A TDNN-based precipitation forecasting method for grassland areas in northern China

Abstract

The invention discloses a method for predicting rainfall in the grassland area of northern China based on TDNN. The method comprises steps: the input variables P (t-1), P (t-2), P (t-3) and output variables P (t) of the TDNN prediction model being determined; obtaining the training data and test data of TDNN prediction model; importing the training data into the TDNN prediction model, and trainingthe TDNN prediction model; the input variables of the test data being imported into the trained TDNN prediction model, the output results of the TDNN prediction model and the actual observed rainfallbeing compared, and the MAE and RMSE being calculated; comparing the training precision of MAE and RMSE of each TDNN prediction model; if MAE and RMSE are less than the expected precision, the training of TDNN prediction model being completed; otherwise, the model parameters being adjusted and retrained. The invention solves the technical problem that the prediction accuracy is not high due to the inability to accurately describe the time relationship between the inputs of the model in the conventional rainfall prediction model.

Description

technical field [0001] The invention relates to the field of rainfall prediction modeling, more specifically, relates to a method for predicting rainfall in grassland areas in northern China based on TDNN. Background technique [0002] In the past 20 years, the rainfall in the northern China grasslands has changed abnormally, and severe droughts or floods have occurred in very few years, which seriously threatens the protection and restoration of the grassland ecological environment in this area, as well as the sustainable development of grassland animal husbandry, especially in the growing season ( The fluctuation of rainfall from June to August every year has a great influence on the vegetation growth in this area. [0003] At present, the prediction technology for rainfall is relatively mature. In addition to the traditional linear statistical model, there are many nonlinear prediction modeling methods, such as particle swarm optimization (PSO), backpropagation neural net...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to overcome the shortcomings in the existing rainfall forecasting technology for the grassland areas in northern China, and to use the observation data of rainfall in the past to propose a method for predicting rainfall in the grassland areas in northern China based on TDNN, which solves the problem of rainfall in the past. In the quantity forecasting model, the technical problem of low forecasting accuracy is caused by the inability to accurately describe the time relationship between the inputs of the model, and the precise forecasting of the monthly average rainfall in the growing season of the research area has been realized

Images