Forecasting reinforcement corrosion rate in concrete by an ensemble EEMD–mRMR–BP framework: Time dependence and environmental drivers

Accurate understanding of the time-dependent process of reinforcement corrosion rate is crucial for the durability design and life-cycle assessment of concrete structures. The corrosion rate is strongly time dependent and responsive to both natural environmental and concrete environmental conditions, producing nonlinear, non-stationary and multi-scale behaviors that challenge prediction. Accordingly, this work presents an ensemble framework — EEMD–mRMR–BP — that integrates Ensemble Empirical Mode Decomposition (EEMD), Max-Relevance Min-Redundancy (mRMR) feature selection and a backpropagation (BP) neural network to predict time-series of corrosion rate in reinforced concrete. EEMD decomposes corrosion rate series into intrinsic mode functions (IMFs) and a residual with clearer periodicity; for each IMF or residual the mRMR algorithm selects the most relevant temporal (including partial field electrochemical measurements) and environmental features to avoid redundancy and overfitting; future of each IMF or residual is then forecast by a BP model and recomposed to yield the final prediction. The prediction is clearly carried out based on an explanatory understanding of the effects of environmental features on the corrosion rate and its time-dependent patterns. The validation is carried out using data from long-term outdoor exposure experiments. The results show that the proposed framework achieves good performance in both single-step and recursive predictions. Environmental temperature and humidity are the primary drivers of the corrosion rate, and a fluctuation period of approximately 90 days, corresponding to seasonal behaviour is observed in the corrosion rate. Simultaneously, the framework’s robustness under recursive prediction indicates that early-stage on-site electrochemical data can effectively support long-term corrosion-rate estimation. The framework provides a practical and experimentally validated tool for predicting the corrosion rate of reinforcement.