency reduction ratio is 30%, the target frequency should be 1.68GHz. However, the unique characteristics of the underwater environment make heat dissipation efficiency closely related to water temperature.

[0059] Specifically, the calculation of the heat dissipation efficiency reduction factor is based on the principle of heat conduction. The standard heat dissipation temperature is typically set at 15 degrees Celsius in a seawater environment. When the actual water temperature rises to 25 degrees Celsius, the temperature difference decreases by 10 degrees, and the heat dissipation efficiency decreases accordingly. Experimental data fitting shows that for every 1 degree Celsius increase, the heat dissipation efficiency decreases by approximately 3%. Therefore, a 10-degree temperature difference leads to a 30% reduction in heat dissipation efficiency, meaning that the originally planned 30% frequency reduction needs to be adjusted to 39% to ensure the processor temperature remains within a safe range.

[0060] It's important to note that the quantitative assessment of task processing capacity directly impacts the overall efficiency of the cluster. Processor frequency and computing power have an approximately linear relationship; a 39% reduction in frequency means a corresponding decrease in the number of instructions executed per unit of time. For computationally intensive tasks such as image recognition, processing time will increase accordingly. However, for I / O-intensive tasks such as data transfer, the impact is relatively small. By statistically analyzing the distribution of different task types, the overall performance loss can be accurately assessed.

[0061] In one embodiment, the calculation of task priority deviation reflects the rationality of resource allocation. Assume the cluster originally had 20 high-priority tasks evenly distributed across 10 robots, with each robot handling 2 tasks. After hot load balancing, some robots reduce their task frequency or pause tasks, leading to a redistribution of high-priority tasks. This may result in an imbalance where some robots handle 4 tasks while others handle only 1. The deviation percentage quantifies this imbalance by calculating the allocation variance.

[0062] For example, real-time tracking of task completion percentage relies on the measurable characteristics of the task. For path planning tasks, the completion percentage equals the planned path length divided by the total path length; for area scanning tasks, it is the ratio of the scanned area to the total area. The task management database continuously updates this progress information, and when it detects that the actual progress of a task is more than 15% behind the planned progress, a progress correction mechanism is triggered.

[0063] Preferably, the final blueprint construction process embodies the concept of multi-dimensional optimization. The task allocation matrix adopts a two-dimensional table format, with rows representing tasks, columns representing robots, and matrix element values ​​representing allocation relationships and execution priorities. The execution time axis divides time into fixed-interval time slices, marking the tasks executed by each robot within each time slice. The resource utilization data table statistically analyzes the CPU utilization, memory utilization, and communication bandwidth utilization of each robot within each time slice. Through comprehensive analysis of these three dimensions, a collaborative operation scheme that considers both task completion efficiency and resource balance is formed.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart thermal management and energy-saving method for underwater robots, characterized in that, The method includes: The thermal sensors built into each robot in the underwater robot cluster collect processor core temperature data and operating load data, construct thermal state vectors, and store them in distributed data nodes to form an initial thermal state record set. Broadcast the thermal state vectors of each robot, and after receiving the thermal state vectors of neighboring nodes, store them according to timestamps to form a distributed thermal state record set; Based on the comparison between the temperature values ​​of each robot in the distributed thermal state record set and the preset overheating threshold, if the temperature value exceeds the standard and the duration exceeds the preset range, it is marked as a thermal risk unit, and a thermal risk unit list is generated. Based on the thermal risk unit list analysis, the probability of task interruption for each robot is analyzed. Combined with the temperature fluctuation frequency and load peak, the ratio of task load to heat dissipation efficiency is calculated to generate a list of tasks with limited capabilities. By analyzing the cluster communication bandwidth, the task queue congestion is analyzed, affected tasks are reallocated, a task takeover allocation table is generated, and the target takeover robot combination is determined. Based on the task takeover assignment table and the underwater environment's water flow resistance and terrain obstruction coefficient, adjust the target takeover robot's navigation trajectory and generate a navigation trajectory group. The task takeover assignment table and navigation trajectory group are distributed to the target takeover robot group, the task execution sequence is updated synchronously, temperature data and fluctuation frequency are monitored in real time, and heat load control instructions are generated. Adjust the peak operating load according to the heat load control command, generate operating status combinations, analyze task priority deviation and takeover delay time, and generate task progress correction commands.

2. The underwater robot swarm thermal management and task coordination method according to claim 1, characterized in that, The construction of the thermal state vector includes: For the processor core temperature data and operating load data collected by each robot, a sliding window of fixed size is set. The temperature change rate is obtained by dividing the temperature difference between adjacent sampling points within the window by the sampling time interval. The load change rate is obtained by dividing the load difference between adjacent sampling points by the sampling time interval. Perform a fast Fourier transform on the temperature data within the window, and take the frequency component with the largest amplitude as the main frequency. If the main frequency exceeds the preset threshold, it is marked as an abnormal fluctuation state. A four-dimensional thermal state vector is constructed based on the current temperature, load, heat dissipation efficiency coefficient, and main frequency.

3. The underwater robot swarm thermal management and task coordination method according to claim 2, characterized in that, The generated thermal risk unit list includes: Each robot's identifier, temperature value, load value, and timestamp are encapsulated into a data frame, which is then sent to all nodes in the cluster. The data frame content is parsed and stored in local memory to form a real-time synchronized cluster hot status dataset; For the cluster thermal state dataset, compare the temperature values ​​of each robot with the preset overheating threshold, record the overheating robot identifier and overheating start time, calculate the duration of thermal overload, if the duration exceeds the preset range, set the thermal risk flag to true, extract the identifier, temperature value, and duration information, calculate the product of the temperature exceeding the threshold and the duration as the risk value, and generate a thermal risk unit list by sorting according to the risk values.

4. The underwater robot swarm thermal management and task coordination method according to claim 3, characterized in that, The generation of the task takeover allocation table, which determines the target takeover robot combination, includes: Based on the task capability limitation list, record the task types and hardware limitation parameters that each robot does not support, and combine the cluster communication bandwidth monitoring data to calculate the available bandwidth matrix for task transfer between nodes. Based on the available bandwidth matrix and task queue length, calculate the sum of the product of the expected task execution time and the priority value, divide by the total available bandwidth between nodes, and generate a load status table containing congestion level values ​​and task type distribution. Identify high-load robots, extract restricted tasks, match low-load candidate takeover robots, construct an allocation cost matrix, solve for the optimal allocation scheme, and generate a task takeover allocation table.

5. The underwater robot swarm thermal management and task coordination method according to claim 4, characterized in that, The adjustment of the target takeover robot's navigation trajectory and the generation of a navigation trajectory group include: Extract the robot's starting and target positions from the task takeover assignment table and generate an initial navigation path; Water flow velocity field data are collected using underwater environmental sensors, and the water flow resistance coefficient is calculated. Obstacle distribution is extracted from underwater topography database, topography occlusion coefficient is calculated, and environmental constraint matrix is ​​formed; The initial navigation path is corrected using the environmental constraint matrix, the path direction is adjusted to reduce water flow resistance, and the position with the minimum obstruction coefficient is selected as the relay point at the boundary of the communication blind zone to generate a set of candidate navigation trajectories. The candidate navigation trajectory set is discretized into a spatiotemporal coordinate sequence, and the spatial coordinate distance between trajectories is detected. If the distance is less than a preset safety threshold, the movement speed is adjusted or a waiting time is inserted to generate a navigation trajectory group.

6. The underwater robot swarm thermal management and task coordination method according to claim 5, characterized in that, The real-time monitoring of temperature data and fluctuation frequency generates heat load control commands, including: Send a data packet containing the task number, execution order, and navigation trajectory coordinate sequence to the target takeover robot assembly; Each robot updates its task execution sequence according to its own number; The core temperature value is obtained by the processor temperature sensor, the communication delay parameter is recorded, the temperature difference is calculated, and the temperature fluctuation frequency is statistically analyzed. If the temperature value exceeds the first proportion of the preset overheat threshold or the fluctuation frequency exceeds the preset upper limit threshold, a set of control instructions is generated based on the task priority and queue length, including reducing the processor frequency, pausing low-priority tasks, and activating auxiliary cooling devices.

7. The underwater robot swarm thermal management and task coordination method according to claim 6, characterized in that, The generated task progress correction instruction includes: The thermal management module reads the processor's operating frequency and task queue status, calculates the heat dissipation efficiency reduction coefficient by combining the ambient water temperature sensor data, adjusts the frequency reduction range, pauses low-priority tasks, and generates a combination of running statuses containing the adjusted frequency value and task number. Calculate the percentage decrease in task processing capacity based on the combination of the operating states, calculate the percentage deviation of task priority, accumulate task transfer communication and startup time, and generate a set of quantitative parameters. The task completion time is corrected by the quantization parameter set, the difference between the completion percentage and the target percentage is calculated, and a progress correction instruction set containing the task execution order and completion time nodes is generated.

Images