Data-driven approaches to prediction of inlet NOx concentrations in denitrification systems

In order to optimize the nonlinear and hysteresis control issues in the SCR denitration system, ensure a stable nitrogen oxide concentration at the system outlet, and reduce ammonia injection operating costs effectively, this study investigates the development of an optimized prediction model using a data-driven approach. Regression prediction tests are conducted on four data-driven machine learning models: random forest, SVR, BP-NN, and LSTM-NN. After identifying the optimal model parameters, time series prediction tests for nitrogen oxide concentration at the system inlet are implemented in industrial application scenarios. Ultimately, SVR and LSTM-NN models are selected as the primary predictive control models. The anticipated future optimization and improvement aim for accurate prediction of inlet nitrogen oxide concentration, precise control of ammonia injection denitration, and consistent achievement of outlet concentration standards.