| 1. |
- Geng, Zhaoquan, et al.
(författare)
-
Zero-Shot Recurrent Graph Neural Networks for Beam Prediction in Non-Terrestrial Networks
- 2022
-
Ingår i: 2022 IEEE GLOBECOM Workshops, GC Wkshps 2022. - : Institute of Electrical and Electronics Engineers (IEEE). ; , s. 1400-1405
-
Konferensbidrag (refereegranskat)abstract
- Beam management has been considered as one of the most challenging issues in mobile communications, especially in non-terrestrial networks with high-speed low-earth orbit satellites. When the user and the satellite are moving, the satellite equipped with multiple antennas needs to sweep different beam directions periodically to provide continuous service to the user. To reduce the signaling overhead in beam sweeping, we develop a recurrent graph neural network (RGNN) to predict the next beam direction that maximizes the signal strength. Compared with state-of-the-art recurrent neural networks with gated recurrent units (GRU), RGNN reduces the number of training parameters by 99.8% by exploiting a graph representation of the beams. To improve the generalization ability of RGNN in satellite communications with dynamic antenna directions, we integrate RGNN with a first-order meta-learning algorithm. After meta training, no sample is required to fine-tune the RGNN in unseen scenarios, and this approach is referred to as zero-shot meta-learning. Our simulation results show that the RGNN outperforms the GRU in terms of the convergence time and generalization ability, and the prediction accuracy with zero-shot meta-learning can be up to 97%. Even for unseen antenna directions, instead of sweeping all the neighboring beam directions, the satellite only needs to send reference signals towards few beam directions (e.g., two out of six neighboring beam directions) according to the output of the RGNN. In this way, the signaling overhead for beam sweeping can be reduced by 66%.
|
|
| 2. |
- Peng, Xin, et al.
(författare)
-
Optimal site selection for the remote-monitoring sulfur content of ship fuels in ports
- 2022
-
Ingår i: Ocean and Coastal Management. - : Elsevier. - 0964-5691 .- 1873-524X. ; 225
-
Tidskriftsartikel (refereegranskat)abstract
- The remote monitoring method based on air-quality monitoring sensors is a common way to monitor the FSC (fuel sulfur content) of oils for ships. Considering the influences of geographical environments, atmospheric conditions, regional ship traffic flow, emission characteristics of ships, and height of monitoring sensors on the monitor station chosen, a new method was proposed to optimize the site selection for monitoring the FSC of fuel oils used by ships in waters of the port. SO2 numeric simulation and observation from sensors were integrated to estimate the FSCs. The proposed method was utilized to recommend the sites of the fixed sniffing monitoring stations in Yantian port, Shenzhen, China from June and July 2018. The results showed that the experimental stations could monitor FSCs, and the relative difference between the estimated and actual FSCs of ships was 16.34%. The proposed method for recommending sites of FSC monitoring sensors contributed to intelligently supervising air pollutants emitted from ships and fuel oils of ships in the emission control areas of China.
|
|
| 3. |
- Sui, Zhongyi, et al.
(författare)
-
An improved approach for assessing marine traffic complexity based on Voronoi diagram and complex network
- 2022
-
Ingår i: Ocean Engineering. - : Elsevier BV. - 0029-8018 .- 1873-5258. ; 266, s. 112884-
-
Tidskriftsartikel (refereegranskat)abstract
- The shipping industry is constantly expanding, and maritime traffic is increasing in numerous navigable waters. For maritime traffic safety, improving Situation Awareness (SA) is critical. However, the majority of relevant research placed a greater emphasis on the risk of a collision between two ships, but failed to capture the complexity of the global traffic situation. In this paper, a framework for evaluating marine traffic complexity is developed. With the combination of the Voronoi diagram and complex network, an improved Marine Traffic Complex Network (MTSCN) is coming into being. The geometric features of Voronoi diagram are used to divide the water area into multiple cells, setting safe distances is avoided by treating ships as neighbors and other ships. The impact of neighbors on the situation awareness and the impact of other ships on the collision risk of own ship are considered. This method can more truly demonstrate the traffic complexity in the water area. Simulations are introduced to investigate the validity of the proposed method with two ships crossing situation, three ships crossing situation and four ships crossing situation. The proposed method is further demonstrated using actual AIS data in the Yangtze River, and the correlation between complexity and other indicators has been analyzed. The result shows that the improved MTSCN may give an intuitive and accessible measure to capture the essence of maritime traffic, which will be helpful in understanding, monitoring, and controlling maritime traffic safety in the future. Additionally, the complexity based on MTSCN could enhance the SA of VTSOs by supporting decision-making.
|
|
