基于CNN-BiLSTM的广西月降水量预测模型研究

Research on Monthly Precipitation Prediction Model of Guangxi Based on CNN-BiLSTM

摘要: 深入研究短期气候预测新技术，提高月降水的预测准确率对国民经济健康发展具有重要意义。针对卷积神经网络(Convolutional Neural Networks, CNN)在降水预报中难以有效提取数据时序特征，而长短期记忆网络(Long-Short-Term Memory, LSTM)难以提取数据空间特征的问题，本文建立卷积神经网络和双向长短期记忆网络组合月降水模型(CNN-BiLSTM)，该模型充分结合CNN提取月降水数据在空间结构上的特征，以及BiLSTM提取月降水数据时序特征的优势，从而实现对月降水数据空间特征的提取，以及非线性长期和短期时间序列依赖关系的有效动态建模。以广西柳州月降水进行实例验证，实验结果表明，CNN-BiLSTM模型可同时有效提取数据空间特征和时序特征，预测精度高，泛化性能高，稳定性好，为降水预报提供了一种可靠、有效的方法。

Abstract: It is of great significance for the healthy development of the national economy to conduct in-depth research on the new technology of short-term climate prediction and to enhance the accuracy of monthly precipitation prediction. This paper addresses the challenge of effectively extracting temporal and spatial features of precipitation data using Convolutional Neural Networks (CNNs) and Long-Short-Term Memory (LSTM) networks, respectively. To this end, a CNN and a Bidirectional Long-Short-Term Memory Network (Bi-LSTM) are established. This paper establishes a Convolutional Neural Network (CNN) and a Bidirectional Long-Short-Term Memory Network (Bi-LSTM), which fully combine the advantages of CNN in extracting spatial features of monthly precipitation data and Bi-LSTM in extracting temporal features of monthly precipitation data. This enables the extraction of spatial features of monthly precipitation data and effective dynamic modeling of the dependence of nonlinear long-term and short-term time series. To illustrate the efficacy of the proposed methodology, we conducted a case study on monthly precipitation data from Liuzhou, Guangxi. The experimental results demonstrated that the CNN-Bi-LSTM model is capable of extracting both spatial and temporal features with high precision, accuracy, and stability. This suggests that the proposed approach can serve as a reliable and effective tool for precipitation forecasting.