An enhancement method for chloride diffusion coefficient long-term prediction based on Hilbert dynamic probabilistic interpolation and BO-LSTM

For marine-exposed concrete, long-term prediction of chloride diffusion is important for durability assessment. The chloride diffusion data exits nonlinear and stochastic characteristic that makes existing model remains challenging at long-term prediction. In this paper, a mix intelligent predictive model integrating the BayesianOptimized Long Short-Term Memory network (BO-LSTM) and the Hilbert Dynamic Probabilistic Interpolation (HDPI) method is proposed to address this problem. experimental data. To enhance the predictability of discrete time series, the HDPI algorithm is employed to dissect the original time series into distinct components, encompassing a single-frequency undulation signal, a stationary signal, and a residual signal. Subsequently, the undulation signal is transmuted from a discrete to a continuous form via the utilization of the Hilbert transform. This continuous signal serves as the training input for the neural network, wherein the hyperparameters are meticulously refined through the BO algorithm. The subsequent forecast of the time series data is derived by aggregating the predictions of all the dissected signals, facilitated by the trained neural network. To evaluate the performance of the proposed HDPI-BO-LSTM method, this paper conducts a case study and implements a comprehensive comparison between the proposed method and BO-LSTM neural network. The results show that the HDPI-BO-LSTM method enhances the prediction accuracy by 8% compared to BO-LSTM method.