通过预测性人工智能分析乱流促进海事安全

This week marks a major milestone in oceanic studies and maritime safety after a team of engineers published a study demonstrating how AI neural networks could help to forecast and warn about rogue waves. Notably, these massive events were once seen as unpredictable and have led to major losses for centuries. Here’s what you need to know.

汹涌波涛

恶浪有很多名字，包括杀手浪、怪兽浪和极端浪。这些海浪的高度是正常海况下的两倍，可能会突然出现，使挡路者陷入极大的危险之中。纵观历史，巨浪一直是海上旅行者的民间传说，他们无疑听说过这些巨浪将海员同伴带入深海的传说。今天，这些巨浪仍然构成重大威胁。

值得注意的是，这些罕见的海洋事件在规模和破坏力上与海啸相似。然而，与人们熟知的同类事件不同，"流氓波 "可能在没有地震等重大明显事件的情况下出现。因此，这些难以预测的巨浪会对海上基础设施、旅客和其他设备造成极大的危险。

海浪是如何形成的

Until recently, there wasn’t enough data and computational power to create reliable rogue wave formation mechanisms. As such, the majority of science regarding the formation of these elusive events was based on observational data collection following events. This info has led researchers to determine the three most prominent ways that rogue waves form.

慢慢积累

Interestingly, a rogue wave can begin to form and build up without the need to combine with other waves. Strong currents like the Gulf Stream can resonate and expand to create a massive sea anomaly. Scientists studying this phenomenon often employ a method called Benjamin-Feir instability to explain the single-wave train’s expansion. This is one of the rarest forms of rogue waves captured by researchers.

多重波浪

最广为人知的 "流氓波 "的形成是由于多波干扰在最佳时机相遇，从而放大了波幅。当波浪积聚并产生共鸣时，它们会放大并加强波流和波浪的凶猛程度。你可以把这想象成滚雪球效应，每一个波浪都会被吸收到更大的动力中，形成一堵巨大的墙，随时准备摧毁路径上的任何东西。

目前确定游离波的方法

尽管这些事件有可能每年造成数十亿美元的损失，但全球并没有可靠的流氓波预报系统。确定这些事件的最常用方法依赖于实时海洋数据，这些数据只能在事件发生时预测，并通知科学家海平面的上升。这种策略在预警和安全预防方面还有很多不足之处。

汹涌波涛研究

的"根据浮标测量结果预测怪浪” study seeks to shed light on these unique ocean events. The researchers set out to create an AI neural network capable of predicting the chances of a rogue wave occurring based on the current ocean state. As part of this approach, researchers sought to determine what pre-factors occurred before the formation of rogue waves.

这种方法的目的是利用现场测量设备和先进的神经网络，展示前几次波浪与最终流氓事件之间的功能关系。更重要的是，研究小组希望捕捉并记录产生流氓波所需的各个波之间的同步阶段。

测试阶段

测试从创建神经网络开始。团队决定使用长短期记忆网络（LSTM）人工智能算法。创建这些人工智能系统是为了简化输入数据和输出数据之间的功能关系。因此，它们非常适合对复杂系统进行数据驱动预测。

海洋浮标

研究人员从位于美国东海岸和太平洋岛屿附近的 172 个浮标上收集了数据，这些浮标的深度从 20 米到 4000 米不等。测试中使用了两种浮标。一种是采样率为 1.28 Hz 的 Datawell Directional wave rider MkIII，另一种是采样率为 2.56 Hz 的 Datawell Directional wave rider MkIII。这些设备都是经过充分测试的，可提供可靠的数据，包括监测垂直位移的加速器。

Source – MATLAB

数据集

The neural network was programmed using a sample size equivalent to 880 years of consecutive data regarding pre-rogue wave conditions. In particular,  14 million 30-minute sea surface elevation measurements were combined with 40,000 sea surface elevation measurements from the same buoys. This data was then scanned to check for abnormalities. The data was then fed back into the algorithm to better train it how to spot these factors.

与海洋条件相关

研究人员希望在测试阶段尽量少用流氓波样本，以确保数据更接近真实世界的场景。随后，研究人员将样本分为非流氓波浪和流氓波浪两类，并对其进行了强化。值得注意的是，最初的数据是通过海岸数据信息计划（CDIP）与斯克里普斯海洋学研究所联合获得的。

扫描骇浪

研究人员利用马里兰大学的超级计算资源扫描了波浪样本数据。这种访问方式使他们能够利用英伟达™（NVIDIA®）A100 GPU 和本地英伟达™（NVIDIA®）Quadro P1000 GPU 来提高人工智能性能、训练时间和准确性。

成果

这项研究的结果可能会对工程师和研究人员今后看待流氓波的方式产生重大影响。利用人工智能系统，这些曾经几乎无法确定的事件能够以 75% 的准确率被挑出来。具体来说，75% 的流氓波在出现后 1 分钟内就能被预测出来。当预警时间延长到 5 分钟时，预测准确率才下降到 73%。

人工智能总共准确预测了近 3k 个流氓波。只有 855 个事件侥幸通过了人工智能的检测，使系统的准确率达到了 23%。考虑到许多研究人员认为不可能确定这些事件，这些结果为引入新的、更有效的预测系统打开了大门。

益处

This study could bring many benefits to the market. For one, it’s the first reliable rogue wave forecast system concept. It takes the idea of these events being too complex to predict and instead, introduces a very reliable and cost-effective solution that could provide valuable advanced warning to ships and offshore platforms when needed.

使用海流浮标数据

Another major benefit of this approach to determining the probability of rogue waves is that there is no need to install new sensors or systems. Buoys have already been deployed and tested for decades. As such, they provide a reliable and proven data source that has a trackable history.  This AI prediction upgrade is software. As such, it can leverage the millions of hours of data provided by these devices to improve and enhance future performance.

通用应用

他们的人工智能训练最有趣的一点是，该模型能够成功应用于原始浮标数据集之外的地点。研究小组能够以极高的概率预测两个异地浮标的流氓波风险。他们在佛罗里达州杰克逊维尔附近的 132 号浮标和洛杉矶海岸圣尼古拉斯岛附近的 067 号浮标上测试了他们的理论。结果表明，该算法可成功应用于其他地点。

自我完善

这种人工智能算法的最大好处之一是，它可以不断改进数据集，加强理解并提高性能。这些系统会随着数据的完善而不断改进。因此，这种方法提供了一种低成本、高效率的方式来提高运营水平。

研究人员

Thomas Breunung and Balakumar Balachandran were the lead researchers in the study. They achieved their goal of demonstrating a way to determine rogue waves with high accuracy globally. Notably,  The University of Maryland provided supercomputing resources and support for the team, which now seeks to improve its results by introducing more data to its model including wind speed, location, and depth. All of these factors could help to improve detection times and even provide a way to determine the height of the event.

今天就能整合这项技术的公司

许多公司可以从这项研究中立即受益。目前，有数十亿美元的海上基础设施和航运船只在海洋中穿行。这些公司肯定会投资任何有助于防止灾难性损失和死亡的技术。

1.Diamond Offshore Drilling Inc.

Diamond Offshore Drilling, Inc. (DO -2.44%)

钻石海上钻井公司（Diamond Offshore Drilling Inc.该公司于 1987 年作为 Diamond M Drilling 公司进入市场，之后经历了多次品牌重塑。目前，该公司拥有 44 个海上钻井平台，包括 32 个半潜式平台和 5 艘钻井船。

Diamond Offshore Drilling Inc. has contracts with many of the largest oil and gas companies in the world, including Hess Corporation, Petrobras, BP, and Occidental Petroleum. Its positioning and the demand for fossil fuels make this stock a strong “持有.” Notably, it has experienced some downsides due to an influx of green energies into the market, but analysts predict future gains as conflict and other factors drive gas prices higher.

2.Sable Offshore Corp

Sable Offshore Corp. (SOC -14.79%)

Sable Offshore Corp 是另一家可以利用海浪预测系统保护其海上钻井平台和钻井站的钻井公司。该公司成立于 2020 年，原名 Flame Acquisition Corp，后更名为 Sable。Sable 公司的近海业务位于加利福尼亚海岸附近的联邦水域。该公司还拥有 7.6 万英亩的海底租约，可以输送原油和天然气。

The firm has many strategic partnerships with industry leaders including ExxonMobil, Canada Ltd, Imperial Oil Resources Limited, and Pengrowth Energy Corporation, to name a few.  The company’s stock has suffered recently due to fluctuations in the market. However, California has a growing demand for these services, positioning Sable Offshore Corp as a premier local energy provider.

海浪预测的未来

This study sheds light on the elusive world of rogue waves. These events are no longer sailor tales but are now predictable events. In the future, these systems will be integrated across the maritime economy to mitigate risk and improve efficiency.  Notably, these systems will get much better and more accurate as they obtain additional reinforced data.

我们可以预见，这种预测方式很快将与区块链网络等系统结合起来，对全球范围内的海量数据进行实时监控。这种改进将使这些系统能够以不可更改的方式记录和跟踪海洋中的实时事件。然后，这些数据可用于提高科学家对这些难以捉摸的现象的理解。

探测可以解决 "乱波 "问题

There’s no way to stop rogue waves from forming as of yet. However, the first step to preventing major losses is determining when and what makes these events occur. These engineers have taken the first steps and laid the groundwork for future research that could save lives. For now, their efforts are making “风浪” across multiple industries.

了解其他很酷的人工智能项目 现在.