Abstract: To obtain a high-precision model for the axial capacity prediction of piles, in this paper, a combined prediction model based on the idea of combining prediction with the long-term and short-term memory neural network model (LSTM) and deep belief network model (DBN) is established. The combined model was further improved based on the Satin blue bowerbird optimization algorithm (SBO), Northern Eagle optimization algorithm (DGO) and Cuckoo optimization algorithm (CSA) to further improve the accuracy of the model. The results showed that: The Q-S curve of piles simulated by different models showed the same trend, and the fitting effect of SBO-DBN-LSTM model between the simulated and the measured values was the best. When the input parameters of the model were 5, the accuracy of the model was generally high; when the input parameters of the model are 4, although the accuracy is reduced, it can still meet the estimation requirements. The SBO-DBN-LSTM models can guarantee high accuracy under different input parameters and can be recommended for predicting the axial capacity of piles.

Key words: Axial Capacity of Piles, Combined Prediction, Long-Term and Short-Term Memory Neural Network, Deep Belief network, Satin Blue Bowerbird Optimization Algorithm