Rainfall prediction based on PCA-BiLSTM algorithm

This paper examines key issues in rainfall prediction under the dual impacts of climate change and human activities. Accurate rainfall prediction is essential for disaster mitigation, infrastructure protection and sustainable development. This study evaluates the performance of PCA-BiLSTM, LSTM and BP algorithms on the Guangxi Nanning dataset. PCA-BiLSTM combines the dimensionality reduction capability of PCA and the time series feature extraction capability of Bi-LSTM, and shows excellent performance in both low and high rainfall scenarios. The experimental results show that PCA-BiLSTM achieves the lowest MSE (20.24), RMSE (4.50), and MAPE (1.12%), which is better than LSTM (MSE: 48.91, RMSE: 6.99, MAPE: 1.46%) and BP (MSE: 49.73, RMSE: 7.05, MAPE: 2.12%) . In terms of classification accuracy, PCA-BiLSTM reaches 92.54%, LSTM reaches 82.09% and BP reaches 73.13%.PCA-BiLSTM is good at capturing complex temporal patterns and categorizing rainfall, and is therefore well suited for precision forecasting tasks, but is computationally expensive. LSTM strikes a balance between accuracy and efficiency, whereas BP is applicable to the resource-limited situations. Overall, PCA-BiLSTM proved to be the most robust and accurate rainfall prediction method.