Dry, Dahlia (School: Benton High School)
This project takes a different approach to meteorological forecasting by investigating underlying patterns in local weather fluctuations and attempting to find a connection between these phenomena that can lead to larger weather effects. Using a dual-part approach, the goal was first to use time series modeling techniques such as the Box-Jenkins method to create a model that can forecast wind shifts to a reasonable degree of accuracy. A Raspberry Pi and sensor apparatus were used to collect data at 60 second intervals for two twelve day trials. A data analysis program created four models which either contained AR or ARIMA parameters and were set in either a static or dynamic environment. In the second part of the experiment, another analysis program used support vector machine algorithms to binarily classify the data points, searching for a relationship between wind direction and temperature. The first analysis program showed that the researcher’s implementation of a dynamic rather than static ARIMA model allowed a dynamic ARIMA model of order (4,4,3) to achieve an average binary accuracy of 93% over two trials, which was significantly higher than that of static DAR models used. The second analysis program classified the compound temperature-wind shift data set with 71% accuracy. Thus, the researcher concluded that creating an environment that enables a time series model to evolve periodically along with a data set can substantially improve the forecasting ability of a model. The results also support the idea that relationships between local weather fluctuations could be used to increase the accuracy of current forecasting methods.
Mu Alpha Theta, National High School and Two-Year College Mathematics Honor Society: Third Award of $1,000.
Fourth Award of $500