Recently, a weather forecast has improved day by day, and accuracy of the forecast has been improving. At the same time, it is necessary to forecast by the super computer which has larger-scale and higher-performance.
This study considers the small-scale weather forecast which can be executed on a personal computer, and this study confirms effectiveness of the weather forecast by classifing weather pattern by the self-organizing map. To forecast, this research uses the idea "if there is a similar day in the observational data to the observation day, the next day tends to resemble the next day of observation day" and "if there is a similar patern of sevral days in the observational data, the next day tends to resemble the next day of observation days".
First, a self-organizingmap learned the observational data of various places as input, and the pattern was classified. Next, another self-organizing map learned next day's observational data as input, and next day's weather was forecasted. On the other hand, another self-organizing map learned several days' observational data as input, and next day's weather was forecasted by the observational data of several days ago.
Error of the forecasted weather and the actual weather was evaluated. Next, improvement of the precision of forecast was tried by input data conversion.
As a result, the rain hitting ratio resulted in about 73%, the error of highest temperature resulted in about 2.82C, and error of lowest temperature resulted in about 2.46C.