Applicable subject 3 – Integration and Implementation – Siting and Power Prediction
A Data Mining Approach for Wind Data Preparation for Data MiningEstimation of Data missing Values in Wind Measurements Data Set
Tavares, Geraldo M. 1, Ebecken, Nelson F. F. 2 and Souza, Fábio T. 3
Prof. Geraldo Martins Tavares, D.Sc.
Head of the “Laboratório de Energia dos Ventos da Universidade Federal Fluminense - LEV”
R. Passo da Pátria 156 – Bloco E – Sala 431 – Niterói – RJ – Brazil – CEP 24210240
Tel/telefax: 55 21 2621-4587; E-mail:
One of the most important data set to predict the production of a wind park is the measured data at the place on which the park will be installed. Data missing is one of the problems existing in this data set. This work presents a data mining approach used to replace the missing data in the wind data set
Abstract
Any project concerning the use of wind energy requires the precise wind patterns knowledge of the wind patterns. The data acquisition process and the preparation of a database are background tasks to find the potentially optimum regions to install the system in the chosen region and the output energy generated expected from that system.
The data set from the measurements of the wind parameters used in this work have been performed obtained from measurements mad since September 2001 in Arraial do Cabo’scounty cit, located at y, Rio de Janeiro stateState, Brazil. Three towers have been installed in the region local of atwo proposed Wind Farms and the wind registers have been collected and averaged for each from ten to ten minutesminutes’ intervals. This work intends to presentshow a data mining approach used to replace the missing data in the wind data set.
The predictions of the missing values had been carriedy out by neural network. requiring previous tasks andwith multivariate analyses to reduce the dimensionality. The wind data sets were analyzed by several methodstechnical modern knowledge extraction, including Clustering, Autocorrelation and Principal Component Analysis (PCA).
The samples with all possible patterns forto the neural network training were extracted with a clustering method. The wind seasonal patterns were identified by Kk-Mmeans algorithm. The technical of Autocorrelation and PCA technicals carried out the choice of the variables used to the neural network training.
These previous technical applied to the wind data set allow that variables have been clustered in similar factors or clusters. This issue promotes the replacing of the missing data by by the application of a neural network andwith improves reduction of the computational performancecost. The training, test and predict of the missing values missing were performed by Multilayer Perceptron networks showing very goods results.
CURRUCULUM VITAE
GERALDO MARTINS TAVARES
Phone/fax: ++55 21 26214587
++55 21 3604 5910
++55 21 3604 5911
Email:
EDUCATION
1970 Electrical Engineering degree – Universidade Federal do Rio de Janeiro -
UFRJ, Brazil
1975 Master Degree in Power Systems – Coordenação dos Programas de Pós-
Graduação da Universidade Federal do Rio de Janeiro – COPPE/UFRJ.
1998 Doctor Degree in Production Engineering - Coordenação dos Programas
de Pós-Graduação da Universidade Federal do Rio de Janeiro – COPPE/UFRJ.
PROFESSIONAL EXPERIENCE
ACADEMIC
1992 - ... Professor at the Electrical Engineering Department of the School
Engineering, Universidade Federal Fluminense, Niterói, Rio de
Janeiro, Brazil (1997 – Head of the Wind Energy Laboratory – LEV).
Teaching(main courses taught)
Electric Circuits II, Electric Circuits IV, Introduction to Wind Energy,
Introduction do Distributed Energy and Energy in Agriculture.
Main R&D Áreas
Energy Policies, Wind Energy, Hydropower Plants, Distributed
Energy, Energy Efficiency, Ecoefficiency, data mining.
2001 - … Research Project: Wind measurement for 2 wind parks, with a total capacity of 180 MW, to be installed, at Arraial do Cabo county, RJ, Brazil. Project leader
2001 – 2001 Research Project “ Identification of Areas in the Rio de Janeiro State
for Wind Parks installation. Project leader.
2002 - ... Research Project “MICROTUR: Definition of a Strategy for Introduction Distributed Generation in the Brazilian Power System”. Project leader.
2002 - ... Research Project “PUFAE: “UFF’s Project for Renewable Energy and Energy Efficiency. Project leader. .
1999 - ... Research Project “Ecoefficiency Program of the Universidade Federal Fluminense”. Project leader.
2001 - 2002 Research Project: Electrical load prevision using data mining techniques. Project member
INDUSTRIAL
From 1971 to 1992 I’ve worked at the Compania Hidrelétrica do São Francisco – CHESF – and at the Centrais Elétricas do Sul do Brasil – ELETROSUL, two of the biggest Brazilian’s utilities, and at MDK Engineering, ENGEVIX Engineering and International Engineering, three of the most important Brazilian’s consulting companies in the area of energy.
PUBLICATIONS
· Doctorate Thesis: A Strategic Approach for the Implementation of the Commercial Production of Electrical Wind Energy in Brazil. presented to the Production Engineering Program of the COPPE/UFRJ, Rio de Janeiro, November 1998.
· Tavares, G.M. Wind Energy in Brazil. Post section of the DEWEK 2002. Germany, October, 2002.
· Tavares, G. M. Wind Energy Environmental Considerations. Presented at the Third International Seminar About Urban Problems - ECO URBS 95. Rio de Janeiro, June, 1995.
· Tavares, G.M. Comparison of the Strategies Used for Wind Energy Implementation. Presented at the XII Electrical Energy Production and Transmition National Seminar. Camboriu, Santa Catarina, Brazil, October 1995.
Niterói, April 22, 2003
The Measure Correlate Predict – algorithm (MCP) based in on principles of artificial intelligence tools aidaids to the reduction of uncertainties relationship into the implementation of Wind Farm sitting and must be used to the extract knowledge from the wind data.
1 – UFF - Federal University Fluminense <
2 – COPPE/UFRJ – Federal University of Rio de Janeiro <>
3 – COPPE/UFRJ – Federal University of Rio de Janeiro <>