Geodesic Distance Distance based based Geodesic Fuzzy Clustering - PowerPoint PPT Presentation

Geodesic Distance Distance based based Geodesic Fuzzy Clustering Clustering Fuzzy Abonyi and and Bal Já ános nos Abonyi Balá ázs zs Feil Feil J University of of Pannonia Pannonia University abonyij@ @fmt.uni fmt.uni- -pannon.hu pannon.hu abonyij www.fmt.uni- -pannon.hu pannon.hu/ /softcomp softcomp www.fmt.uni

Administrative details Administrative details � Webpage : Webpage : � � www.fmt www.fmt. .uni uni- -pannon pannon. .hu/softcomp hu/softcomp � • You can download • You can download � Transparencies Transparencies � � Related papers in PDF Related papers in PDF � � Demonstration programs Demonstration programs � � Please, send an e Please, send an e- -mail to: mail to: abonyij@fmt. abonyij@fmt.uni uni- -pannon pannon.hu .hu � � Book: Book: � � Fuzzy Clustering for Data Mining Fuzzy Clustering for Data Mining � and System Identification (coming soon! !) ) and System Identification (coming soon � MATLAB MATLAB Toolbox Toolbox � � File exchange: more than File exchange: more than 7 7000 users worldwide 000 users worldwide � 2/ /47 47 2

Location Location DPE Laboratory DPE Laboratory Veszpré ém m Veszpr DPE offices DPE offices Buildings of the University of Pannonia Pannonia Buildings of the University of 3/ /47 47 3

Industrial Process Development Industrial Process Development � Process engineering covers all the necessary � Process engineering covers all the necessary knowledge required for defining, designing, defining, designing, knowledge required for implementing and optimizing any process implementing and optimizing any process � Nowadays, the task of process engineers is to design, Nowadays, the task of process engineers is to design, � construct and operate complete complete systems. systems. construct and operate SZABÁLYOZÓ SZABÁLYOZÓ RÁTÁPLÁLÁS RÁTÁPLÁLÁS (Vc) BECSÜLT (Vc) BECSÜLT τ 1 , K 1 τ 1 , K 1 SLAVE SLAVE C 1 C 1 T T τ 2 , K 2 τ 2 , K 2 MASTER MASTER C 2 C 2 A A M M V 2 NYITVA V 2 NYITVA FOLYAMAT MODELL FOLYAMAT MODELL T J T J MÉRÉS MÉRÉS STRUKTÚRA STRUKTÚRA ÁLLAPOTBECSLÉS ÁLLAPOTBECSLÉS T T T W T W T, T J T, T J VÁLASZTÁS VÁLASZTÁS IDENTIFIKÁLÁS IDENTIFIKÁLÁS V 1 ZÁRVA V 1 ZÁRVA τ 4 , K 4 τ 4 , K 4 MASTER MASTER T T C 4 C 4 τ 3 , K 3 τ 3 , K 3 SLAVE SLAVE T J T J C 3 C 3 H Ű TÉS H Ű TÉS 4/ /47 47 4

Department of Process Engineering at the University of Veszprém dr. Szeifert dr. Chován dr. Abonyi dr. Németh dr. Árva dr. Moser dr. Lakatos dr. Nagy Ferenc Tibor János Sándor Péter Károly Béla Lajos Industrial Batch Process Tailored Hierarchical Modeling Process Process Data Process Process and Control Automation Design Mining Simulators Modeling of Crystallizers dr. Janos Madar, Ferenc P. Pach, Balazs Feil , Balázs Balaskó The Optimization of Operating Processes Project: www.fmt.vein.hu/softcomp/procopt The EAsy MATLAB Toolbox www.fmt.vein.hu/softcomp/EAsy The GP MATLAB Toolbox www.fmt.vein.hu/softcomp/gp 5/ /47 47 5

Computational Intelligence in Data Mining CI in modeling and control Advanced Model Based Process Engineering Tools www.fmt.uni-pannon.hu/softcomp 6/ /47 47 6

Optimization of operating technologies Optimization of operating technologies Problem Description (e.g.) Problem Description (e.g.) HDPE- -I I plant of TVK Ltd plant of TVK Ltd HDPE � Operating Operating � multi- - product technology product technology multi � Large volume (60.000t/y), � Large volume (60.000t/y), high value added products high value added products (more than (more than ten ten) ) � Using the same process Using the same process � equipment, equipment, (process transitions) (process transitions) � Goal Goal: : � optimization of of the the optimization technological parameters technological parameters 7/ /47 47 7

Know- -how: how: Know Process Data Data Warehouse Warehouse Process Enterprise Information Enterprise Information System System � T � The mountains of data he mountains of data that that computer- -controlled plants controlled plants computer 3. generate must be used generate must be used Adequate reports Adequate reports � by the operator support by the operator support � 2. based on electronic based on electronic stored data stored data systems to distinguish normal systems to distinguish normal 1. from abnormal operating from abnormal operating Consistent Consistent information sources, information sources, conditions conditions regulated access regulated access � to � to optimize the technology optimize the technology Partially overlapping, Partially overlapping, non-consistent, mainly non-consistent, mainly and the product and the product paper based reports paper based reports � to plan and schedule to plan and schedule � sequences of operating steps. sequences of operating steps. � The aim of the The aim of the data � data warehouse warehouse is to organize and store the data is to organize and store the data taken from different information taken from different information sources by different sampling sources by different sampling times and allow the users to times and allow the users to query, analyze and group these query, analyze and group these data. data. 8/ /47 47 8

Know- -how: how: Know Interactive Process data mining Process data mining Interactive Data mining (knowledge discovery in databases): Extraction of interesting (non-trivial, implicit, previously unknown and potentially useful) information or patterns from data in large Prior databases Knowledge, goals � Learning the Learning the application domain: application domain: � � relevant prior knowledge and goals of relevant prior knowledge and goals of � Pattern Evaluation application application � Creating a target data set: data selection Creating a target data set: data selection � Data Data Mining � Data cleaning Data cleaning and preprocessing: and preprocessing: � Analysis (may take 60% of effort!) (may take 60% of effort!) Task-relevant Data � Data reduction and transformation � Data reduction and transformation: : Database � Find useful features, Find useful features, � Data � Choosing functions of data mining Choosing functions of data mining � Warehouse Selection � summarization, classification, � summarization, classification, regression, association, clustering. regression, association, clustering. � Choosing the mining algorithm(s) Choosing the mining algorithm(s) � Data Cleaning � Data mining: Data mining: search for patterns of interest search for patterns of interest � � Pattern evaluation and knowledge Pattern evaluation and knowledge � Data Integration presentation presentation � visualization, transformation, removing � visualization, transformation, removing Technology Databases redundant patterns, etc. redundant patterns, etc. � Use of discovered knowledge � Use of discovered knowledge 9/ /47 47 9

Why Mine Data? Why Mine Data? Commercial Viewpoint Commercial Viewpoint � Lots of data is being collected Lots of data is being collected � and warehoused and warehoused � UC Berkeley 2003 estimate: UC Berkeley 2003 estimate: � 5 exabytes 5 exabytes (5 million terabytes) of new data was created in 2002. (5 million terabytes) of new data was created in 2002. • Relational database Relational database • • Huge data warehouses are under construction Huge data warehouses are under construction • • WWW: A huge, hyper WWW: A huge, hyper- -linked, linked, • dynamic, global information system dynamic, global information system • e- -commerce commerce • e • Text (documents, emails) Text (documents, emails) • and multimedia databases and multimedia databases � Competitive Pressure is Strong Competitive Pressure is Strong � � Provide better, customized services for an � Provide better, customized services for an edge edge (e.g. in Customer Relationship Management) (e.g. in Customer Relationship Management) � Computers have become cheaper and more powerful Computers have become cheaper and more powerful � 10/ /47 47 10

Why Mine Data? Scientific Viewpoint Why Mine Data? Scientific Viewpoint � Data collected and stored at � Data collected and stored at enormous speeds (GB/hour) enormous speeds (GB/hour) � Time Time- -series data series data � � microarrays microarrays generating gene generating gene � expression data expression data � scientific simulations scientific simulations � generating terabytes of data generating terabytes of data � Traditional techniques infeasible Traditional techniques infeasible � for raw data for raw data � Data mining may help scientists Data mining may help scientists � � in classifying and segmenting data in classifying and segmenting data � � in in Hypothesis Formation � Hypothesis Formation 11/ /47 47 11