DEVELOPMENT OF AN ALGORITHM FOR FINDING ANOMALIES IN WEATHER REPORTS TO IMPROVE THE QUALITY OF FORECASTS
The issues of using data analysis methods to find and correct errors in the reports issued by meteorologists are considered. The features of processing various types of meteorological messages are studied. The advantages and disadvantages of existing methods of classification of text information are considered. The classification methods are compared in order to identify the optimal method that will be used in the developed algorithm for analyzing meteorological messages. The prospects of using each of the methods in the developed algorithm are described. An algorithm for processing the source data is proposed, which consists in using syntactic and logical analysis to preclean the data from various kinds of noise and determine format errors for each type of message. After preliminary preparation the classification method correlates the received set of message characteristics with the previously trained model to determine the error of the current weather report and output the corresponding message to the operator in real time. The software tools used in the algorithm development and implementation processes are described. A complete description of the process of processing a meteorological message is presented from the moment when the message is entered in a text editor until the message is sent to the international weather message exchange service. The developed software is demonstrated, in which the proposed algorithm is implemented, which allows to improve the quality of messages and, as a result, the quality of meteorological forecasts. The results of the implementation of the new algorithm are described by comparing the number of messages containing various types of errors before the implementation of the algorithm and after the implementation.