Study On The Application Of Rough Set Attribute Reduction Algorithm In Thermal System

Rough set theory is used as a tool for dealing with uncertain information effectively.It is widely used in many fields and provides a new method for modeling and fault classification of complex thermal systems.This paper describes the combination of rough set and neural network,the steps of the method are preprocessing of raw data,attributes reduction to the construction of network.Among them,for the discretization of continuous attribute,an improved k-means algorithm is proposed,which does not need to specify the number of clusters.The concept of consistency degree is introduced to determine the penalty parameter of the algorithm,Fully considered the relationship between the properties.Aiming at the attribute reduction of decision table,an reduction algorithm based on approximate decision entropy is proposed,which combines the approximate precision with the traditional information entropy.It solves the shortcoming of the traditional information entropy which simply start from the point of information theory.In order to solve the problems that real-time modeling of thermal objects is difficult,the model accuracy is not high,and the convergence rate of neural network is reduced greatly due to the increase of inputs.In this paper,a method of BP network modeling based on approximate decision entropy is proposed.The method is applied to the modeling of main steam temperature and NOx emission concentration.The results show that this method has high precision and lowers the dimension of BP network input layer,the network structure is simplified and the training speed is more fast,which has important practical value for modeling thermal system.In order to realize the fast and accurate judgment of equipment fault types in power plant,the attribute reduction algorithm based on the approximate decision entropy model is combined with the BP network to classify the faults.The method is applied to fault classification of steam turbine vibration.The result shows that the combined method effectively reduces the input dimension of network and simplifies the structure,the high classification accuracy is obtained,the computing cost is reduced.It is of practical value to judge the type of fault quickly and improve the efficiency of maintenance.