In this study, time-series data of corrosion current for Q235 carbon steel, continuously collected under the marine atmospheric environment in Qingdao, along with various environmental and meteorological factors, were utilized to construct and analyze two time-series prediction models: Long Short-Term Memory (LSTM) and Transformer. During the modeling process, feature variables were first standardized, and a logarithmic transformation was applied to the corrosion current data. Key features were selected through Spearman correlation analysis, which revealed a high correlation coefficient of 0.83 between temperature and dew point temperature; therefore, the highly collinear dew point temperature was excluded. Subsequently, the preprocessed data were subjected to stationarity analysis using the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, indicating that the data were relatively stationary and suitable for subsequent modeling. After multiple iterations, the optimal parameters for the LSTM model were determined: epochs = 10 and batch size = 30, achieving an R2 value of 0.96 and RMSE of 0.56; for the Transformer model, the optimal parameters were epochs = 10 and batch size = 50, with an R2 value of 0.94 and RMSE of 0.63. The results demonstrate that the predictive performance of the LSTM model surpasses that of the Transformer model. Consequently, the LSTM model was selected as the time-series prediction model for the corrosion current of Q235 carbon steel in the marine atmospheric environment. The findings of this study provide a reference for corrosion prediction in complex future environments and offer a scientific basis for the formulation of protective measures.