Deep Learning Prediction Model of Drought and Flood in Summer Based on Random Forest and Attention Mechanism

Abstract

The traditional linear statistical forecasting approach is frequently utilized to tackle the significant interannual variability and nonlinear traits of summer drought and flood patterns in Guangxi. However, its forecasting precision tends to be inadequate. In this research, the year-to-year change in the average precipitation for the month of August was employed as a means to predict trends in droughts and floods. The correlation factor between the forecast and the previous 500 hPa monthly average height field was calculated, yielding 81 preliminary forecast factors for the early monthly average circulation field. First, the random forest algorithm was utilized to assess and prioritize the significance of the 81 predictive elements and indicators. Subsequently, the top six crucial variable components were chosen to feed into the deep learning LSTM network's forecasting framework. Second, an attention mechanism was employed to provide different attention values to the input variables of the model. Third, a forecasting model for the year-to-year change in average summer rainfall in Guangxi was constructed, incorporating a random forest and an attention mechanism-enhanced LSTM network (RF-LSTM-Attention). Upon applying this predictive model to forecast the average summer rainfall for the eight-year period in Guangxi (June–August) from 2013 to 2020, the model exhibited an average absolute percentage error of 9.49%. The relative error of prediction in the six years, especially in the years with maximum and minimum precipitation in the eight-year return sample, did not exceed 15%. The RF-LSTM-Attention model was approximately doubled and showed better forecast accuracy in qualitative and quantitative forecasting.