Accession Number:

ADA383780

Title:

A Neural Network Solution to Predicting Wind Speed at Cape Canaveral's Atlas Launch Pad

Descriptive Note:

Master's thesis

Corporate Author:

AIR FORCE INST OF TECH WRIGHT-PATTERSONAFB OH

Personal Author(s):

Report Date:

2000-03-01

Pagination or Media Count:

133.0

Abstract:

This thesis demonstrates the potential for using time-delay neural networks to provide Launch Weather Officers LWOs at 45th Weather Squadron 45 WS with advance warning of wintertime November-March peak wind speeds at the Atlas launch pad. The 45 WS provides weather support to the United States space program at Cape Canaveral Air Station, NASAs Kennedy Space Center, and Patrick Air Force Base. Due to the complex wintertime environment produced by the effects of friction and instability, 45 WS LWOs consider wintertime launch pad winds their toughest forecast challenge. Neural networks were developed, trained, and tested using observations of wintertime peak wind speed, wind direction, and directional deviation collected from March 1995 through March 1999 by 45th Space Wings Weather Inforrnation Network Display System. Using current and past values of the observed elements, the networks produced 16 forecasts of peak wind speed. The first forecast was valid for 30 minutes past forecast start time, the second for 1 hour past start time, etc., up to 8 hours past start time, for any start time. Network performance was compared to three other forecasting options persistence, climatology, and randomly selecting wind speeds from a climatologically based distribution. For forecasts at the end of the forecast period, networks that were tested with data near in time to the networks training data showed skill over the other forecasting options. A new confidence measure for neural network forecasts based on mean absolute error was also developed. Confidence for shorter forecast times was not necessarily higher than for longer forecast times. The results of this thesis provide a baseline for measuring future attempts at forecast improvements and establish the neural network approach as a potential means of enhancing peak wind speed prediction accuracy, especially for forecasts late in the forecast period.

Subject Categories:

  • Cybernetics
  • Unmanned Spacecraft

Distribution Statement:

APPROVED FOR PUBLIC RELEASE