Data mining diagnosis system based on rough set theory for boilers in thermal power plants

Abstract

Large amounts of data in the SCADA systems  databases of thermal power plants have been used for monitoring, control and over-limit alarm, but not for fault diagnosis. Additional tests are often required from the technology support center of manufacturing companies to diagnose faults for large-scale equipment, although these tests are often expensive and involve some risks to equipment. Aimed at difficulties in fault diagnosis for boilers in thermal power plants, a hybrid-intelligence data-mining system based only on acquired data in SCADA systems is structured to extract hidden diagnosis information directly from the SCADA systems  databases in thermal power plants. This makes it possible to eliminate additional tests for fault diagnosis. In the system, a focusing quantization algorithm is proposed to discretize all variables in the preparation set to improve resolution near the change between normal value to abnormal value. A reduction algorithm based on rough set theory is designed to find minimum reducts from all discrete variables in the preparation set to represent diagnosis rules succinctly. The diagnosis rules mining from SCADA systems  database are expressed directly by variables in the database, making it easy for engineers to understand and use in industry applications. A boiler fault diagnosis system is designed and realized by the proposed approach, its running results in a thermal power plant of Guangdong Province show that the system can satisfy fault diagnosis requirement of large-scale boilers and its accuracy rangers from 91% to 98% in different months.