Toward building an expert system for weather forecasting operations

摘要

Extracting knowledge from a human expert has long been recognized as the most time-consuming and complicated task in the development of an expert system and is termed as the bottleneck of expert system development. The success in acquiring knowledge automatically by using machine learning techniques has to a certain extent overcome this problem, and hence, has found applications in many fields. Machine-learning techniques have been applied successfully to several real-world problem domains. The application of machine learning techniques to weather forecasting has received only little attention. One reason for this has been the difficulty in obtaining suitable weather forecasting data sets. This paper describes the application of machine-learning techniques on weather data sets to acquire knowledge automatically for the development of an expert system to predict the occurrence and mean depth of rainfall over Melbourne City and its suburbs in Australia during a 24-hour period. The weather data sets were assembled from the archives of the Australian Commonwealth Bureau of Meteorology.