TESLA & MeteoGroup Partnership Tackles Accurate Load Forecasting for the Japanese Power Market

Significant changes have already happened in the Japanese power market. It became fully liberalized in April 2016, opening the market and widening choices for consumers. Before this, consumers could only purchase power from their regional power company.


Pre- April 2016, there were only ten power retailers in Japan- one for each region. Now there are approximately 450 power retailers. Moreover, as the fifth-largest electricity producer in the world, the Japanese market still has unexplored potential, demonstrated by the electricity futures contracts launch on 17 September 2019 by TOCOM.


Increased competition = increased appetite for highly accurate load forecasting


As we've seen in other deregulated power markets around the world, increased competition typically translates to an increased appetite for highly accurate load forecasting.


This scenario proved to be true in Japan, with the Tokyo Electric Power Company (TEPCO), Japan's largest vertically integrated power utility, holding its first Electric Load Forecasting Technology Contest in 2017.


TESLA & MeteoGroup partnership entering the Japanese market


Out of 100+ participants, including 40 from overseas, the TESLA/MeteoGroup entry was the only overseas participant to place. This contest was TESLA's first time forecasting in Japan, with TESLA utilizing just a single Tokyo weather station. The contest was very close; Toshiba only surpassed TESLA on the last day of the competition.


Since Toshiba used multiple weather stations, the outcome suggested that the TESLA forecast may have performed better with more than one.


Optimizing to find the right combination


Using the learnings from the contest, based on a detailed analysis of historical weather station observation data, TESLA and MeteoGroup collaborated to determine the optimal number of weather stations for the Tokyo region.


To train the load forecasting model, MeteoGroup provided historical weather station data, which was further supplemented by synthetically generated data. Given the sparsity of weather stations equipped with cloud /solar radiation sensors, this synthetically generated data was especially necessary for achieving a good representation of embedded solar generation, and its corresponding impacts on the load forecast.


TESLA tested many different combinations of weather stations and settled on a total of eight for optimal forecasting accuracy. These were chosen using TESLA’s proprietary weather station selection optimization algorithms. Since the deployment of the eight-weather station model, we've observed a  20% improvement in forecasting accuracy for the Tokyo region.


Refining and extending the offering


We have now developed forecasts for nine regions in Japan. The forecast horizon is 14 days with an hourly model solve as new load, weather observations, and forecast data become available. MeteoGroup weather forecasts power the TESLA forecasts, and the tool is available in both English and Japanese. TESLA and MeteoGroup are currently exploring the potential of adding additional weather stations to enhance the forecasting suite further. There is also the potential to link MeteoGroup’s industry-leading sub-seasonal and seasonal forecast capabilities with the TESLA load forecast.





TESLA & MeteoGroup: A proven partnership


For more than ten years, MeteoGroup analysts have provided TESLA analysts with accurate weather forecasts, historical data, and weather models to create value-adding solutions. The combination of TESLA and MeteoGroup creates robust and integrated products that offer clients a sophisticated modeling approach with a hyperlocal, accurate, and actionable weather forecast. The net result is that clients have access to precise forecasts, which in turn, improves business competitiveness, risk management, and customer loyalty. They complement each other and form a partnership that's greater than the sum of its parts.


The system-level solution is readily available for a two-week trial. For retailers needing a load forecast for their specific customer portfolio, this service is also available. Please get in touch.


Contact Us