Radar echo extrapolation method based on sequential convolutional neural network

Abstract

The invention discloses a radar echo extrapolation method based on a sequential convolutional neural network. The method is realized through the following three phases: a data processing phase: decomposing a radar echo image into a plurality of images through discrete cosine transform according to different frequency bands, carrying out standard operation on the decomposed images, dividing the decomposed images into a plurality of samples to obtain a training sample set or a test sampling set; a neural network training phase: establishing and initializing the sequential convolutional neural network, training the neural network through the training sample set, obtaining a predicted value and calculating loss through network forward propagation, and adjusting parameters through back propagation to enable the neural network to converge; and a neural network test phase: processing the radar echo image to be subjected to extrapolation through the method of the data processing phase to obtain a test sample set, inputting the test sample set into the specific neural network to obtain the predicted value, and restoring the predicted value to a radar echo to obtain a predicted radar echo. The method overcomes the defect of weak precipitation particle attenuation and enhancement modeling capability of a conventional method.

Description

technical field [0001] The invention belongs to the technical field of surface meteorological observation in atmospheric detection, and in particular relates to an extrapolation method based on a time series convolutional neural network. Background technique [0002] Convective precipitation nowcasting has always been an important research issue in the field of weather forecasting. The purpose of this task is to predict the precipitation for a short period of time (usually 0 to 6 hours) in a certain area. It is important for many human production Activities have an important role, such as the operation of airports, the management of water conservancy facilities. The existing prediction methods can be roughly divided into two types: methods based on numerical weather prediction (Numerical Weather Prediction, NWP) and methods based on radar echo extrapolation maps. The former has a very stable effect in the case of long-term forecasting, but the forecast accuracy for small ar...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to solve existing problems such as the limited expression ability of the traditional radar echo map extrapolation model, and the poor effect of the convolutional neural network on processing timing issues. , a radar echo extrapolation method based on temporal convolutional neural network is proposed, which includes the following steps:

Images