首页 > 数据分析 > 航空业如何通过机器学习来平滑IT动荡

[悬赏]航空业如何通过机器学习来平滑IT动荡 (已翻译0%)

查看 (475次)
英文原文:How the airline industry can smooth IT turbulence through machine learning
标签: 数据分析
admin 发布于 2017-04-17 13:42:08 (共 5 段, 本文赏金: 19元)
参与翻译(0人): 暂无 默认 | 原文

【待悬赏】 赏金: 3元

the comedian Louis C.K. has a bit about air travel, in which he mocks travelers for their impatience around delays. Even a relatively short wait of 40 minutes, while sitting in a plane on the tarmac, sends many people into a tizzy, he says. Yet, no one seems to appreciate the part where they fly through the clouds, like a bird: "New York to California in six hours? It used to take 30 years to do that ... and a bunch of you would have died on the way."

It's funny because it's (mostly) true.

Still, given the technological advancements we all enjoy in the realms of communication and medicine, it does sometimes seem incredible that air travel plans are regularly beseeched by delays — oftentimes far longer than 40 minutes — and cancellations.

Sure, weather plays a major role. But as the news website Quartz has been documenting since late 2015, glitches within airline IT systems regularly cause problems, ranging from short delays to worldwide outages that knock thousands of flights off schedule.



【待悬赏】 赏金: 2元

Could machine learning (ML) and other artificial intelligence (AI) tools change that? It's still early days, but some airline industry analysts and IT providers are hopeful that they can. Mark Jaggers, a research director with Gartner, sees three areas within aviation IT that are ripe for improvement through cutting-edge data tools:

  • Testing techniques, including simulations
  • Automatic and proactive alerts around critical infrastructure and processes
  • Automated workload and resiliency programs

"Today, a lot of these are manual processes, so AI and ML would automate these steps," he said.

Sabre, which provides technology for the travel industry, is beginning to test some machine learning tools — anomaly detection and clustering — within its R&D department, according to Mark McSpadden, director of Sabre Labs.



【待悬赏】 赏金: 2元

The goal is to employ these tools to trigger alerts to potential problems before they happen, and therefore avoid IT glitches in the first place. "Anomaly detection is about finding outliers — what is happening outside the curve — within a set of data," McSpadden said. "That data may be specific to an airline, or to what is happening on the network, [or it] might be specific to what is going on in application."

Through clustering, data is sorted into groups via various methods and then analyzed. "We see promise in these kinds of techniques," he says.



【待悬赏】 赏金: 6元

The big picture

Airlines have suffered some monumental outages in recent years. In January, a glitch took down Delta's website and mobile app along with airport information screens and computers at its reservation desks. It canceled nearly 300 flights.

But that paled in comparison to a power outage the airline suffered at its operations center in Atlanta the previous August. That set off a system failure that resulted in around 2,000 flight cancellations over several days.

Delta suffered reputational damage, certainly, but Jaggers says that on balance, airlines are not seeing an uptick in major IT issues. In fact, Gartner is tracking an opposite trend. "We are seeing a reduction in the number of outages, but the corollary is that IT systems are getting more complex. So, when you have an issue, it's more difficult to triage and address," he says.

All that complexity also means there is "less slack in the system," Jaggers explains. Airline IT networks must juggle so many points of integration, from schedules to maintenance to TSA, that even a small interruption in service can lead to a cascade of delays or other problems.

Part of the complexity is derived from the fact that airlines, like many businesses, run a mixture of legacy and more modern computing systems. Add to that the high rate of consolidation that the industry has undergone in recent years. Plus, Jaggers noted, the transportation sector, as a whole, spends less on IT than some other industries.

For example, IT spending for the banking and financial services industry as a percent of revenue was in 6.8% in 2016, whereas the transportation industry's was 3.2%.

On top of all this is another key dimension: consumers are interacting more — and more directly — with airlines than ever before. Travelers book online, rather than through agents. They receive text messages with flight updates. They might use the airline app to access airport or Wi-Fi networks. So, when IT problems arise and disrupt those touch points, consumers oftentimes respond loudly and through social media.



【待悬赏】 赏金: 6元

Rise of the machines

One outcome of the industry's shift toward more direct electronic interaction with consumers is a reduction in the number of agents airlines hire for customer-facing positions, Jaggers says. As a result, when things go awry, travelers might feel more on their own. Yet, they also benefit from many conveniences technology  provides, such as online check-in, or text message alerts.

As was made quite obvious when a typo took down the Amazon S3 last month, humans sometimes screw things up. And as Wired recently reported, airlines are developing machine learning tools that will analyze data from the thousands of sensors mounted inside aircraft and crunch the data to power predictive maintenance alerts or even subvert human error that could lead to safety problems.

Likewise, airlines including EasyJet, Emirates, KLM, and Korean Air, are looking into AI tools to keep reservation systems and other parts of their IT networks humming along, and a 2016 survey by air travel IT provider SITA found that 24% of airports are planning to test AI tools over the next five years.

In addition to anomaly detection and clustering, Sabre is also researching generative testing, in which an ML engine sends streams of unexpected data into a network as a means of testing its robustness.

"Our systems are bringing in data from all types of different places and sometimes [that data] looks [different] than you expected it to look," said McSpadden. But Sabre is still working on these tools internally. "Some of these are core machine learning techniques, and the math behind them has been there for a while, but compute power is now catching up and [we're starting to be] able to run these in real time."

As these machine learning tools prove themselves in the lab, Sabre will reach out to customers to begin pilot tests. "You can do a lot in the lab," he says. "But you can't run a pilot without a customer."


GMT+8, 2018-11-16 17:56 , Processed in 0.039400 second(s), 11 queries .