Monthly precipitation prediction based on quadratic decomposition and improved parrot algorithm

Abstract The amount of precipitation directly affects the ecological balance and the economic benefits of the region. However, the highly nonlinear and stochastic nature of precipitation time series data limits the accuracy of predictions. Therefore, improving the prediction accuracy of regional precipitation is crucial for formulating disaster prevention and mitigation measures, as well as for responding to climate change. To achieve a scientific and effective prediction of regional precipitation, this study proposed a precipitation prediction model based on the CEEMDAN-TVMD-IPO-BiLSTM framework. The model first decomposed the original precipitation data using the CEEMDAN decomposition algorithm, output the modal components and residual components, and then used the topology optimization algorithm (TTAO) to optimize the VMD, and decomposed the high-dimensional sequence in the first decomposition result for the second time. An improved parrot optimizer (IPO) algorithm based on chaotic Cat and Cauchy-Gaussian variation was introduced to optimize the bidirectional long short-term memory neural network (BiLSTM). This precisely constructed prediction model was utilized to predict regional precipitation, with historical monthly precipitation data from three representative cities in China—Guangzhou in the east region, Changsha in the central region, and Emeishan in the west region—used to validate the model’s accuracy and robustness. Experimental results indicated that the proposed CEEMDAN-TVMD-IPO-BiLSTM model achieved RMSE values of 32.373, 14.445, and 22.447 for the three cities, respectively, with corresponding R² values of 0.960, 0.972, and 0.977, outperforming other models. This demonstrated its advantages in monthly precipitation prediction, allowing for a better characterization of precipitation fluctuation patterns and providing scientific references for formulating policies to combat droughts and floods.