Research On The Key Problems Of Anomaly Detection System Of Gas Turbine

In the process of progress for industrialization, informatization and intelligent, it iscrucial to have some intelligent but practical diagnosis technology to monitor an equip-ment and then to prevent accidents through the whole operation process. As new corepower equipment, the gas turbine starts to be used more widely after this gets the atten-tion from the world. Obviously, when the gas turbine is in work status, in order to predictand then to avoid significant changes, it must be very important to know how to monitorequipment and how to diagnose problem efectively during the operation process. Due tothe great engineering application value and theoretical value for the research of the gasturbine circuit fault diagnosis, more countries are committed to the research and develop-ment of gas turbine diagnosis systems.The objective of this paper is to study the key technical issue through the designof gas turbine diagnosis system based on the data form, including some testing methodsfor point form and sequence form for gas turbine fault, and the cold start problem duringearly construction time, as well as the growth method for anomaly detection system.This paper starts with a brief description for the performance of gas turbine break-down, and then analyzes whether the anomaly detection method could be helpful for thediagnosis. Also, an anomaly detection method using afnity propagation clustering hasbeen put forward to solving the anomaly point form of the combustion engine failure.With less parameter settings, it will be more easily to add some priori information to amachine under this detection methodThen a method based on weighted manifold embedded is proposed to detect abnor-mal sequential form. Followed by the description of this method, more comprehensiveanalysis and comparison of the performance and impact factors are illustrated. In anapplication of assisting human brain to recognize the fault data, a combustion enginesequence sample is embedded into a two-dimensional space to be displayed visually.Furthermore, this article focuses on resolving the cold start problem during earlyconstruction time of anomaly detection system. The cold start problem relates to systemdependency to data. In the early stage of system construction, the data in the system doesnot have enough data resources. Thus, a classification method based on neural network complex structure has been used for anomaly detection of the gas turbine data. Con-sidering of preliminary improving the accuracy for the model, an idea has been postedthat a single model will decide whether to receive the samples, the model will also applyanomaly detection mechanism under false rejection framework to filter and select data.In the end, this paper shows the growth problem for anomaly detection system.Based on the enrichment of applied models of anomaly detection system, dynamic multi-model fusion detection method will be used to detect the operation status of the gas turbinein order to increase the accuracy rate. With sufcient base detection models, the adaptiveprocessing strategy is used for each sample to be detected, and then dynamically selectingthe best suited base model and judging the final integration of all results.