Research On Chinese Textual Data Mining Techniques And Reliability Applications In Power Systems

As the development of intelligence and informazation in power grids,the volumn of data is increasing dramasticly in the database.It is the issue of big data in power systems.In this theis,all the data could be named life cycle data,where are generated from lifecycle process of power production and electric equipment.Life cycle data in smart grid includes structured data and unstructured data.The previous researches were mainly focusing on structured data mining,and someone employed the image recognition techniques in smart grids.However,few discussed the textual data mining in power system domain.This thesis focus on text mining techniques,and then establish the lifecycle data mining framework to acquire the reliability indices,including statistics,health index,failure rate and system reliability for asset management in power grids.1.According to the characteristics of reliability and asseset management,it introduced the concepts and framework of Natural Language Processing for power grid(NLP4PG).We compiled a domain-specific lexicon,ontology-based lexicon for transformer and domain-specific corpus,which partly have been uploaded in the website for cooperation.The NLP4SG has embedded with ontology theory and domain knowledge.Several examples of linguistc analysis showed the components and potential applications in NLP4PG,which filled the void of Chinese text mining in power system domain.2.The third section combined the statistical learning off-line and rule-based syntactic technique on-line together,so as to establish a text mining framework based on ontology-lexicon for reliability statistics automata(RSA)of transformer.The example showed that it could solve the problem of Chinese sentence hard to segmentation and mixed with quatifiers.RSA could aid the power company to calculate the reliability indices in some degree.3.Considering the multiple sources and types of Health Index(HI),the kNN algorithms has been improved to interval-based self-learning text classification model.It could classify the text into HI rigorously by studying the similar texts of assets.Then,the proportional health fusion model considered the historical information by text mining.In some degree,this machine learning tool could assess the health index instead of subjective evaluations.4.A non-parametric failure-rate model based on martigale process was proposed.It could not only fuse the HI mined from lifecycle data,but also handle the multi-types recurrent events.The case study of transformers showed that it could depict the short-term failure rate throughout parameter estimation,martingale residule testing and sensitivity analysis.The extended failure rate provided the basic reliability index for remaining useful life prediction and maintenance optimization.5.Additionally,a generalized reliability assessment model based on ANP(Analytic Network Process)was built appropriate for calculating reliability of new system with few lifecycle data,such as SSAS(Smart Substation Automation System).Ontology knowledge was summarized from the IEC61850 standards,design documents and literatures.The ontology platform of SSAS was divided into three parts,physical ontology,logic ontology and information ontology.Accordingly,the system reliability contained reliability degree of physical ontology,logic ontology,information ontology and economy.Many qualitatively and quantitatively indices belonged to the four macro-indices above.The results and sensitivity analysis showed that this ontology knowledge-based model could help the designers to select the scenarios from candidates.