Predicting gusts with radar and AI Strengthening safety measures in response to derailment accidents JR East

It has been 15 years since the derailment accident of the JR Uetsu Line in 25 days.

The accident is attributed to a local gust, which is not easy to predict.

JR East is strengthening safety measures by developing the world's first system that uses AI = artificial intelligence to predict windstorms and stop train operations, as well as a dedicated radar that captures the movement of the wind based on the lessons learned from the accident. I will.

In December 2005, the limited express train derailed on the JR Uetsu Line in Shonai Town, Yamagata Prefecture, killing 5 people and seriously injuring 33 people, and it was concluded that the cause was a local gust.

Based on the lessons learned from the accident, JR East has developed a new system.

The mechanism is to catch a "wind vortex" that may cause a local gust such as a tornado with Doppler radar, and stop the train immediately if there is a possibility of reaching the track.

Introduced three years ago, in order to improve accuracy, we used AI = artificial intelligence to learn about 36,000 images such as false "wind vortex" radar images.

As a result, we started operation last month, saying that we can expect to reduce the number of cases that are mistakenly recognized by about 20% and the number of cases that we missed.

According to JR East, it is the first system in the world to use AI to predict gusts and stop train operations.

JR East is strengthening safety measures based on the lessons learned from accidents, such as significantly increasing the number of anemometers throughout the jurisdiction.

How did the derailment accident and countermeasures take place?

The derailment of the limited express train on the JR Uetsu Line, which connects Akita and Niigata prefectures, occurred 15 years ago on December 25, 2005, at around 7:14 pm.

The limited express "Inaho" bound for Niigata, carrying 46 passengers and crew, derailed while traveling in Shonai Town, Yamagata Prefecture at a speed of about 100 km / h, killing 5 men and women and injuring 33 seriously.

According to the report of the Accident Investigation Commission, the cause was a local gust called a tornado or downburst, and it is believed that it was capsized by a gust with an instantaneous wind speed of about 40 meters.

It was concluded that the anemometers installed along the railway lines near the site at the time of the accident did not reach the regulation values ​​for suspending train operation or slowing down, and it was not easy to predict.

After the accident, JR East, with the cooperation of the Meteorological Research Institute of the Japan Meteorological Agency, will generate gusts from coastal areas to inland areas using approximately 30 meteorological observation devices such as anemometers, barometers, and small weather radars. I worked on elucidating the mechanism.

After more than 10 years of research, it was found that most of the gusts that occur in this area occur over the Sea of ​​Japan and land due to monsoons.

In response to this, in 2017, three years ago, we installed a Doppler radar that captures the movement of the wind over the Sea of ​​Japan in detail, and stopped the train by detecting the vortex of the wind that may cause a gust such as a tornado. I made a system.

However, the problem was that there were errors such as misrecognizing the vortex of the wind and mistakenly recognizing the spray of waves as a vortex, and we were working to improve the accuracy.

Prediction of gusts What is the mechanism?

Predictions of gusts such as tornadoes are made using Doppler radar, which captures the movement of the wind in detail.

That's how it works.

Winds approaching the radar are shown in red, winds moving away from the radar are shown in blue, and when the winds in the opposite directions of red and blue are combined, it means that the wind is blowing in a spiral.

It is difficult to capture gusts such as tornadoes with radar, but since they often occur under this "wind vortex", the key to prediction is how quickly this "wind vortex" can be detected. I will.

However, it is difficult to accurately capture complex wind movements, and errors may occur due to splashing waves, so improving accuracy has been an issue.

The research group of JR East and the Meteorological Research Institute uses AI to learn about 36,000 images, such as images that were mistakenly recognized as "wind vortices" for the past two years, and accuracy. Has been improved.

This is the screen of the system created by JR East based on this.

The wind vortex detected on the map and the range reached after 10 minutes are predicted and displayed.

If the track enters the predicted range, it sounds an alarm that it may pass on the track and instructs the driver to stop the train.

This is the world's first system that uses AI to predict gusts and stop trains, and JR intends to further improve accuracy by learning while actually operating it.

The person in charge of JR East said, "It is a very rudimentary thing in AI that simply makes AI learn vortices and things that are not vortices and outputs the learning results, and experts make vortices. I think that if AI can learn the information used for judgment, the accuracy will be further improved. I would like to continue researching and improve the ability of AI itself and improve the safety of railways. " I was talking.