The magnitude of precipitation is closely related to people's lives. To enhance the precision of precipitation prediction, a monthly precipitation prediction model for Guangxi based on EVO-CNN-LSTM-Attention is proposed in this paper. Firstly, the model utilizes convolutional neural network to capture spatial features of the input sequence, followed by capturing temporal correlations through long short term memory neural networks. Secondly, Attention is used to better understand internal correlations of the sequence. Finally, an energy valley optimization algorithm is introduced to optimize parameters such as learning rate, convolutional kernel size, and number of neurons that are difficult to determine in the model, making the structure of the model more reasonable, optimize model structure and training parameters. Using the historical precipitation data of Liuzhou city, Guangxi province, EVO-CNN-LSTM-Attention is compared with CNN, LSTM, CNN-LSTM and CNN-LSTM-Attention. The findings reveal that EVO-CNN-LSTM-Attention model surpasses other comparative models, exhibiting both high precision and stability.