Wall Temperature Monitoring And Anomaly Analysis Based On Data Driven Framework

With the vigorous development of China’s energy industry,the parameters of thermal power plants are getting larger and larger,and the requirements for safe operation of power plants are also getting higher and higher.Among them,boiler tube explosion accidents occur frequently in thermal power plants,which are mainly caused by high metal wall temperature and large steam temperature fluctuation.For this problem,researchers have done a lot of work,such as the design of on-line monitoring system,optimization of measuring point layout,wall temperature calculation and life evaluation of heating surface,optimization of wall temperature calculation model,etc.,but there are still the following problems: the work content of equipment point inspection is complex,the equipment state evaluation depends on subjective evaluation,the lack of parameter trend analysis,one-sided attention to parameter limit alarm,the lack of automatic monitoring system,the lack of automatic monitoring system.It is rare to find the potential fault of equipment actively,and lack of diagnosis experience accumulation and sharing.Now artificial intelligence has been applied in all fields of life,and there are countless applications for power plant direction,but most of them are used in load forecasting,system optimization and so on.This paper uses data-driven method to predict and detect the wall temperature,mainly for the early warning of the state of the equipment,based on the idea of data mining equipment state data analysis,to show the data-driven dynamic analysis standard rather than the fixed standard.From traditional alarm to early warning:from fixed upper and lower limits of actual value to dynamic band of deviation value.The main research work includes:1、ARIMA model is used to predict the boiler wall temperature,and sarimax model is added to predict the exogenous variable load.Finally,the ARIMA simulation results are used for anomaly detection based on the model.The specific method is to use the Euclidean distance between the predicted value and the real value as the anomaly score to judge the anomaly.2、The distance based anomaly detection algorithm is used to detect one-dimensional wall temperature data,in which Euclidean distance is used to compare with Sax method.Results both methods can identify abnormal temperature data.3、The abnormal temperature data of high dimension are detected by DTW.When 300 groups of wall temperature data are detected,abnormal temperature data can be identified.Finally,comparing the time used in various methods,it is found that the speed of the anomaly detection program is much higher than that of real-time data,which can realize the on-line anomaly detection.4、The abnormal detection algorithm based on model and distance is used to analyze the temperature data of 209 measuring points.The results show that the temperature of drum and final superheater fluctuates greatly.Which of the eight devices has the biggest temperature fluctuation.Finally,the two methods are compared,and the results show that the distance based anomaly detection method is better than the model-based anomaly detection method.At the same time,it also proves the feasibility of the algorithm.