Abstract: Current denitration systems of thermal power plants has some problems such as delayed measurement at denitration inlets and outlets， distorted values at some NOx measuring points， and inconsistency between values measured at denitration reactor outlets and denitration environmental protection devices. To deal with these problems， a 300 MW power plant is taken as an example for the intellectualization and optimization of its denitration control system. The optimized denitration system can meet the requirements of ultra-low emission， and prevent ammonia escape led by excessive ammonia. Artificial intelligence technology is used in building the SCR denitration system model which can be continuously improved by powerful self-taught learning and intelligence development of the technical optimization control system（TOCS）.Then，the operation state can be predicted and the control quantity can be accurately calculated.After the running of TOCS， quality control of denitration systems can be largely improved with reduced ammonia spraying volume and mitigated ammonia escape on the premise of NOx emission conforming to standards.