| Due to the characteristics of low cos,high flexibility,and high line-of-sight probability,unmanned aerial vehicles(UAVs)can play an important role in 5G/6G networks.However,due to the low load capacity,UAVs are limited by the battery capacity and the antennas power.UAVs-enabled air-to-air(A2A)communications also face several common challenges in ensuring service time,transmission rate and communication range.To address the above issues,the specific works carried out are as follows:(1)An A2 A communication system framework with collaborative beamforming(CB)is constructed,in which two UAVs swarms constitute two virtual antenna arrays by using CB for long-distance communication that was originally inaccessible.Considering the system lifetime and antennas power of the UAVs-enabled A2 A communication system are extremely limited in practical applications,an A2 A communication multi-objective optimization problem(A2ACMOP)is formulated for improving duplex transmission rates and minimize the total energy consumptions.According to the line-of-sight propagation model,array antenna gain model,and energy consumption model,the objectives are mainly affected by the UAVs excitation currents,the locations,and the status of receivers.Due to the large number of independent UAVs in the virtual antenna arrays and the flexibility of UAVs in three-dimensional space,A2 ACMOP is a large-scale hybrid nonlinear optimization problem that is difficult to solve by traditional multi-objective optimization algorithms.Moreover,duo to the objectives compete with each other,and the problem is NP-hard,it is difficult to convert the problem into multiple single objective optimization problems by optimizing each objective in turn to obtain the optimal solution.(2)The non-dominated sorting genetic algorithm-III(NSGA-III)lacks an effective population initialization method and an effective population update method for A2 ACMOP,resulting in premature convergence of NSGA-III.Therefore,an improved non-dominated sorting genetic algorithm-III(INSGA-III)is proposed for A2 ACMOP.Driven by the knowledge of A2 ACMOP,INSGA-III fully utilizes the physical meaning and correlation of objective functions and various decision variables.In the initialization phase of the solution,the solution initialization method of oppositionbased learning based on the objective functions can improve the diversity and performance of the initial population.During the iteration process,the discrete and continuous update operators based on the location information of UAVs in CB and the continuous value update strategy based on the ant lion optimizer enhance the directionality of the random algorithm in exploring solution spaces,accelerating the search process of large-scale hybrid solution spaces.For the iterative results,the black hole operators designed based on electromagnetic principles and CB prior knowledge dynamically adjusts the topology of the virtual antenna array from a global perspective,making the final UAV deployment plan have better spatial aggregation and topology.The above improvements improve the convergence speed and performance of the solutions.(3)Due to the computing complexity of INSGA-III is dominated by the communication model,a residual neural network agent model(Rate Res)is proposed which can quickly calculate the transmission rates and improve optimization efficiency.An improved residual structure and the batch normalization(BN)layers are added to the Rate Res.Inspired by the calculation process of the original communication model,the improved residual structure changes the way identity mapping works to map the original input data to different deep neural network layers,enabling deep neurons to have advanced features while avoiding the loss of important information caused by feature extraction,and mitigating model degradation;The BN layers effectively eliminate the internal covariate shift of different batches of data randomly generated by A2 ACMOP during training and the impact of different dimensions. |
PDF Full Text Request