Numerous surrogate-assisted evolutionary algorithms have been proposed for expensive optimization problems. However, each surrogate model has its own characteristics and different applicable situations, which caused a serious challenge for model selection. To alleviate this challenge, this paper proposes an adaptive surrogate-assisted particle swarm optimization (ASAPSO) algorithm by effectively combining global and local surrogate models, which utilizes the uncertainty level of the current population state to evaluate the approximation ability of the surrogate model in its predictions. In ASAPSO, the transformation between local and global surrogate models is controlled by an adaptive Gaussian distribution parameter with a gauge of the advisability to improve the search process with better local exploration and diversity in uncertain solutions. Four expensive optimization benchmark functions and an airfoil aerodynamic real-world engineering optimization problem are utilized to validate the effectiveness and performance of ASAPSO. Experimental results demonstrate that ASAPSO has superiority in terms of solution accuracy compared with state-of-the-art algorithms.
In response to the issues of unreachable targets, local minima, and insufficient real-time performance in drone path planning in urban low-altitude complex scenarios, this paper proposes a fusion algorithm based on digital twin, integrating LSTM (long short-term memory), DQN (Deep Q-Network), and APF (artificial potential field). The algorithm relies on a twin system, integrating multi-sensor fusion technology and Kalman filtering to input obstacle information and UAV trajectory predictions into the DQN, which outputs action decisions for intelligent obstacle avoidance. Additionally, to address the blind search problem in trajectory planning, the algorithm introduces exploration rewards and heuristic reward components, as well as adding velocity and acceleration compensation terms to the attraction and repulsion functions, reducing the path deviation of UAVs during dynamic obstacle avoidance. Finally, to tackle the issues of insufficient training sample size and simulation accuracy, this paper leverages a digital twin platform, utilizing a dual feedback mechanism from virtual and physical environments to generate a large number of complex urban scenario samples. This approach effectively enhances the diversity and accuracy of training samples while significantly reducing the experimental costs of the algorithm. The results demonstrate that the LSTM-DQN-APF algorithm, combined with the digital twin platform, can significantly improve the issues of unreachable goals, local optimality, and real-time performance in UAV operations in complex environments. Compared to traditional algorithms, it notably enhances path planning speed and obstacle avoidance success rates. After thorough training, the proposed improved algorithm can be applied to real-world UAV systems, providing reliable technical support for applications such as smart city inspections and emergency rescue operations.
Network latency is a crucial factor affecting the quality of communications networks due to the irregularity of vehicular traffic. To address the problem of performance degradation or instability caused by latency in vehicular networks, this paper proposes a time delay prediction algorithm, in which digital twin technology is employed to obtain a large quantity of actual time delay data for vehicular networks and to verify autocorrelation. Subsequently, to meet the prediction conditions of the ARMA time series model, two neural networks, i.e., Radial basis function (RBF) and Elman networks, were employed to construct a time delay prediction model. The experimental results show that the average relative error of the RBF is 7.6%, whereas that of the Elman-NN is 14.2%. This indicates that the RBF has a better prediction performance, and a better real-time performance than the Elman-NN.
In response to the issue of low positioning accuracy and insufficient robustness in small UAVs (unmanned aerial vehicle) caused by sensor noise and cumulative motion errors during flight in complex environments, this paper proposes a multisource, multimodal data fusion method. Initially, it employs a multimodal data fusion of various sensors, including GPS (global positioning system), an IMU (inertial measurement unit), and visual sensors, to complement the strengths and weaknesses of each hardware component, thereby mitigating motion errors to enhance accuracy. To mitigate the impact of sudden changes in sensor data, a high-fidelity UAV model is established in the digital twin based on the real UAV parameters, providing a robust reference for data fusion. By utilizing the extended Kalman filter algorithm, it fuses data from both the real UAV and its digital twin, and the filtered positional information is fed back into the control system of the real UAV. This enables the real-time correction of UAV positional deviations caused by sensor noise and environmental disturbances. The multisource, multimodal fusion Kalman filter method proposed in this paper significantly improves the positioning accuracy of UAVs in complex scenarios and the overall stability of the system. This method holds significant value in maintaining high-precision positioning in variable environments and has important practical implications for enhancing UAV navigation and application efficiency.
Unmanned surface vehicles (USVs) face significant challenges in long-term operations in complex and dynamic marine environments. These include abnormal attitudes, low accuracy in navigation attitude prediction, and difficulties in maintaining operational stability and equipment safety. To address these issues, this paper proposed a USV navigation attitude prediction method that integrates Unscented Kalman Filtering (UKF) with a digital twin model. First, a three-degree-of-freedom mathematical model is constructed based on the motion characteristics of the USV to establish an initial digital twin model. Then, the UKF algorithm is improved with a dynamic sliding window approach and integrated with real vessel experimental data to achieve dynamic model parameter updates, further enhancing prediction accuracy. The updated twin model is subsequently used for USV navigation attitude prediction. Experimental results demonstrate that this method significantly improves prediction accuracy and robustness, even under complex sea conditions and sensor data loss, providing crucial support for the safety and reliability of USV autonomous navigation.
Due to the advantages of a centralized critic to estimate the Q-function value and decentralized actors to optimize the agents’ policies, counterfactual multi-agent (COMA) stands out in most multi-agent reinforcement learning (MARL) algorithms. The sharing of policy parameters can improve sampling efficiency and learning effectiveness, but it may lead to a lack of policy diversity. Hence, to balance parameter sharing and diversity among agents in COMA has been a persistent research topic. In this paper, an effective training method for a COMA policy network based on a differential evolution (DE) algorithm is proposed, named DE-COMA. DE-COMA introduces individuals in a population as computational units to construct the policy network with operations such as mutation, crossover, and selection. The average return of DE-COMA is set as the fitness function, and the best individual of policy network will be chosen for the next generation. By maintaining better parameter sharing to enhance parameter diversity, multi-agent strategies will become more exploratory. To validate the effectiveness of DE-COMA, experiments were conducted in the StarCraft II environment with 2s_vs_1sc, 2s3z, 3m, and 8m battle scenarios. Experimental results demonstrate that DE-COMA significantly outperforms the traditional COMA and most other multi-agent reinforcement learning algorithms in terms of win rate and convergence speed.