Electricity Load Forecasting Technology Contest

Starting on June 20, 2017, TEPCO held its First Electricity Load Forecasting Technology Contest with the goal of discovering innovative methods and new approaches.

Dealing with two types of forecasting as well as reporting, the contest presented challenging tasks, yet we were delighted to receive over 100 entries. Many contestants participated from overseas, and we were impressed by the momentum in the data science field and the potential for growth both domestically and abroad.

Nine entries demonstrating highly precise forecasting advanced to the Final Judging stage, and the prize winners were determined at the final meeting of the Judging Committee held on October 19.

We are grateful to everyone who participated in the contest, and we hope to enjoy your continued support in the future.

Application Guidelines: https://cuusoo.com/projects/50136


Judging Committee

  • Head Judge
    • Yutaka Matsuo (Project Associate Professor, Department of Technology Management for Innovation, Graduate School of Engineering, The University of Tokyo)
  • Judges
    • Kazuhiko Ogimoto (Project Professor, Collaborative Research Center for Energy Engineering Institute of Industrial Science, The University of Tokyo)
    • Hiroshi Okamoto (Executive Vice President, TEPCO Power Grid, Inc.)
    • Tomomichi Seki (Managing Executive Officer, Tokyo Electric Power Company Holdings, Inc.)
    • Tomohisa Azuma (Manager, Demand-Supply Planning Group, Power System Operation Department, TEPCO Power Grid, Inc.)
    • Tomoki Okamoto (Manager, Energy Economics Group, TEPCO Research Institute, Tokyo Electric Power Company Holdings, Inc.)

Prizes

  • Best award¥1,500,000

    TOSHIBA Corporation

    Comments from Judging Committee

    Load forecasting methods based on weather forecasts are easily affected by times of seasonal change, such as the period during which the contest was held. This team successfully dealt with that shortcoming by using weather forecasts from multiple locations. The entry also employed ensemble learning to achieve even greater precision, impressing the judges with its ingenuity and potential. We are looking forward to future developments.

  • 2nd place¥500,000

    TESLA Asia Pacific, Ltd.

    Comments from Judging Committee

    Although this team had limited access to the weather forecast data in Japan and were not able to use a local weather vendor during the Actual Forecast, they derived very high accuracy in both forecast of the Annual Forecast and the Actual Forecast. This suggests that if the weather data from a local weather vendor was used, this team may improve their performance in the prediction.

  • 3rd place¥300,000

    Japan Meteorological Corporation

    Comments from Judging Committee

    We took note of concepts such as temperatures weighted for each region’s population. Although it is simple, this highly precise method holds promise for the future.

  • Presentation award¥500,000

    Wataru Uda (Youworks Co.,Ltd)

    Comments from Judging Committee

    Analyzing problems according to each factor that affects load is a fascinating concept. It is a simple yet attractive method. Incorporating indicators such as social activity was a creative approach.

  • Special prize¥100,000

    CHIYODA Corporation

    Comments from Judging Committee

    Despite using data that can be obtained free of charge, their method devised a means of achieving stable load forecasts even when weather forecasts are inaccurate. This creative approach uses many explanatory variables and a fully coupled neural network. We are looking forward to future developments.

  • Special prize¥100,000

    Japan Weather Association

    Comments from Judging Committee

    This team produced a highly precise Annual Forecast. Taking advantage of the strengths of weather forecasting and optimizing its application for load forecasting were defining characteristics of their approach.


Summary

In addition to their 3 submissions, the top-ranked contestants also turned in presentation videos and materials as their final tasks.


Final Judging

The Judging Committee met on Thursday, October 19 to discuss the contents of the presentation videos, after which the rankings were determined.


General Comments from Judging Committee

  • Yutaka Matsuo

    (Project Associate Professor, Department of Technology Management for Innovation, Graduate School of Engineering, The University of Tokyo)

    This contest provided a beneficial overview of the different methods that are possible. If these efforts continue to develop, it may be possible to recruit technical experts who will help us become the firm with the world’s most accurate forecasting. In that sense, the contest seems to hold great potential for the future.

    In a contest of a different field, deep learning method gradually became so dominant that it is not even possible to earn high ranks without incorporating this method. It will be interesting if the same trend develops after we hold our contest a few more times.

  • Kazuhiko Ogimoto

    (Project Professor, Collaborative Research Center for Energy Engineering Institute of Industrial Science, The University of Tokyo)

    This time, we achieved excellent results thanks to the contestants who participated.

    It is important to think carefully about what we are trying to forecast in the contest, and for what purpose. As the structure of the market for electric power itself changes, it may be possible to make forecasts based on a solid grasp of those changes. In fact, setting themes with these factors into consideration will make future contests even more interesting.

  • Hiroshi Okamoto

    (Executive Vice President, TEPCO Power Grid, Inc.)

    The contest was a success in that we were able to receive many entries and variety of intriguing methods. Not many electric power companies around the world have attempted anything like this contest, but as a big data company, we hope to hold more contests in the future.

  • Tomomichi Seki

    (Managing Executive Officer, Tokyo Electric Power Company Holdings, Inc.)

    It goes without saying that the contest offered many insights, and as we move forward, we will think of ways to build upon the results and make future contests even more lively. We may find ourselves operating as a firm that provides data, together with issues that need to be solved, to data scientists who are looking for issues to tackle with sufficient data.

  • Tomohisa Azuma

    (Manager, Demand-Supply Planning Group, Power System Operation Department, TEPCO Power Grid, Inc.)

    The contest provided many practical business insights, and we were reminded again of the importance of using meteorological data. On the other hand, we also took note of novel approaches such as the use of neural networks. We plan to use the results from the contest to raise the precision of our forecasting.

  • Tomoki Okamoto

    (Manager, Energy Economics Group, TEPCO Research Institute, Tokyo Electric Power Company Holdings, Inc.)

    We received entries from contestants working at comparatively high levels. The forecasting methods we use at TEPCO differ from those used by the contestants, but we maintain our traditionally high level of precision. On the other hand, the contest reminded us of the importance of working to improve our precision without becoming complacent or settling for our current methods. We thank everyone who participated.