Research On Intelligent Analysis And Modeling Of Control Valve Big Data Based On Parallel Computing

Process control is an important branch of modern industrial-controlling field,and control valve is the most important terminal component of the process control system.With the development of automation and information technology,a large process control system with thousands of control valves can collect massive real-time signals and store them as large historical data.At present,due to the complexity and time-consuming of big-data analysis,the control valve big-data has not been effectively used,resulting in the phenomenon of“rich data and lack of information,.With the support of the National 863 Plan and the National Natural Science Foundation of Chinath,this paper combines Internet of things technology,distributed computing technology and artificial intelligence technology,to monitor,control and analyze the real-time big-data and historical big-data of the control valve.The main research works are listed as follows.The design of the embedded control system of control valve.For the direct-drive electro-hydraulic control valve,an embedded control system based on Internet of Things technology is designed.The system uses ARM Cortex-M3 SCM as the control core,and realizes the functions that are sensor signal acquisition,touch screen display/input,PWM servo motor driving,Ethernet communication,control valve opening control,and so on.A discrete control algorithm is used for the opening control.The dead zone phase,the linear phase and the lag phase in the running process of the valve are analyzed,respectively.A digital controller is used to correct the lag phase.The positioning accuracy of the valve is less than 0.1%FS.A fault diagnosis method for control valves based on parallel SDP algorithm and large data driving is proposed.Simulating the working condition of a control valve in Jigang cycle power generation project,a control valve experimental platform was built and some fault states were simulated.The samples of the normal running state and seven types of failure states were collected.Each sample includes five characteristics:front pressure,post valve pressure,load pressure,valve opening and flow rate.In this paper,first use the t-SNE data visualization algorithm to preprocess the data,and map the high-dimensional data to a low dimensional space.Then,according to the data distribution in the low-dimensional space,the Searching Density Peaks(SDP)clustering algorithm is selected to establish the fault diagnosis model.Because of the high time-consuming of SDP algorithm when dealing with big-data,this paper combines the SDP algorithm with Spark MapReduce framework to realize a parallel SDP(pSDP)algorithm.The experimental results show that,compared with other methods,the model established in this paper has a high prediction accuracy,and Spark MapReduce has a good acceleration effect on SDP,which greatly reduces the time consumption of modeling.A large-scale real-time fault diagnosis method for control valves based on Spark Streaming and fault diagnosis reasoning model is proposed.The real-time data collected by a process control system with a large number of control valves has the characteristics that are large flow,multi class and poor synchronization.Due to the limited computing power and fault tolerance,it is very dificult for traditional methods to analyze real-time big-data.In this paper,a distributed real-time fault diagnosis system based on Spark Streaming and the pSDP model is designed.The system also integrates two components:the distributed message collector Flume and the distributed message subscriber/publisher Kafka.The maximum nunlber of real-time signals that the system can process at different signal sampling intervals and CPU core numbers is experimentally studied.The results show that the system has good performance in processing sub-second real-time big-data.Several main factors affecting the performance of the system are analyzed,including the size of the fault diagnosis model,the sampling time interval and the number of equipment types.A method of predicting the back pressure of control valve based on parallel decision tree regression analysis and large data driving is proposed.Establishing an accurate mathematical model is very important for the automatic control and the fault diagnosis of control valve.Because of the high coupling and nonlinearity of the control valve structure,it is very difficult to model the control valve by mechanism,the researchers usually use the regression analysis method to model control valve.The common algorithms,such as neural network,support vector machine(SVM)and other regression algorithms,have high computational complexity.They are only suitable for small sample processing,which makes it difficult to excavate more profound rules in big-data.This paper uses linear regression,decision tree regression,random forest regression and progressive gradient tree regression algorithm in Spark MLlib to build control valve regression models on a large dataset contain six million samples..The models can predict the back pressure of control valve according to the input parameters,such as the load pressure and flow rate.The performances of the four algorithms is analyzed by experiments.The results show that decision tree regression has obvious advantages in modeling precision and modeling speed,which is the most suitable for control valve modeling.An intelligent operation system(IOC)for regulating valve network is developed,which integrates and visualizes the equipment information and data analysis results(fault diagnosis,Pressure prediction,etc.)mentioned above,and provides decision-making planning for the operation and management of the network.On one hand,traditional monitoring systems can only be carried out on the fixed location and fixed equipment in a computer monitoring room.On the other hand,their intelligence level is low,and there is no comprehensive data mining analysis.By using SSH server,iOS,Web and other technologies,the IOC can fully support web pages,mobile terminals and desktop terminals,and it can monitor devices anytime and anywhere.A parallel ant colony algorithm based on Spark MapReduce is used to realize efficient optimal route planning for pipeline inspection.The system aims to enhance the engineering personnels’perception and insight of the pipe network,improves the level of decision-making analysis and emergency command,and provides a reliable guarantee for the stabilities of production and life.