Anomaly Monitoring Of Thermal Power Plant Equipment Based On Machine Learning

Thermal power generator set is a large complicated system,its safety and stablity are extremely crucial to electricity production.Due to its high complexity and coupling of each equipment,the abnormal or mal-function of a single equipment can lead to the decrease of reliability and efficiency of the whole unit.Therefore,anomaly monitoring for thermal power plant equipment is a fundamental task to ensure the safety and efficiency of thermal power production.With the development of the industrial informatization,large-scale thermal power generator sets are generally equipped with a great number of sensors to monitor the running status of each equipment in real time.Therefore,the identification and diagnosis of the abnormal operating conditions of the unit based on the analysis of the real-time operating variables have attracted widespread attention in the academic and industrial societies.Considering the coal mill and steam turbine in thermal power units,this paper studies anomaly monitoring method based on real-time operating variable analysis.First,this paper reviews the existing studies of fault detection,diagnosis and prediction technologies based on machine learning,with the discussions about their advantages and disadvantages.Seconde,aiming at abnormal performance detection for thermal power plant equipment,an anomaly detection framework is designed.Furthermeore,ean "operating variable/performance indicator" correlation model is proposed,and an estimation method based on support vector regression is used for unit performance anomaly detection.On this basis,the selection strategies of different control limits are discussed.For the performance anomalies,a similarity-based method is utilized to diagnose abnormal variables.This method is verified and analyzed by the real operating data from some coal mills.Third,as for the abnormal condition prediction,considering the correlation between the multieple variables and the correlation in the time series,multivariable analysis and attention mechanism are introduced into the recurrent neural network model.Two network structure are designed for single-step-predict and multi-step-predict of the target variable.This method is then adopted for predicting failure time of a steam turbine.The experimental results validate the effectiveness of the algorithm.Fourth,targeting at monitoring and analysis of milling performance,an online monitoring platform for milling system is developed and tested.This system is used to improve the intelligent level and ensure the safe and efficient operation of the milling system.Finally,we summary the work of this paper,and discuss the potential research work that can be done in the future.