Fault Pattern Recognition of Thermodynamic System Based on SOM

Thermodynamic system has great effect on security and efficiency of modern thermal power station. According to experiences, thirteen fault types and nine kinds of symptom parameters were summarized and abstracted to diagnose faults. After fuzzy processing and normalization, feature pattern knowledge base was set up on the base of the processed data. Self-organization feature map (SOM) was chosen to establish an expert system that was completed by MATLAB, with nine neurons in input layer and 8×8 two-dimensional competition layer. The expert system was trained by the data of knowledge base. To test the actual performance of SOM, six input samples were used for simulation by MATLAB. Five samples were diagnosed accurately and only one sample was fuzzy. It corresponds with the result of simulation machine. Furthermore, according to simulation result, the number of arranged neurons in competition layer is important to SOM performance. Through mixed programming and application programming interface (API) provided by MATLAB, the expert system can easily be developed to online diagnosis system. Now the online expert system has been used for about one year in a thermal power station and the performance is satisfactory.