This study focuses on developing an optimal setting method for the first integrated coking flue gas desulphurization and denitration device in China. Maintaining the denitration process in a state of optimal economic efficiency has become an issue in production optimization control. This paper proposes a data-based two-stage nonparametric optimization method to optimize the operation of the denitration process. A principal component regression (PCR)-based multiple case fusion case-based reasoning (CBR) method is proposed to obtain the initial optimization set points. To overcome the steady-state modelling difficulties associated with the process, a local modelling method for the coking flue gas denitration process is developed using an improved just-in-time learning (JITL) algorithm. Taking the preset values obtained above as the initial value of an active set algorithm, the optimization problem can be solved in a timely and precise manner. The intelligent setting software has been developed for running industrial applications, and the results demonstrate the effectiveness of the proposed optimization approach.