Research On Schema Structure Learning And Application Driven By Relational Data Interpretation

Relational data interpretation is an important content for large data processing.It tries to obtain the representation and reasoning of undertain knowledge by mining associative rules among observed data.In existing relational data mining studies,mixed relational rules of observed data consisting of continuous numerical and label features can not be effectively extracted.For nonintrusive power load monitoring problem,current relational data modeling and recognition methods still needs to be improved to gain higher detection performance.In view of the problems of existing relational data rule mining,two developed algorithms are studies.In order to model and analyze the mixed feature data containing continuous numeric and label features,a new data representation model,self-explainable reduction model SRM is proposed.In the proposed model,a novel reduction data model objective is designed to realize adaptive discrete division for continuous data dimension.Several experimental analysis are executed on standard datasets,simulated data mining problems,and a practical application problem.Experimental results indicate that the proposed model is very useful and valuable to extend the abilities of existing data mining algorithms.For non-intrusive power load monitoring problem,the key trouble is that there contains complex multiple types of electrical equipments in a power load environment.In this thesis,two suppositions are firstly considered: 1)power signal characteristics at any time point could be modeled by Gaussian mixture model of the singal characteristics of independent electic equipments,2)the signal characteristics shoud keep stable for each loaded equipment with a same running state for continuous times.Thus,a coupled allocation mechanism on mixed probability state labels is designed to realize a new non-intrusive power load monitoring method.Wherein,interative filter processing algorithm is designed to solve optimal state allocation of different loaded equipments.According to professional silulation experiments,corresponding results indicate that better comprehensive performance can be obtained compared to latest fuzzy clustering method and hidden Markov method.