Settlement prediction based on CNN-BiLSTM model integrated with BKA optimization algorithm: a case study of Xining City

Accurate ground subsidence prediction is crucial for disaster prevention and infrastructure safety. Although CNN and BiLSTM models are widely applied in this field, their effectiveness often depends on well-tuned hyperparameters, which are difficult to determine manually. This study proposes a BKA-CNN-BiLSTM model that integrates the Black Kite Algorithm (BKA) with CNN-BiLSTM to perform automated hyperparameter optimization. Using Xining City as a case study, the model is validated through comparative experiments. Results show that the proposed model outperforms traditional BiLSTM and CNN-BiLSTM models, with R² improvements ranging from 5.01%–65.54% and 1.82%–23.66%, respectively. It also achieves lower MAE, MSE, and RMSE values, indicating enhanced prediction accuracy. Overall, the BKACNN-BiLSTM model offers a robust and effective solution for ground subsidence forecasting, with strong potential for broader geohazard monitoring applications.