Research on diagnosis for vibration faults in steam turbines using IRF algorithm

The random forest algorithm， known for its robustness against noise and powerful computational capabilities， is widely employed in diagnosing vibration faults in rotating machinery. However， when applied in industrial settings， the algorithm encounters challenges such as limited sample sizes， the inability to integrate prior knowledge， and comparatively lower accuracy. To tackle these issues， a method for diagnosing vibration faults in steam turbines using an improved random forest （IRF） algorithm is proposed. This approach incorporates prior knowledge to optimize decision trees， utilizing analytical hierarchy process （AHP） and information entropy. Genuine operational datasets from the data center of a million-kW thermal power plant are utilized to validate the efficacy and reliability of the proposed method. Computational findings indicate that， in comparison to the traditional random forest algorithm， IRF achieves higher accuracy and a reduced miss rate， with a 33% decrease in the number of decision trees. Moreover， the operational time is slashed to just 11.4% of that taken by the traditional random forest algorithm. These results suggest that IRF holds significant practical value for real-time， precise vibration fault diagnosis in thermal power units.