Intelligent high-reliability semantic communication method oriented to low-altitude intelligent mobile network

Abstract

The invention discloses an intelligent high-reliability semantic communication method oriented to a low-altitude intelligent mobile network, and relates to a low-altitude edge computing network assisted by an unmanned aerial vehicle. According to the method, balancing between task delay of ground equipment and energy consumption of the unmanned aerial vehicle is realized in an interference environment by jointly optimizing task unloading scheduling, computing resource allocation and a three-dimensional trajectory of the unmanned aerial vehicle. For a non-convex optimization problem caused by discrete and continuous variable coupling, an accurate penalty alternating algorithm and a deep reinforcement learning optimization framework (EPA-SP3O) based on soft penalty are designed. According to the method, the average time delay and energy consumption of the system can be remarkably reduced on the premise of meeting semantic accuracy constraints, and the interference resistance and robustness in a complex urban environment are improved.

Description

Technical Field

[0001] This invention belongs to the fields of UAV communication, semantic communication, mobile edge computing and anti-interference optimization, and in particular relates to an intelligent and highly reliable semantic communication method for low-altitude intelligent mobile networks in urban low-altitude scenarios. Background Technology

[0002] With the rapid popularization of smart city applications such as autonomous driving and augmented reality (AR), the demand for low-latency and high-performance computing services in urban environments is growing exponentially. Limited by computing resources and energy consumption, many edge devices (such as surveillance cameras and traffic signal controllers) struggle to independently complete complex computing tasks. Mobile edge computing (MEC) has been proposed as an effective solution [C. Xu, C. Zhan, J. Liao and J. Gong, "ComputationThroughput Maximization for UAV-Enabled MEC with Binary ComputationOffloading," ICC 2022 - IEEE International Conference on Communications ,Seoul, Korea, Republic of, 2022, pp. 4348-4353, doi: 10.1109 / ICC45855.2022.9838879.] and [Q. Wu, M. Cui, G. Zhang, F. Wang, Q. Wu and X. Chu, "Latency Minimization for UAV-Enabled URLLC-Based Mobile Edge ComputingSystems," in IEEE Transactions on Wireless Communications [, vol. 23, no. 4, pp. 3298-3311, April 2024, doi: 10.1109 / TWC.2023.3307154.] MEC (Multi-access Edge Computing) offloads computing, storage, and networking functions to edge servers closer to the data source, enabling tasks to be processed at the edge closer to the terminal, thereby reducing communication latency, alleviating pressure on the core network, and improving service quality. By offloading tasks to nearby edge servers, terminal devices can significantly reduce their own computational burden while ensuring timely task execution.

[0003] However, in practical deployments, MEC faces significant challenges such as scarce spectrum resources. With the rapid increase in data volume and the widespread application of large-scale models, the bandwidth requirements of traditional bit-level transmission-based MEC systems are further exacerbated. To alleviate this problem, Semantic Communication (SemCom) has been proposed as a new communication paradigm [C. Chaccour, W. Saad, M. Debbah, Z. Han and H. Vincent Poor, "Less Data, More Knowledge: Building Next-Generation Semantic Communication Networks," in IEEE Communications Surveys&Tutorials [, vol. 27, no. 1, pp. 37-76, Feb. 2025, doi: 10.1109 / COMST.2024.3412852.]. Semantic communication emphasizes ensuring information availability at the semantic level, aiming to maximize the receiver's correct understanding of task-related semantics. This is achieved by semantically encoding the raw data at the source, transmitting only task-related semantic features, and then performing semantic recovery at the receiver. This approach reduces reliance on bandwidth resources and improves transmission efficiency. Existing research mainly focuses on offloading terminal tasks to ground-edge servers via semantic communication, achieving offloading performance superior to traditional bit-level transmission schemes. Reference [Z. Ji, Z. Qin, X. Tao and Z. Han, "Resource Optimization for Semantic-Aware Networks With Task Offloading," in...] IEEE Transactions on Wireless Communications , vol. 23, no. 9, pp. 12284-12296, Sept. 2024, doi: 10.1109 / TWC.2024.3390407.] and literature [Z. Ji and Z. Qin, "Energy-Efficient Task Offloading for Semantic-Aware Networks," ICC 2023 - IEEE International Conference on Communications [Rome, Italy, 2023, pp. 3584-3589, doi: 10.1109 / ICC45041.2023.10279646.] proposes a framework for integrating semantic communication into the MEC architecture and achieves significant improvements in task offloading efficiency.

[0004] However, deploying fixed base stations as edge servers in urban areas for terrestrial MEC still faces inherent limitations. In particular, the high density of infrastructure and buildings in urban environments leads to signal congestion, shadowing, and multipath fading, reducing the stability of communication links and consequently affecting the reliability of task offloading. Furthermore, fixed edge servers struggle to meet the regional fluctuations in computing demands in terms of service coverage and deployment flexibility. Against this backdrop, unmanned aerial vehicles (UAVs), with their strong line-of-sight (LoS) communication links, high mobility, and wide coverage capabilities, are gradually becoming an important enabling technology for low-altitude MEC networks [G. Cheng, X. Song, Z. Lyu and J. Xu, "Networked ISAC for Low-Altitude Economy: Coordinated Transmit Beamforming and UAV Trajectory Design," in IEEE Transactions on Communications [, vol. 73, no. 8, pp. 5832-5847, Aug. 2025, doi: 10.1109 / TCOMM.2025.3541027.]. As flexible aerial edge servers, drones offer new possibilities for dynamic and efficient edge computing services. Existing research includes [M. Zheng, H. Yang, S. Liu, K. Lin, L. Xiao and Z. Han, "Reliable Semantic Communication With QoE-Driven ResourceScheduling for UAV-Assisted MEC," in...] IEEE Transactions on Vehicular Technology , vol. 74, no. 7, pp. 11484-11489, July 2025, doi: 10.1109 / TVT.2025.3542775.] and [S. Liu, H. Yang, M. Zheng, L. Xiao, Z. Xiong and D.Niyato, "UAV-Enabled Semantic Communication in Mobile Edge Computing UnderJamming Attacks: An Intelligent Resource Management Approach," in IEEE Transactions on Wireless Communications[F. Pervez, A. Sultana, C. Yang and L. Zhao, "Energy and Latency Efficient Joint Communication and Computation Optimization in a Multi-UAV-Assisted MEC Network," vol. 23, no. 11, pp. 17493-17507, Nov. 2024, doi: 10.1109 / TWC.2024.3454073.] This paper explores a UAV-assisted MEC architecture that integrates semantic communication and jointly optimizes UAV trajectory and resource allocation to improve communication efficiency. However, most of these studies focus on single-modal tasks and have not fully considered multi-modal task scenarios in complex urban environments. On the other hand, existing UAV-assisted MEC systems generally simplify the UAV's movement model, considering only two-dimensional trajectory design [F. Pervez, A. Sultana, C. Yang and L. Zhao, "Energy and Latency Efficient Joint Communication and Computation Optimization in a Multi-UAV-Assisted MEC Network," in IEEE Transactions on Wireless Communications , vol. 23, no. 3, pp. 1728-1741, March 2024, doi: 10.1109 / TWC.2023.3291692. ] and [F. Lu et al ., "Resource andTrajectory Optimization for UAV-Relay-Assisted Secure Maritime MEC," in IEEE Transactions on Communications [, vol. 72, no. 3, pp. 1641-1652, March 2024, doi: 10.1109 / TCOMM.2023.3330884.]. Low-altitude flight reduces communication distance and enhances channel gain, improving link quality; however, it also increases the risk of signal congestion and collisions with buildings. Conversely, high-altitude flight expands coverage and avoids obstacles, but leads to higher propulsion energy consumption and poorer link reliability. Therefore, achieving energy-efficient and adaptive 3D trajectory planning in complex urban scenarios is crucial for reliable semantic task offloading.

[0005] Furthermore, UAV line-of-sight channels are susceptible to malicious interference, and semantic communication is sensitive to information integrity; impaired transmission significantly reduces semantic accuracy [H. Yang, K. Lin, L. Xiao, Y. Zhao, Z. Xiong and Z. Han, "Energy Harvesting UAV-RIS-Assisted Maritime Communications Based on DeepReinforcement Learning Against Jamming," in IEEE Transactions on Wireless Communications [Y. Zeng and R. Zhang, "Energy-Efficient UAV Communication With TrajectoryOptimization," in vol. 23, no. 8, pp. 9854-9868, Aug. 2024, doi: 10.1109 / TWC.2024.3367034.], thereby weakening task execution and overall network stability. Therefore, improving the anti-interference capability of semantic communication in UAV-enabled MEC networks is a key link in ensuring system reliability. Meanwhile, there is a contradiction between task completion latency and UAV energy consumption. To minimize transmission latency, UAVs tend to fly closer to the target terminal device to improve link quality and task completion efficiency; however, this approach usually requires more frequent maneuvers, thus increasing energy consumption [Y. Zeng and R. Zhang, "Energy-Efficient UAV Communication With TrajectoryOptimization," in IEEE Transactions on Wireless Communications [, vol. 16, no.6, pp. 3747-3760, June 2017, doi: 10.1109 / TWC.2017.2688328.]. Therefore, in complex environments, how to balance task latency and UAV energy consumption while ensuring semantic accuracy is a key challenge for the deployment of UAV-enabled multimodal task offloading systems. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in current low-altitude intelligent mobile networks, such as insufficient anti-interference capability, difficulties in optimization and coordination, and inadequate trade-off between semantic accuracy and system performance, during semantic communication and task offloading. This invention provides an intelligent and highly reliable semantic communication method for low-altitude intelligent mobile networks. This method utilizes a penalty iterative algorithm based on convex optimization and a deep reinforcement learning algorithm to optimize the three-dimensional spatial trajectory of the UAV, task offloading scheduling, and computational resources under semantic similarity and computational resource constraints. This reduces task transmission latency and UAV energy consumption in the low-altitude mobile edge computing network and improves the UAV's anti-interference performance.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solutions.

[0008] A highly reliable intelligent semantic communication method for low-altitude intelligent connected mobile networks includes the following steps:

[0009] 1) System parameter initialization: Set up multiple ground devices randomly distributed on the ground plane, configure jammers in fixed positions, and configure the flight and communication parameters of the UAV;

[0010] 2) Time slot allocation: Divide the total service time of the UAV evenly into slots. Each time slot records the three-dimensional position coordinates of the UAV in each time slot, and the system status observation, action selection and related parameter update operations are completed sequentially within each time slot;

[0011] 3) Dual-modal task generation and link modeling: Each ground device generates a dual-modal task consisting of text and images in each time slot and sets a minimum semantic accuracy threshold. The communication link is modeled using an elevation angle-dependent line-of-sight / non-line-of-sight model, and the link transmission rate is determined by combining transmit power, interference, etc.

[0012] 4) Task processing latency calculation: If the task is not unloaded, the total task latency is determined by the ratio of the total computing resources required by the text and image tasks to the maximum CPU frequency of the ground equipment; if the task is unloaded, the total task latency is the sum of the transmission latency and the UAV computing latency, where the transmission latency is the ratio of the task data volume to the link transmission rate, and the UAV computing latency is determined by the ratio of the total computing resources of the task to the maximum CPU frequency of the UAV.

[0013] 5) System optimization objective setting: The optimization objective is to minimize the weighted sum of the average task latency of ground equipment and the total flight energy consumption of UAV. The weights of the two objectives are adjusted by the trade-off coefficient and the dimensional balance parameter.

[0014] 6) Exactly Penalized Alternating Iterative Optimization (EPA): Introducing auxiliary penalty variables and penalty coefficients, the binary constraint of task offloading scheduling is transformed into a continuous constraint, and an equivalent objective function containing penalty terms is constructed; the task offloading scheduling variables and auxiliary penalty variables are solved alternately until convergence;

[0015] 7) Soft-penalized deep reinforcement learning (SP3O): Based on the task offloading scheduling results of the output of precise penalty alternating iterative optimization, a Markov decision process is constructed. The proximal policy optimization (PPO) algorithm is used to construct an objective function that includes a semantic precision soft penalty term and a computational resource constraint soft penalty term. The UAV trajectory and computational resources are optimized, and convergence is achieved through network training and experience replay.

[0016] In step 1), the ground equipment is a terminal randomly and uniformly distributed in a square planar area, and the jammer at the fixed position continuously performs broadband interference on the semantic offload link with constant power.

[0017] In step 2), the time slot length is the total service time divided by the number of time slots, and each time slot completes system status observation, action selection and parameter update.

[0018] In step 3), the bimodal task includes a text task and an image task, and a minimum semantic accuracy threshold is used to constrain the task recovery quality; the communication link is modeled based on an elevation angle-dependent line-of-sight and non-line-of-sight link model.

[0019] In step 4), the transmission latency is the ratio of the task data volume to the link transmission rate, and the UAV computing latency is the ratio of the total task computing resources to the maximum CPU frequency of the UAV.

[0020] In step 5), the optimization objective function is composed of the weighted sum of the average task latency of ground equipment and the energy consumption of UAV, and the trade-off factor is used to adjust the priority of the trade-off optimization between latency and energy consumption.

[0021] In step 6), the auxiliary penalty variable and penalty coefficient are used to transform the original binary unloading constraint into a continuous constraint, thereby constructing an equivalent optimization problem and achieving convergence through alternating updates.

[0022] In step 7), the Markov decision process is modeled with a triplet of state, action, and reward, and the semantic precision and computational resource soft penalty terms ensure that the optimization results maintain a balance between accuracy and feasibility.

[0023] Compared with the prior art, the present invention has the following outstanding advantages:

[0024] Unlike traditional UAV semantic communication anti-interference schemes, this invention addresses the semantic transmission requirements of bimodal tasks (text + image). To balance the trade-off between device task latency and UAV energy consumption, it jointly optimizes task offloading metrics, computational resources, and the UAV's 3D trajectory, minimizing the weighted sum of average task node completion time and UAV energy consumption. To solve this joint optimization problem, an EPA algorithm based on precise penalty and an SP3O algorithm based on deep reinforcement learning are proposed. Specifically, a balanced constraint precise penalty framework is constructed to handle discrete offloading decisions, transforming them into differentiable forms while maintaining solution feasibility. Subsequently, by modeling the original non-convex problem as an MDP and employing the PPO algorithm based on a soft penalty method to handle the high-dimensional action space, the coupling problem between the UAV's 3D trajectory and computational resources is resolved. Simulation results show that, due to the dynamic optimization of the UAV's 3D trajectory, the proposed algorithm outperforms other comparative schemes in balancing task completion time and UAV energy consumption, and reveals an interesting trade-off between elevation angle and distance. Attached Figure Description

[0025] Figure 1 This is a system model diagram of the intelligent and highly reliable semantic communication method for low-altitude intelligent connected mobile networks as described in an embodiment of the present invention.

[0026] Figure 2 This is a comparison chart of the rewards of the reliable semantic communication algorithm and the comparison algorithm described in this embodiment of the invention at different iteration numbers.

[0027] Figure 3 This diagram illustrates the trade-off between average task completion latency and UAV energy consumption in the reliable semantic communication method described in this embodiment of the invention. It demonstrates how the system dynamically adjusts its behavior based on trade-off coefficients to achieve a balance between two competing objectives. In practical engineering applications, the weighting factors can be adjusted according to specific requirements for task latency and energy consumption to adapt to different task priorities.

[0028] Figure 4 This is a top-down view of the trajectory of a UAV after anti-interference optimization, which is the reliable semantic communication method described in this embodiment of the invention.

[0029] Figure 5 This is a 3D trajectory diagram of a UAV after anti-interference optimization for the reliable semantic communication method described in this embodiment of the invention. The diagram illustrates how the UAV-assisted reliable semantic communication method avoids interference by optimizing the UAV's flight trajectory. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments will be used in conjunction with the accompanying drawings to further illustrate the invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Rather, the invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims.

[0031] Figure 1 This is a system model diagram of the intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks described in an embodiment of the present invention. The system includes five ground devices capable of generating tasks (such as text translation or image recognition). These tasks can be processed locally by the devices or, after semantic encoding, offloaded to a UAV for processing. Simultaneously, a static malicious jammer with semantic attack capabilities attempts to disrupt the task offloading process by locking onto the semantic features of the transmission. This jammer is assumed to be in a fixed position with constant transmission power, corresponding to a worst-case continuous interference scenario. To ensure reliable task offloading under such interference conditions, the UAV optimizes its three-dimensional trajectory to mitigate the impact of continuous interference and enhance transmission robustness.

[0032] The embodiments of the present invention include the following steps:

[0033] Step 1: Configure the system to include Several ground devices, randomly and evenly distributed in On a square ground plane of a certain size; a jammer, positioned at a fixed location, continuously interferes with the equivalent broadband semantic offloading link between ground equipment and the UAV at a constant jamming power. The jamming power is... The flight altitude range of the drone is within arrive The total service time provided by the drone is The maximum horizontal and vertical speeds of the drone are respectively and .

[0034] Step 2: Set the total service time interval Average score One time slot, of which Let be the time slot length; denoted as _i_1_1_2_2_3_2_4_3_2_4_3_2_4_3_3_4 ... The location of the drone in each time slot is , It is the first Two-dimensional horizontal position vector of a time-slot UAV It is the transpose of this two-dimensional horizontal position vector (used for concatenation with the vertical height). This refers to the vertical altitude of the UAV in that time slot. It represents a 3x1 vector space defined over the real number field; it performs state observation, action selection, and parameter updates within each time slot.

[0035] Step 3: Set system bandwidth The noise power is The reference channel power gain is The link between the UAV and ground equipment adopts a line-of-sight / non-line-of-sight probability model, and its probability is related to the elevation angle of the ground equipment based on urban environment experience; within each time slot, the transmission rate of the task offloaded from the ground equipment to the UAV is:

[0036]

[0037] in, The transmission power of ground equipment. Indicates the first Within the first time slot Channel gain between ground equipment and drones Indicates the first The interference power of the jammer on the link within each time slot.

[0038] Step 4: For each ground device, generate a bimodal task combining text and images within each time slot, and consider whether to offload it to a UAV for auxiliary computation; the number of sentences and images in the text task are... and The average semantic entropy of each sentence and each image is and The average number of semantic symbols per sentence and per image is and The minimum semantic precision threshold is .

[0039] Step 5: If in a time slot If internal tasks do not consider unloading (i.e., they are computed locally), then task unloading scheduling is... The average computational resources of the ground equipment for each sentence and each image are set to 0; and The maximum CPU frequency for each device is set to The maximum CPU frequency of the drone is set to If the task is not unloaded, the local task latency is calculated.

[0040] Step 6: If in a time slot If the internal tasks are offloaded to the drone for auxiliary computation, then task offloading scheduling is implemented. The value is 1; the task latency is the sum of the transmission latency between the device and the drone and the computation latency of the drone; at this time, the upper limits of drone speed, altitude, flight energy consumption and CPU are simultaneously in effect.

[0041] Step 7: The system optimization objective is to minimize the weighted sum of the average task completion latency of ground equipment and the total energy consumption of the UAV, i.e. ,in, This represents the total number of ground equipment. Representative equipment Task completion delay, Represents the total flight energy consumption of the drone. It is a parameter with balanced dimensions. It is a weighted coefficient; the system minimizes this weighted sum by jointly optimizing the UAV's three-dimensional flight trajectory, task offloading scheduling, and computing resources, while ensuring semantic accuracy constraints.

[0042] Step 8: Propose an Alternating Iterative Optimization (EPA) framework with precise penalties to address the duality of task offloading scheduling variables; and introduce auxiliary penalty variables. And set the penalty coefficient to By using the Lagrange dual function transformation, the bivariate constraint is equivalent to a differentiable continuous constraint, and a regularization penalty term is added to the objective function, making the solution converge to... The set is used to ensure the feasibility of the original problem; the constructed equivalent objective function is:

[0043]

[0044] in, It is a trade-off coefficient. This represents the total number of ground equipment. Representative equipment Task completion delay, Represents the total flight energy consumption of the drone. It is a parameter with balanced dimensions. It is the penalty coefficient. The total number of time slots, It is the first The device in the Task offloading scheduling variables for each time slot It is an auxiliary penalty variable that is introduced.

[0045] Step 9: Propose a soft-penalty-based deep reinforcement learning algorithm (SP3O) to address the high-dimensional continuity of trajectory and computational resources; using the binary task offloading scheduling obtained in Step 8, construct a Markov decision process (MDP) from the continuous variables (UAV three-dimensional trajectory and computational resources), and employ a proximal policy optimization algorithm (PPO) based on deep reinforcement learning to intelligently optimize the trajectory and computational resources; the constructed equivalent objective function is:

[0046]

[0047] in, and This is a trade-off factor for soft penalty items; For semantic precision; For drones Time slot position Position of the first time slot The Euclidean distance is used to constrain the range of the drone's trajectory and avoid unreasonable trajectory dispersion.

[0048] Step 10: Initialize the maximum number of iterations for the exact penalty alternating iterative optimization framework Set trade-off factors Let the initial number of alternating iterations be . Initialize Actor network parameters and Critic network parameters Initialize the experience replay pool The number of samples in a small batch, and other relevant parameters.

[0049] Step 11: If If the EPA is not optimized, proceed to the alternating iterative optimization stage (step 12); otherwise, proceed to the deep reinforcement learning stage (step 16).

[0050] Step 12: Given a local point Under the given conditions, solve for the task unloading scheduling variables. .

[0051] Step 13: Based on task unloading scheduling variables Solve for the auxiliary penalty variable. .

[0052] Step 14: Update the number of iterations .

[0053] Step 15: Repeat steps 12-15 until the EPA algorithm converges.

[0054] Step 16: Set the number of training rounds to... For each round number To obtain the system's initial observations of the environment. .

[0055] Step 17: For each time slot Observe the current position of the drone Task unloading and scheduling Current energy consumption of drones Link channel information between the UAV and ground equipment And the jammer's channel information for the link between the drone and the equipment. Construct the system state vector .

[0056] Step 18: Under the self-guided exploration strategy, based on the strategy Select with exploratory noise action Drones performing actions It can choose to fly to the next location or hover, and adjust its computing resources.

[0057] Step 19: Calculate the reward function based on the performed action. That is, the objective function constructed in step 9.

[0058] Step 20: Observe the new state of the environment feedback , experience Stored in the experience replay pool middle.

[0059] Step 21: If time slot Less than If the condition is met, return to step 17; otherwise, continue to step 22.

[0060] Step 22: Set the random sampling method from the experience replay pool based on the number of samples in the small batch sampling. According to the near-end strategy objective, namely:

[0061]

[0062] Update the Actor network, where It is a small truncation parameter for the magnitude of the control strategy update. For Actor network parameters, For the desired operation, This represents the generalized advantage estimation (GAE). It is the probability ratio of the old and new strategies. It's a cropping operation. It is a minimization operation. This objective function ensures that the new policy does not deviate excessively from the old policy, thereby avoiding instability during training.

[0063] Step 23: Return to the value objective, i.e.:

[0064]

[0065] Update the Critic network, where This is an estimated value. Let $\frac{ ...

[0066] Step 24: Repeat steps 16-23 until the SP3O algorithm converges.

[0067] Simulation results:

[0068] Experimental results are as follows Figure 2 , Figure 3 , Figure 4 and Figure 5 As shown. Figure 2 The average reward convergence of EPA-SP3O, EPA-TD3, EPA-SD3, EPA-DDPG, HEPA-SP3O, SEPA-SP3O, and the Random trajectory method were compared. The results show that, except for the Random and SEPA-SP3O methods, the average reward of the other algorithms gradually increases with the number of training epochs and then stabilizes after a certain number of epochs. Among them, the EPA-SP3O method proposed in this invention achieves the highest average reward after convergence, verifying the effectiveness of the proposed method. EPA-TD3, EPA-SD3, and EPA-DDPG methods also show convergence, but their reward values ​​are consistently lower than those of EPA-SP3O. HEPA-SP3O, due to its lack of optimization for UAV altitude and lower motion dimension, has the fastest convergence speed, but its final performance is inferior to EPA-SP3O. The Random and SEPA-SP3O methods have the lowest overall performance due to the lack of an effective trajectory optimization mechanism. These results fully demonstrate that UAV trajectory optimization plays a crucial role in improving the overall system performance. Figure 3 Indicates the influence of trade-off factors As the value increases from 0.1 to 0.9, the system exhibits a clear trade-off between "task completion latency" and "drone energy consumption." At this time, the system focuses more on reducing the average task completion latency of ground equipment. The task unloading and computation processes are accelerated, resulting in a significant decrease in average completion latency, but the energy consumption of the UAV increases accordingly. At that time, the system optimization objective shifted to reducing UAV energy consumption, resulting in a significant reduction in energy consumption, but a slight increase in the average task completion latency of ground equipment. This result indicates that adjusting the trade-off factors can flexibly select between prioritizing task latency reduction or energy consumption reduction based on different application requirements, thereby achieving an adaptive performance balance. Figure 3 This demonstrates how the system dynamically adjusts its behavior based on trade-off coefficients to achieve a balance between two competing objectives. In practical engineering applications, the weighting factors can be adjusted according to the specific requirements of task latency and energy consumption to adapt to different task priorities. Figure 4 The top view represents the optimized 2D UAV trajectory, while Figure 5 This represents the corresponding three-dimensional trajectory. Experimental results show that the method proposed in this invention effectively improves the performance of horizontal trajectory planning and altitude control, making the flight path smoother and more efficient. In an interference-free state (i.e., Under these conditions, drones tend to get closer to ground equipment. To maximize communication quality. Conversely, when malicious interference exists (i.e. When the drone actively deviates... The drone adaptively modulates its trajectory by adjusting its altitude to bypass interference areas. This adaptive behavior allows it to effectively mitigate the impact of interference and maintain reliable mission offload. Notably, the drone exhibits an undulating flight path, maintaining a relatively high altitude until approaching the ground receiving point. This phenomenon reveals a dynamic trade-off between distance and elevation angle: altitude affects elevation angle, which in turn affects the probability of line-of-sight communication. In other words, the dominant factors affecting air-to-ground channel quality change dynamically throughout the drone's flight: as the drone approaches ground equipment, the reduced distance becomes crucial, leading to a decrease in altitude; conversely, as the drone moves away from ground equipment, the influence of elevation angle increases, prompting the drone to climb.

[0069] In summary, the experimental results demonstrate that the proposed solution can achieve a flexible balance between task completion latency and energy consumption under various trade-off factors, and its convergence performance is significantly superior to the comparative algorithms. This method effectively improves the system's anti-interference capability and resource optimization level in complex low-altitude environments and under interference conditions, providing an efficient and scalable optimization approach for UAV-enabled semantic communication and edge computing networks.

[0070] The above embodiments are merely preferred embodiments of the present invention and should not be considered as limiting the scope of the present invention. All equivalent variations and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A highly reliable intelligent semantic communication method for low-altitude intelligent connected mobile networks, characterized in that... Includes the following steps: 1) System parameter initialization: setting A ground device randomly distributed on the ground plane is configured with a jammer that emits broadband interference at a fixed position and constant power to initialize the flight and communication parameters of the UAV. 2) Time slot allocation: Divide the total service time of the UAV evenly into slots. Each time slot records the three-dimensional position coordinates of the UAV in each time slot, and the system status observation, action selection and related parameter update operations are completed sequentially within each time slot; 3) Dual-modal task generation and link modeling: Each ground device generates a dual-modal task consisting of text and images in each time slot and sets a minimum semantic accuracy threshold. The communication link is modeled using an elevation angle-dependent line-of-sight / non-line-of-sight model, and the link transmission rate is determined by combining the transmit power and interference. 4) Task processing latency calculation: If the task is not unloaded, the total task latency is determined by the ratio of the total computing resources required by the text and image tasks to the maximum CPU frequency of the ground equipment. If the task is selected to be unloaded, the total task latency is the sum of the transmission latency and the drone computing latency. The transmission latency is the ratio of the task data volume to the link transmission rate, and the drone computing latency is determined by the ratio of the total task computing resources to the drone's maximum CPU frequency. 5) System optimization objective setting: The optimization objective is to minimize the weighted sum of the average task latency of ground equipment and the total flight energy consumption of UAV. The weights of the two objectives are adjusted by the trade-off coefficient and the dimensional balance parameter. 6) Exactly Penalized Alternating Iterative Optimization (EPA): Introducing auxiliary penalty variables and penalty coefficients, the binary constraint of task offloading scheduling is transformed into a continuous constraint, and an equivalent objective function containing penalty terms is constructed; the task offloading scheduling variables and auxiliary penalty variables are solved alternately until convergence; 7) Soft-penalized deep reinforcement learning (SP3O): Based on the task offloading scheduling results of the output of precise penalty alternating iterative optimization, a Markov decision process is constructed. The proximal policy optimization (PPO) algorithm is used to construct an objective function that includes a semantic precision soft penalty term and a computational resource constraint soft penalty term. The UAV trajectory and computational resources are optimized, and convergence is achieved through network training and experience replay.

2. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 1), the ground equipment is a terminal randomly and uniformly distributed in a square planar area, and the jammer at the fixed position continuously performs broadband interference on the semantic offload link with constant power.

3. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 1), the flight and communication parameters of the UAV include: the maximum horizontal speed of the UAV. Maximum vertical speed of the drone Maximum flight altitude of drones Minimum flight altitude of drones System bandwidth Noise power Reference channel power gain .

4. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 2), the time slot length is the total service time divided by the number of time slots, and each time slot completes system status observation, action selection and parameter update.

5. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 3), the bimodal task includes a text task and an image task, and a minimum semantic accuracy threshold is used to constrain the task recovery quality; the communication link is modeled based on an elevation angle-dependent line-of-sight and non-line-of-sight link model. The link transmission rate is: in, The transmission power of ground equipment. Indicates the first Within the first time slot Channel gain between ground equipment and drones For noise power, For system bandwidth, Indicates the first The interference power of the jammer on the link within each time slot.

6. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 4), the transmission delay is the ratio of the task data volume to the link transmission rate, and the UAV computing delay is the ratio of the total task computing resources to the maximum CPU frequency of the UAV.

7. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 5), the optimization objective function is composed of the weighted sum of the average task latency of ground equipment and the energy consumption of the UAV. The trade-off factor is used to adjust the priority of the trade-off optimization between latency and energy consumption. The optimization objective function is as follows: in, This represents the total number of ground equipment. Representative equipment Task completion delay, Represents the total flight energy consumption of the drone. It is a parameter with balanced dimensions. It is a weighted coefficient; the system minimizes this weighted sum by jointly optimizing the UAV's three-dimensional flight trajectory, task offloading scheduling, and computing resources, while ensuring semantic accuracy constraints.

8. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 6), the auxiliary penalty variable and penalty coefficient are used to transform the original binary unloading constraint into a continuous constraint, thereby constructing an equivalent optimization problem and achieving convergence through alternating updates.

9. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 6), the constructed equivalent objective function is: in, and As a trade-off factor for soft penalty items, For semantic precision, For drones Time slot position Position of the first time slot The Euclidean distance is used to constrain the range of the drone's trajectory and avoid unreasonable trajectory dispersion.

10. The intelligent high-reliability semantic communication method for low-altitude intelligent connected mobile networks as described in claim 1, characterized in that... In step 7), the Markov decision process is modeled with a triplet of state, action, and reward, and the semantic precision and computational resource soft penalty terms ensure that the optimization results maintain a balance between accuracy and feasibility.

Images