Research On Early Warning And Diagnosis Of Abnormal Health State Of Heavy Gas Turbines

In the context of the current large-scale development of global natural gas resources,the increasing demand for power grid peak shaving,and the rapid development of distributed energy systems,gas turbines are becoming increasingly important in China’s power generation field.In order to ensure that the gas turbine can continue to operate safely,steadily and reliably for a long time,its health status is becoming an important research content.At present,big data technology is booming,and it is a general trend to integrate it into the research on the health status of heavy gas turbines.Therefore,by monitoring real-time data of gas turbines,early abnormalities of the health state are discovered,and the development trend of the health state is predicted by means of data,and abnormal events that affect the health of the unit are processed in time,which is of great significance for improving the life of the unit and reducing maintenance costs.This paper has carried out research on early warning and diagnosis technology of abnormal health status of heavy gas turbines.1)Clarify the structure and function of the health status warning and diagnosis system,and determine the general idea of the whole study.Based on the analysis of the gas turbine equipment and mechanism,the health status indicator that can characterize the health state of the gas turbine is determined,and a gas turbine health state abnormal mode system that can affect the health status indicator is established.2)Filter the data,divide the operating conditions of the gas turbine operating conditions,and apply the random forest algorithm to select health feature parameters.3)Use the isolated forest algorithm to perform abnormal detection on the health feature parameters,and then determine the starting node of the prediction and early warning work.Based on the time series prediction theory,extreme learning machines are used to perform multi-step time series prediction on the health feature parameters backward,Use the forecast value as input to the health prediction model.The GA-BP neural network is used to build a health state prediction model.After the training data is used to construct a complex nonlinear relationship between the health feature parameters and the health state quantity into a mathematical model,The future health status can be determined by inputting health feature parameters predicted by time series and determining the future health status.4)Constructed network reasoning model and based on abnormal health system,the diagnosis and decision of abnormal health after the early warning of the healthy state is realized.