A dynamic link joint optimization method, system, terminal, and storage medium driven by Ascend computing power

Abstract

The present invention discloses a dynamic link joint optimization method, system, terminal and storage medium driven by Ascend computing power. The method includes: a link prediction network predicts the performance changes of each link based on communication link performance data and historical data; an environment perception network predicts the impact of the environment on link performance based on the real-time status data of the drone and environmental data; a flight control task solution module obtains the flight control instruction priority based on the task type and task requirements; and a reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority. Based on Ascend computing power, the present invention integrates priority instructions based on flight control task requirements, predicts the link status, and dynamically switches the communication mode in real time according to the prediction results, achieving highly reliable, low-latency, and highly robust communication guarantees in low-altitude economic scenarios, providing core support for users such as drones and air traffic management.

Description

Technical Field

[0001] The present invention relates to the field of link optimization technology, and in particular to a dynamic link joint optimization method, system, terminal, and computer-readable storage medium driven by Ascend computing power. Background Art

[0002] With the development of the Internet of Things and 5G technology, the combination of multimode communication modules and edge computing has become a key path to solving the demand for large-scale and large-volume data processing. In the low-altitude economy (such as drone logistics and urban air traffic management), drones can use multimode communication protocols to achieve beyond-line-of-sight control and real-time data processing.

[0003] However, existing technologies suffer from high latency during multi-mode protocol switching, severely limiting application reliability. For example, when a drone flying at low altitude needs to switch from 5G to satellite communications due to building obstructions, traditional static allocation methods lack a dynamic coordination mechanism for edge computing power. This switching process can easily result in significant delays in the migration of navigation data, failing to meet the real-time control requirements of drones. This makes it difficult for drones or drone clusters to seamlessly switch protocols, leading to risks such as course deviation and mission interruption. This issue significantly limits the large-scale application of the low-altitude economy.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a dynamic link joint optimization method, system, terminal and computer-readable storage medium driven by Ascend computing power, aiming to solve the problem in the prior art that high latency during multi-mode protocol switching seriously restricts application reliability.

[0006] To achieve the above objectives, the present invention provides a dynamic link joint optimization method driven by Ascend computing power, which includes the following steps:

[0007] Acquire communication link performance data and historical data, and input the communication link performance data and the historical data into a link prediction network, wherein the link prediction network predicts performance changes of each link based on the communication link performance data and the historical data;

[0008] Acquire real-time status data and environmental data of the drone, input the real-time status data and environmental data of the drone into an environmental perception network, and the environmental perception network predicts the impact of the environment on link performance based on the real-time status data and environmental data of the drone;

[0009] Obtaining a task type and a task requirement, inputting the task type and the task requirement into a flight control task solving module, wherein the flight control task solving module obtains a flight control instruction priority according to the task type and the task requirement;

[0010] The performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction are input into a reinforcement learning decision module, and the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction.

[0011] Optionally, the Ascend computing power-driven dynamic link joint optimization method, wherein the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction, further includes:

[0012] Output a dynamic switching decision and feed the dynamic switching decision back to the link prediction network, the environment perception network, and the flight control task solving module to optimize network performance.

[0013] Optionally, in the dynamic link joint optimization method driven by Ascend computing power, the communication link performance data includes: 4G / 5G, Beidou and Tiantong signal strength, latency, channel application response time, single message success rate, transmission rate, remaining power and data security level;

[0014] The historical data includes: the number and timing of communication link switching during flight, message response time, Tiantong transmission rate and bit error / packet loss rate;

[0015] The link prediction network predicts the performance change of each link based on the communication link performance data and the historical data, specifically including:

[0016] Integrate historical data, hardware device status data, and environmental data to obtain multi-source data, and use the Ascend NPU to align the spatiotemporal coordinates of the multi-source data.

[0017] Long-short-term memory networks and the Transformer algorithm are used to integrate historical data with environmental features for time series data prediction. The long-short-term memory network is used to capture the long-term strong correlation between communication links through a gating mechanism, predicting the delay trend within a preset time in the future, thereby achieving trend prediction of time series data. The Transformer algorithm is used to analyze the correlation changes between multiple communication links.

[0018] The communication link is switched based on the prediction results.

[0019] Optionally, in the Ascend computing power-driven dynamic link joint optimization method, the real-time status data of the drone includes: drone position, speed, attitude, and altitude;

[0020] The environmental data includes: surrounding environment images, obstacles and no-fly zone information;

[0021] The environment perception network predicts the impact of the environment on the link performance based on the real-time status data of the drone and the environmental data, specifically including:

[0022] Analyze the drone's environment and its own flight status and trends based on the drone's real-time status data and the environmental data;

[0023] The environment-aware network algorithm is used to predict the transmission status, attenuation and enhancement trends of link signals in the surrounding space.

[0024] Optionally, in the Ascend computing power-driven dynamic link joint optimization method, the context-aware network algorithm includes:

[0025] Feature analysis layer: The input environment image is locally scanned through a convolution kernel array. Each convolution kernel automatically extracts basic features and analyzes the overall shape and size parameters of obstacles through feature fusion.

[0026] Decision output layer: Maps spatial features into obstacle attribute parameters, outputs position, size, and shape classification results, combines spatial features with link status data, and evaluates the transmission and occlusion of communication link signals and data in the environmental space.

[0027] Optionally, in the Ascend computing power-driven dynamic link joint optimization method, the task types include: inspection, logistics distribution, and emergency rescue;

[0028] The mission requirements include: obstacle avoidance, cruising, and ascent and descent;

[0029] The flight control task solving module obtains the flight control instruction priority according to the task type and the task requirement, specifically including:

[0030] Based on the task type and task requirements, analyze the bandwidth, latency, and data type requirements of the data transmission corresponding to the task, and output the required communication link type, priority, and redundancy strategy;

[0031] The Ascend NPU analyzes task requirements and classifies multimodal data, maps task types in parallel to communication link indicators, dynamically selects the main link from the available links, forcibly binds redundant links, and dynamically adjusts protocol parameters.

[0032] Optionally, the Ascend computing power-driven dynamic link joint optimization method, wherein the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions, specifically includes:

[0033] Based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions, a reinforcement learning algorithm is used to select the optimal communication mode and ensure redundant links;

[0034] Based on the link status prediction results and task requirements, a reward function including bandwidth, latency, and switching penalties is set. The deep Q-network algorithm is used to dynamically select the optimal communication mode. The bandwidth, transmission interval, primary link, and redundant links are dynamically allocated through the proximal strategy optimization algorithm. The Ascend NPU is used to complete the communication link allocation decision within the latency requirements.

[0035] In addition, to achieve the above objectives, the present invention also provides a dynamic link joint optimization system driven by Ascend computing power, wherein the dynamic link joint optimization system driven by Ascend computing power includes:

[0036] Link prediction network, used to predict performance changes of each link based on communication link performance data and historical data;

[0037] Environmental awareness network, used to predict the impact of the environment on link performance based on the real-time status data of the drone and environmental data;

[0038] The flight control task solving module is used to obtain the flight control instruction priority according to the task type and task requirements;

[0039] A reinforcement learning decision module is used to dynamically select the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions.

[0040] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an Ascend computing power-driven dynamic link joint optimization program stored on the memory and runnable on the processor. When the Ascend computing power-driven dynamic link joint optimization program is executed by the processor, the steps of the Ascend computing power-driven dynamic link joint optimization method as described above are implemented.

[0041] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a dynamic link joint optimization program driven by Ascend computing power, and when the dynamic link joint optimization program driven by Ascend computing power is executed by a processor, the steps of the dynamic link joint optimization method driven by Ascend computing power as described above are implemented.

[0042] In the present invention, communication link performance data and historical data are obtained, and the communication link performance data and the historical data are input into a link prediction network, and the link prediction network predicts the performance changes of each link based on the communication link performance data and the historical data; the real-time status data and environmental data of the drone are obtained, and the real-time status data and the environmental data of the drone are input into an environmental perception network, and the environmental perception network predicts the impact of the environment on the link performance based on the real-time status data and the environmental data of the drone; the task type and task requirements are obtained, and the task type and the task requirements are input into a flight control task solving module, and the flight control task solving module obtains the flight control instruction priority based on the task type and the task requirements; the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority are input into a reinforcement learning decision module, and the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority. This invention is based on Ascend computing power, integrates priority instructions based on flight control mission requirements, predicts link status, and dynamically switches communication modes in real time based on the prediction results, achieving highly reliable, low-latency, and highly robust communication guarantees in low-altitude economic scenarios, providing core support for users such as drones and air traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flowchart of a preferred embodiment of the dynamic link joint optimization method driven by Ascend computing power of the present invention;

[0044] Figure 2 This is a schematic diagram of a preferred embodiment of the dynamic link joint optimization system driven by Ascend computing power of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clear and distinct, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] The dynamic link joint optimization method driven by Ascend computing power described in the preferred embodiment of the present invention is as follows: Figure 1 and Figure 2 As shown, the Ascend computing power-driven dynamic link joint optimization method includes the following steps:

[0047] Step S10: Acquire communication link performance data and historical data, input the communication link performance data and the historical data into a link prediction network, and the link prediction network predicts performance changes of each link based on the communication link performance data and the historical data.

[0048] Specifically, the communication link performance data include: signal strength, latency, channel application response time, single message success rate, transmission rate, remaining power and data security level of 4G / 5G, Beidou and Tiantong; the historical data include: the number and timing of communication link switching during flight, message response time, Tiantong transmission rate and bit error / packet loss rate, etc.; the link prediction network is based on the communication link performance data and historical data input, and predicts the time-series change pattern of the three-mode communication link status through the link prediction network algorithm. The time-series change pattern of the communication link covers periodic fluctuations (such as the difference in base station load during the day and night), trend attenuation (signal weakening with distance) and sudden interference (obstruction by obstacles).

[0049] The link prediction network predicts the performance change of each link based on the communication link performance data and the historical data, specifically including:

[0050] (1) Multi-source data fusion and feature extraction: Fusion of historical data (historical delay, bandwidth, etc. of three communication modes, namely Tiantong satellite communication, Beidou third-generation short message communication, and 4G / 5G communication), hardware device status data, and environmental data (satellite angle, real-time weather and surrounding environment, etc.) to obtain multi-source data, and align the spatiotemporal coordinates of the multi-source data through the Ascend NPU.

[0051] (2) Spatiotemporal model prediction and NPU accelerated reasoning: Using the long short-term memory network (LSTM, Long Short-Term Memory, is a special recurrent neural network designed specifically to solve the gradient vanishing / explosion problem of traditional RNN when processing long sequence data, and can effectively capture long-term dependencies) and the Transformer algorithm, historical data and environmental features are integrated to predict time series data. Among them, the long short-term memory network is used to capture the long-term strong correlation between communication links through the gating mechanism, and predict the delay trend within a preset time in the future (for example, within the next 10 seconds) (such as the direct causal relationship such as the increase in base station load leading to the increase in delay) to achieve trend prediction of time series data. The Transformer algorithm is used to analyze the correlation changes between multiple communication links (such as the impact of 5G signal attenuation on Tiantong load).

[0052] (3) Dynamic decision-making: Switch the communication link based on the prediction results. Assuming that the 5G delay is predicted to exceed the threshold in 5 seconds, the main link will be switched to the Tiantong satellite 200ms in advance (active switching), and Beidou will be bound as redundancy to avoid the risk of signal interruption.

[0053] Step S20: Acquire real-time status data and environmental data of the drone, input the real-time status data and environmental data of the drone into the environmental perception network, and the environmental perception network predicts the impact of the environment on the link performance based on the real-time status data and environmental data of the drone.

[0054] Specifically, the real-time status data of the drone includes: the drone's position, speed, attitude and altitude; the environmental data includes: surrounding environment images, obstacles and no-fly zone information, etc.; the environmental perception network analyzes the drone's environment and its own flight status and trends based on the drone's real-time status data and environmental data input, and predicts the transmission status and attenuation and enhancement trends of the link signal in the environmental space through the environmental perception network algorithm.

[0055] Environmental Perception Network Algorithm: For a large amount of environmental image input, a convolutional neural network (CNN) is used to extract spatial features in the environmental data, such as the location, size, and shape of obstacles.

[0056] Feature parsing layer: The input environment image is locally scanned using a convolution kernel array. Each convolution kernel automatically extracts basic features such as obstacle edges, retaining significant features while adapting to obstacle position offsets. Feature fusion is then used to parse the overall shape and size parameters of the obstacle.

[0057] Decision output layer: This layer maps spatial features into obstacle attribute parameters and outputs location, size, and shape classification results. It combines spatial features with link status data to assess the transmission and obstruction of communication link signals and data in the environment.

[0058] Step S30: Obtain a task type and a task requirement, input the task type and the task requirement into a flight control task solving module, and the flight control task solving module obtains a flight control instruction priority according to the task type and the task requirement.

[0059] Specifically, the mission types include inspection, logistics distribution, and emergency rescue; the mission requirements include obstacle avoidance, cruising, ascent and descent, and other specific scenario requirements; the flight control mission solution module analyzes the bandwidth, latency, and data type requirements of the data transmission corresponding to the mission based on the mission type and mission requirements, and outputs the required communication link type and priority and redundancy strategy (i.e., sorting the priorities from high to low in the order of 4G / 5G communication, Beidou-3 short message communication, and Tiantong satellite communication), as shown in Table 1:

[0060] Table 1: Allocation relationship between primary links and redundant links

[0061]

[0062] The flight control task solving module uses the Ascend NPU to analyze task requirements (such as obstacle avoidance, inspection, logistics, and emergency response) and classify multimodal data (such as weather and environmental images collected by drone sensors, and its own status). It then maps task types in parallel to communication link indicators such as latency, bandwidth, and reliability. It then dynamically selects the main link from the available links, forcibly binds redundant links (i.e., forcing dual-path communication, with one path serving as a redundant backup to prevent complete communication interruption), and dynamically adjusts protocol parameters.

[0063] Step S40: Input the performance changes of each link, the impact of the environment on the link performance, and the flight control instruction priority into the reinforcement learning decision module, and the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the flight control instruction priority.

[0064] Specifically, the reinforcement learning decision module is based on the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority (i.e., the changes in the performance indicators of each link output by the link prediction network, such as latency, etc.; the characteristic results of environmental obstacles, occlusions, etc. output by the environmental perception network and the judgment conclusion on the impact on the link signal; the current task output by the flight control task solution module most requires a link channel. For example, if the current task requires real-time image transmission, then it is necessary to try to ensure that 4G / 5G has a signal. At the flight control level, it is necessary to avoid obstacles and occlusions as soon as possible to restore 4G / 5G). The optimal communication mode is selected through a reinforcement learning algorithm and redundant links are guaranteed. Based on the link status prediction results and task requirements, a reward function is set that includes bandwidth, latency, and switching penalties. (In reinforcement learning algorithms, penalty terms related to action or state switching are introduced into standard rewards to constrain the stability of the agent's behavior. The main reward functions include real-time rewards (rewards for low latency), switching frequency rewards (rewards for low frequency), and signal interruption rewards (rewards for no interruption)). A deep Q-network algorithm is used to dynamically select the optimal communication mode. The proximal policy optimization algorithm is used to dynamically allocate bandwidth, transmission intervals, primary links, and redundant links. The Ascend NPU is used to complete communication link allocation decisions within the latency requirements.

[0065] Furthermore, if Figure 2 As shown, a dynamic switching decision is output and fed back to the link prediction network, the environment perception network and the flight control task solving module to optimize network performance.

[0066] This invention is based on Ascend computing power, integrates priority instructions based on flight control mission requirements, predicts link status, and dynamically switches three-mode communications in real time based on the prediction results, achieving "high reliability, low latency, and strong robustness" communication guarantees in low-altitude economic scenarios, providing core support for users such as drones and air traffic management.

[0067] Furthermore, if Figure 2 As shown, based on the above-mentioned Ascend computing power-driven dynamic link joint optimization method, the present invention also provides an Ascend computing power-driven dynamic link joint optimization system, wherein the Ascend computing power-driven dynamic link joint optimization system includes:

[0068] Link prediction network, used to predict performance changes of each link based on communication link performance data and historical data;

[0069] Environmental awareness network, used to predict the impact of the environment on link performance based on the real-time status data of the drone and environmental data;

[0070] The flight control task solving module is used to obtain the flight control instruction priority according to the task type and task requirements;

[0071] A reinforcement learning decision module is used to dynamically select the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions.

[0072] Furthermore, based on the above-mentioned Ascend computing power-driven dynamic link joint optimization method and system, the present invention also provides a terminal, which includes a processor, a memory, and a display.

[0073] In some embodiments, the memory may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory may also be an external storage device of the terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal. Furthermore, the memory may include both an internal storage unit of the terminal and an external storage device. The memory is used to store application software and various types of data installed on the terminal, such as program codes of the installation terminal. The memory may also be used to temporarily store data that has been output or is to be output. In one embodiment, a dynamic link joint optimization program driven by Ascend computing power is stored on the memory, and the dynamic link joint optimization program driven by Ascend computing power can be executed by a processor, thereby realizing the dynamic link joint optimization method driven by Ascend computing power in this application.

[0074] In some embodiments, the processor may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to run program code or process data stored in the memory, such as executing the dynamic link joint optimization method driven by Ascend computing power.

[0075] In some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display is used to display information on the terminal and to display a visual user interface. The processor, memory, and display of the terminal communicate with each other via a system bus.

[0076] In one embodiment, when the processor executes the Ascend computing power-driven dynamic link joint optimization program in the memory, the following steps of the Ascend computing power-driven dynamic link joint optimization method are implemented.

[0077] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a dynamic link joint optimization program driven by Ascend computing power, and when the dynamic link joint optimization program driven by Ascend computing power is executed by a processor, the steps of the dynamic link joint optimization method driven by Ascend computing power as described above are implemented.

[0078] In summary, the present invention provides a dynamic link joint optimization method, system, terminal and computer-readable storage medium driven by Ascend computing power. The method includes: obtaining communication link performance data and historical data, inputting the communication link performance data and the historical data into a link prediction network, and the link prediction network predicts the performance changes of each link based on the communication link performance data and the historical data; obtaining drone real-time status data and environmental data, inputting the drone real-time status data and the environmental data into an environment perception network, and the environment perception network predicts the impact of the environment on link performance based on the drone real-time status data and the environmental data; obtaining task type and task requirements, inputting the task type and the task requirements into a flight control task solving module, and the flight control task solving module obtains the flight control instruction priority based on the task type and the task requirements; inputting the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority into a reinforcement learning decision module, and the reinforcement learning decision module dynamically selects the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance and the flight control instruction priority. This invention is based on Ascend computing power, integrates priority instructions based on flight control mission requirements, predicts link status, and dynamically switches communication modes in real time based on the prediction results, achieving highly reliable, low-latency, and highly robust communication guarantees in low-altitude economic scenarios, providing core support for users such as drones and air traffic management.

[0079] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal comprising the element.

[0080] Of course, those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware (such as a processor, controller, etc.) through a computer program. The program can be stored in a computer-readable storage medium that can be read by a computer. When executed, the program can include the processes in the above-described method embodiments. The computer-readable storage medium can be a memory, a magnetic disk, an optical disk, etc.

[0081] It should be understood that the application of the present invention is not limited to the above examples. For those skilled in the art, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A dynamic link joint optimization method driven by Ascend computing power, characterized by: The Ascend computing power-driven dynamic link joint optimization method includes: Acquire communication link performance data and historical data, and input the communication link performance data and the historical data into a link prediction network, wherein the link prediction network predicts performance changes of each link based on the communication link performance data and the historical data; Integrate historical data, hardware device status data, and environmental data to obtain multi-source data, and use the Ascend NPU to align the spatiotemporal coordinates of the multi-source data. Long-short-term memory networks and the Transformer algorithm are used to integrate historical data with environmental features for time series data prediction. The long-short-term memory network is used to capture the long-term strong correlation between communication links through a gating mechanism, predicting the delay trend within a preset time in the future, thereby achieving trend prediction of time series data. The Transformer algorithm is used to analyze the correlation changes between multiple communication links. Switching the communication link according to the prediction results; Acquire real-time status data and environmental data of the drone, input the real-time status data and environmental data of the drone into an environmental perception network, and the environmental perception network predicts the impact of the environment on link performance based on the real-time status data and environmental data of the drone; Based on the real-time status data of the UAV and the environmental data, the UAV's environment and its own flight status and trend are analyzed; the transmission status, attenuation, and enhancement trends of the link signal in the environmental space are predicted through the environmental perception network algorithm; The environment perception network algorithm includes: Feature analysis layer: The input environment image is locally scanned through a convolution kernel array. Each convolution kernel automatically extracts basic features and analyzes the overall shape and size parameters of obstacles through feature fusion. Decision output layer: Maps spatial features into obstacle attribute parameters, outputs position, size, and shape classification results, combines spatial features with link status data, and evaluates the transmission and obstruction of communication link signals and data in the environment space; Obtaining a task type and a task requirement, inputting the task type and the task requirement into a flight control task solving module, wherein the flight control task solving module obtains a flight control instruction priority according to the task type and the task requirement; Inputting the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction into a reinforcement learning decision module, the reinforcement learning decision module dynamically selecting an optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction; Based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions, a reinforcement learning algorithm is used to select the optimal communication mode and ensure redundant links; Based on the link status prediction results and task requirements, a reward function including bandwidth, latency, and switching penalties is set. The deep Q-network algorithm is used to dynamically select the optimal communication mode. The bandwidth, transmission interval, primary link, and redundant links are dynamically allocated through the proximal strategy optimization algorithm. The Ascend NPU is used to complete the communication link allocation decision within the latency requirements.

2. The Ascend computing power-driven dynamic link joint optimization method according to claim 1, characterized in that: The reinforcement learning decision module dynamically selects the optimal communication mode according to the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instruction, and then further includes: Output a dynamic switching decision and feed the dynamic switching decision back to the link prediction network, the environment perception network, and the flight control task solving module to optimize network performance.

3. The Ascend computing power-driven dynamic link joint optimization method according to claim 1, characterized in that: The communication link performance data includes: 4G / 5G, Beidou and Tiantong signal strength, latency, channel application response time, single message success rate, transmission rate, remaining power and data security level; The historical data includes: the number and timing of communication link switching during flight, message response time, Tiantong transmission rate and bit error / packet loss rate.

4. The Ascend computing power-driven dynamic link joint optimization method according to claim 1, characterized in that: The real-time status data of the UAV includes: the position, speed, attitude and altitude of the UAV; The environmental data includes: surrounding environment images, obstacles and no-fly zone information.

5. The Ascend computing power-driven dynamic link joint optimization method according to claim 1, characterized in that: The types of tasks include: inspection, logistics distribution and emergency rescue; The mission requirements include: obstacle avoidance, cruising, and ascent and descent; The flight control task solving module obtains the flight control instruction priority according to the task type and the task requirement, specifically including: Based on the task type and task requirements, analyze the bandwidth, latency, and data type requirements of the data transmission corresponding to the task, and output the required communication link type, priority, and redundancy strategy; The Ascend NPU analyzes task requirements and classifies multimodal data, maps task types in parallel to communication link indicators, dynamically selects the main link from the available links, forcibly binds redundant links, and dynamically adjusts protocol parameters.

6. A dynamic link joint optimization system driven by Ascend computing power, characterized by: The Ascend computing power-driven dynamic link joint optimization system is used to implement the Ascend computing power-driven dynamic link joint optimization method according to any one of claims 1 to 5. The Ascend computing power-driven dynamic link joint optimization system includes: Link prediction network, used to predict performance changes of each link based on communication link performance data and historical data; Environmental awareness network, used to predict the impact of the environment on link performance based on the real-time status data of the drone and environmental data; The flight control task solving module is used to obtain the flight control instruction priority according to the task type and task requirements; A reinforcement learning decision module is used to dynamically select the optimal communication mode based on the performance changes of each link, the impact of the environment on the link performance, and the priority of the flight control instructions.

7. A terminal, characterized in that: The terminal includes: a memory, a processor, and an Ascend computing power-driven dynamic link joint optimization program stored in the memory and executable on the processor. When the Ascend computing power-driven dynamic link joint optimization program is executed by the processor, the steps of the Ascend computing power-driven dynamic link joint optimization method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a dynamic link joint optimization program driven by Ascend computing power. When the dynamic link joint optimization program driven by Ascend computing power is executed by a processor, the steps of the dynamic link joint optimization method driven by Ascend computing power are implemented as described in any one of claims 1 to 5.

Images