Developing Method to Forecast the Short-term Instantaneous Wind Speed for Railways Operation Control under Strong Winds

Japanese railway companies monitor strong winds along their lines with anemometers installed at intervals of several kilometers to several tens of kilometers on average, in order to prevent railway disasters on railways caused by strong winds. When strong winds are observed by anemometers monitoring each section, train operations are controlled. Specifically, if the instantaneous wind speed observed by an anemometer exceeds a certain threshold (usually 25 m/s, on conventional lines in Japan), the speed of train traveling on that section is limited (slowed down) or suspended. However, if strong winds continue intermittently, train operation is repeatedly suspended and resumed, resulting in reduced transportation service. Therefore, we are studying to develop a method for forecasting the short-term instantaneous wind speeds that can be directly used for operation control under strong winds. The Japan Meteorological Agency (JMA) and private weather companies provide wind speed forecast information, but their forecasts are averaged wind speeds and generally have a time resolution of more than one hour. On the other hand, operation control under strong winds requires forecasts of instantaneous wind speeds with a time resolution of about ten minutes.

We have developed a method for forecasting 10-minute maximum instantaneous wind speeds using instantaneous wind speed data observed at anemometers and time series analysis. Our method can forecast 10-minute maximum instantaneous wind speeds up to two hours ahead by using different parameters for the time series analysis depending on the synoptic scales that cause strong winds: typhoon, low pressure, front, winter monsoon, and others. We have also developed a method to automatically classify these five types of synoptic scales from surface weather maps using deep learning.

We have confirmed that the synoptic scale can be automatically classified from surface weather maps with an accuracy of about 70%. As a result of forecasting 10-minute maximum instantaneous wind speeds for 50 cases of strong winds, it was confirmed that 10-minute maximum instantaneous wind speeds can be forecast up to one hour ahead with an RMSE of about 5 m/s or less, regardless of the synoptic scale. Furthermore, we have obtained the prospect of reducing the number of operation control under strong winds and the total suspension time by using our forecasts, compared to the current operation control under strong winds that uses only observed wind speeds.