Estimating NOx Sensor Ammonia Cross-Sensitivity Factor Using a Data-Driven Approach in SCR Systems

Selective Catalytic Reduction (SCR) is a module used in the aftertreatment system to mitigate the nitrogen oxides (NOx) emissions, and it can be affected by cross-sensitivity, resulting in a decrease in control performance and even controller failure. A commonly adopted approach is estimating the cross-sensitivity factor to correct the downstream NOx concentration, but the current estimation strategies often incur relatively high costs. In this paper, we propose a novel and more cost-effective method based on data-driven for estimating the cross-sensitivity factor. Specifically, we design a network architecture based on Long Short-Term Memory (LSTM) and attention mechanism. LSTM is utilized to capture the non-linear relationship between input features, historical information, and the cross-sensitivity factor. The introduced attention mechanism allows the model to focus more on important information, thereby further improving the estimation accuracy. This method eliminates the need for complex mechanism modeling and parameter calibration, and reduces the computational burden during online operation, resulting in significant time and computational resource savings.The simulation experiments conducted transient bench data validate the excellent performance of the designed cross-sensitivity factor estimator, as evidenced by comparative analysis with three alternative methods, and demonstrate the rationality of feature selection through the feature ablation analysis.