NRIET fog forecasting method based on machine learning

Abstract

The invention discloses an NRIET fog forecasting method based on machine learning. The method comprises steps that historical and forecast data for model training and operational forecasting in and around the forecast area is collected; the collected data is analyzed and processed to be a sample set for training a machine learning model; for problems, an xgboost model algorithm based on a decisiontree is selected, key influence factors generated by the airport fog are screened, and a relationship model between the fog and the influence factors is established; a training set and a verificationset are trained; according to the parameter adjusting sequence, model parameters are repeatedly adjusted till a forecasting model with optimal performance is acquired; an integrated learning method is utilized, and intersect verification and iteration training are carried out for the model to further improve model forecasting performance; observed real-time data is inputted to the forecasting model to acquire the fog forecasting result.

Description

technical field [0001] The invention relates to a heavy fog forecasting method based on NRIET machine learning, and belongs to the field of heavy fog objective forecasting systems. Background technique [0002] Heavy fog is the main weather phenomenon that causes low visibility. Improving the level of fog forecasting technology is an important measure to ensure traffic safety. However, in the forecast of various weather phenomena, the forecast of fog is still very difficult, and it is still a worldwide problem. Especially for non-meteorological industry users such as civil aviation and transportation, the accuracy and operational level of fog forecasts need to be improved urgently. [0003] At present, the fog forecasting methods mainly include synoptic interpretation, numerical forecasting and statistical forecasting methods. [0004] Synoptic interpretation is based on the forecast results of the situation field provided by the previous weather situation or numerical mod...

AI Technical Summary

Problems solved by technology

[0008] Aiming at the problem that the synoptic interpretation mainly used in the current operational forecast relies too much on the subjective experience of the forecaster, and the accuracy and reliability of the forecast are low, this method provides an advanced objective forecast technology based on machine learning, which does not rely on Subjective forecasting experience, which can provide accurate positioning of heavy fog forecasts in real time, and the accuracy rate of 30-minute heavy fog level forecasts can reach more than 98%.

[0009] Aiming at the problem that the numerical prediction method cannot fully understand and describe the complexity of the physical process of heavy fog and consumes a large amount of computing resources, this method is based on an advanced machine learning method from the perspective of data statistics and probability, and emphasizes the actual application effect and downplays the Theoretical analysis, let the computer automatically "learn", automatically analyze the law from the data, and use the law to predict the unknown data

Images