We present an open-source system that can optimize compressed trajectory representations for large fleets of vessels. We take into account the type of each vessel in order to choose a suitable configuration that can yield improved trajectory synopses, both in terms of approximation error and compression ratio. We employ a genetic algorithm that converges to a fine-tuned configuration per vessel type without any hyper-parameter tuning. These configurations can provide synopses that retain less than 10% of the original points with less than 20m approximation error in a real world dataset; in another dataset with 90% less samples than the previous one, the synopses retain 20% of the points and achieve less than 80m error. Additionally the level of compression can be chosen by the user, by setting the desired approximation error. Our system also supports incremental optimization by training in data batches, and therefore continuously improves performance. Furthermore, we employ a composite event recognition engine to efficiently detect complex maritime activities, such as ship-to-ship transfer and loitering; thanks to the synopses generated by the genetic algorithm instead of the raw trajectories, we make the recognition process faster while also maintaining the same level of recognition accuracy. Our extensive empirical study demonstrates the effectiveness of our system over large, real-world datasets.

我们提出了一个开源系统，可以优化大型船队的压缩轨迹表示。我们考虑到每艘船的类型，以便选择合适的配置，以在近似误差和压缩比方面产生改进的轨迹概要。我们采用了一种遗传算法，该算法在没有任何超参数调整的情况下收敛到每种容器类型的微调配置。这些配置可以提供在真实世界数据集中保留不到 10% 的原始点且近似误差小于 20m 的概要；在另一个样本比前一个少 90% 的数据集中，概要保留了 20% 的点并实现了小于 80m 的误差。此外，用户可以通过设置所需的近似误差来选择压缩级别。我们的系统还支持通过批量训练进行增量优化，从而不断提高性能。此外，我们采用复合事件识别引擎来有效检测复杂的海上活动，例如船对船转移和游荡；由于遗传算法生成的概要而不是原始轨迹，我们使识别过程更快，同时保持相同的识别精度水平。我们广泛的实证研究证明了我们的系统在大型真实世界数据集上的有效性。由于遗传算法生成的概要而不是原始轨迹，我们使识别过程更快，同时保持相同的识别精度水平。我们广泛的实证研究证明了我们的系统在大型真实世界数据集上的有效性。由于遗传算法生成的概要而不是原始轨迹，我们使识别过程更快，同时保持相同的识别精度水平。我们广泛的实证研究证明了我们的系统在大型真实世界数据集上的有效性。