Abstract
With the rapid development of information technology, the informatization, integration and complexity of more and more large equipment are increasing day by day, so it is very important to carry out fault diagnosis for such complex equipment. In the traditional way, expert system technology is usually used for fault diagnosis of complex equipment. However, with the increasing of equipment data information, traditional methods cannot solve the fault diagnosis requirements in the case of a large amount of data. Therefore, data-driven fault diagnosis method can solve this problem, The carrier of data-driven fault diagnosis is a large amount of engineering data, and its focus is to explore new methods of fault diagnosis from a large amount of historical data. In this paper, the classical random forest algorithm is selected as the basic model, and aiming at the imbalance of complex equipment data, the improved random forest voting mechanism based on the fault ratio is proposed to optimize the model, which makes the final model diagnosis accuracy more than 95%, and has good application value.