the results to larger systems due to prop'erties of the projection algorithm. First, the number of hidden nodes must

Similar documents
Enhancing Performance of MLP/RBF Neural Classifiers via an Multivariate Data Distribution Scheme

Chapter 3: Cluster Analysis

COMP 551 Applied Machine Learning Lecture 11: Support Vector Machines

Determining the Accuracy of Modal Parameter Estimation Methods

NUROP CONGRESS PAPER CHINESE PINYIN TO CHINESE CHARACTER CONVERSION

initially lcated away frm the data set never win the cmpetitin, resulting in a nnptimal nal cdebk, [2] [3] [4] and [5]. Khnen's Self Organizing Featur

CS 477/677 Analysis of Algorithms Fall 2007 Dr. George Bebis Course Project Due Date: 11/29/2007

Dataflow Analysis and Abstract Interpretation

A Scalable Recurrent Neural Network Framework for Model-free

Churn Prediction using Dynamic RFM-Augmented node2vec

Revision: August 19, E Main Suite D Pullman, WA (509) Voice and Fax

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

Pattern Recognition 2014 Support Vector Machines

k-nearest Neighbor How to choose k Average of k points more reliable when: Large k: noise in attributes +o o noise in class labels

Least Squares Optimal Filtering with Multirate Observations

Lecture 2: Supervised vs. unsupervised learning, bias-variance tradeoff

Simulation of Line Outage Distribution Factors (L.O.D.F) Calculation for N-Buses System

Bootstrap Method > # Purpose: understand how bootstrap method works > obs=c(11.96, 5.03, 67.40, 16.07, 31.50, 7.73, 11.10, 22.38) > n=length(obs) >

Early detection of mining truck failure by modelling its operation with neural networks classification algorithms

1996 Engineering Systems Design and Analysis Conference, Montpellier, France, July 1-4, 1996, Vol. 7, pp

Tree Structured Classifier

Technical Bulletin. Generation Interconnection Procedures. Revisions to Cluster 4, Phase 1 Study Methodology

Aerodynamic Separability in Tip Speed Ratio and Separability in Wind Speed- a Comparison

Engineering Approach to Modelling Metal THz Structures

Surface and Contact Stress

Design and Simulation of Dc-Dc Voltage Converters Using Matlab/Simulink

Numerical Simulation of the Thermal Resposne Test Within the Comsol Multiphysics Environment

Learning to Control an Unstable System with Forward Modeling

February 28, 2013 COMMENTS ON DIFFUSION, DIFFUSIVITY AND DERIVATION OF HYPERBOLIC EQUATIONS DESCRIBING THE DIFFUSION PHENOMENA

Exam #1. A. Answer any 1 of the following 2 questions. CEE 371 October 8, Please grade the following questions: 1 or 2

Resampling Methods. Cross-validation, Bootstrapping. Marek Petrik 2/21/2017

Resampling Methods. Chapter 5. Chapter 5 1 / 52

COASTAL ENGINEERING Chapter 2

What is Statistical Learning?

SIZE BIAS IN LINE TRANSECT SAMPLING: A FIELD TEST. Mark C. Otto Statistics Research Division, Bureau of the Census Washington, D.C , U.S.A.

FIELD QUALITY IN ACCELERATOR MAGNETS

5 th grade Common Core Standards

MINIMIZATION OF ACTUATOR REPOSITIONING USING NEURAL NETWORKS WITH APPLICATION IN NONLINEAR HVAC 1 SYSTEMS

ANALYTICAL SOLUTIONS TO THE PROBLEM OF EDDY CURRENT PROBES

Slide04 (supplemental) Haykin Chapter 4 (both 2nd and 3rd ed): Multi-Layer Perceptrons

THERMAL-VACUUM VERSUS THERMAL- ATMOSPHERIC TESTS OF ELECTRONIC ASSEMBLIES

A Matrix Representation of Panel Data

^YawataR&D Laboratory, Nippon Steel Corporation, Tobata, Kitakyushu, Japan

and the Doppler frequency rate f R , can be related to the coefficients of this polynomial. The relationships are:

3D FE Modeling Simulation of Cold Rotary Forging with Double Symmetry Rolls X. H. Han 1, a, L. Hua 1, b, Y. M. Zhao 1, c

Time, Synchronization, and Wireless Sensor Networks

Exam #1. A. Answer any 1 of the following 2 questions. CEE 371 March 10, Please grade the following questions: 1 or 2

Copyright Paul Tobin 63

More Tutorial at

CAUSAL INFERENCE. Technical Track Session I. Phillippe Leite. The World Bank

Activity Guide Loops and Random Numbers

The standards are taught in the following sequence.

Chapter 3 Kinematics in Two Dimensions; Vectors

Figure 1a. A planar mechanism.

A New Evaluation Measure. J. Joiner and L. Werner. The problems of evaluation and the needed criteria of evaluation

Performance Bounds for Detect and Avoid Signal Sensing

3 / Generalized Immittance Based Stability Analysis

Analysis on the Stability of Reservoir Soil Slope Based on Fuzzy Artificial Neural Network

THE TOPOLOGY OF SURFACE SKIN FRICTION AND VORTICITY FIELDS IN WALL-BOUNDED FLOWS

Artificial Neural Networks MLP, Backpropagation

x 1 Outline IAML: Logistic Regression Decision Boundaries Example Data

Methods for Determination of Mean Speckle Size in Simulated Speckle Pattern

Physical Layer: Outline

Web-based GIS Systems for Radionuclides Monitoring. Dr. Todd Pierce Locus Technologies

IAML: Support Vector Machines

Study Group Report: Plate-fin Heat Exchangers: AEA Technology

Heat Management Methodology for Successful UV Processing on Heat Sensitive Substrates

Verification of Quality Parameters of a Solar Panel and Modification in Formulae of its Series Resistance

Particle Size Distributions from SANS Data Using the Maximum Entropy Method. By J. A. POTTON, G. J. DANIELL AND B. D. RAINFORD

Chapter Summary. Mathematical Induction Strong Induction Recursive Definitions Structural Induction Recursive Algorithms

Instructions: Show all work for complete credit. Work in symbols first, plugging in numbers and performing calculations last. / 26.

1. INTRODUCTION ARTIFICIAL INTELLIGENCE APPLICATIONS IN ON-LINE DYNAMIC SECURITY ASSESSMENT. ABS'fRACT

Optimal Locations and Sizing of Capacitors for Voltage Stability Enhancement in Distribution Systems

T Algorithmic methods for data mining. Slide set 6: dimensionality reduction

Turing Machines. Human-aware Robotics. 2017/10/17 & 19 Chapter 3.2 & 3.3 in Sipser Ø Announcement:

Review Problems 3. Four FIR Filter Types

Floating Point Method for Solving Transportation. Problems with Additional Constraints

Chapter 2 GAUSS LAW Recommended Problems:

Module 4: General Formulation of Electric Circuit Theory

Sections 15.1 to 15.12, 16.1 and 16.2 of the textbook (Robbins-Miller) cover the materials required for this topic.

ENG2410 Digital Design Sequential Circuits: Part A

3.4 Shrinkage Methods Prostate Cancer Data Example (Continued) Ridge Regression

Checking the resolved resonance region in EXFOR database

Computational modeling techniques

CHAPTER 3 INEQUALITIES. Copyright -The Institute of Chartered Accountants of India

Accreditation Information

COMP 551 Applied Machine Learning Lecture 9: Support Vector Machines (cont d)

(1.1) V which contains charges. If a charge density ρ, is defined as the limit of the ratio of the charge contained. 0, and if a force density f

Multiple Source Multiple. using Network Coding

Dead-beat controller design

Section I5: Feedback in Operational Amplifiers

5.4 Measurement Sampling Rates for Daily Maximum and Minimum Temperatures

THREE DIMENSIONAL SPACE-TIME Lu Shan No.1144, East of Jiuzhou Avenue, Zhuhai , Guangdong Province P. R. China

Lead/Lag Compensator Frequency Domain Properties and Design Methods

Eric Klein and Ning Sa

Cambridge Assessment International Education Cambridge Ordinary Level. Published

Chapter 4. Unsteady State Conduction

Image Processing 1 (IP1) Bildverarbeitung 1

PSU GISPOPSCI June 2011 Ordinary Least Squares & Spatial Linear Regression in GeoDa

Transcription:

M.E. Aggune, M.J. Dambrg, M.A. El-Sharkawi, R.J. Marks II and L.E. Atlas, "Dynamic and static security assessment f pwer systems using artificial neural netwrks", Prceedings f the NSF Wrkshp n Applicatins f Artificial Neural Netwrk Methdlgy in Pwer Systems Engineering, April 8-10, 1990, Clemsn University, pp.26-30. DYNAMIC AND STATIC SECURITY ASSESSMENT OF POWER SYSTEMS USING MTIFICIAL NEURAL NEmORXS H.E. Aggune, H.J. Dambrg, L A. El-Sharkawi, R.J. Harks XI, L.E. Atlas Department f Electrical ~ngineering university f Washingtn Seattle, WA 98195 What fllws is a summary f sme f the authrsf wrk n tw types f security prblems. In this wrk we used a layered perceptrn with tw training algrithms, a prjectin algrithm and aa errr backprpagatin algrithm. The prjectin algrithm is based n the least squares apprximatin technique and the errr backprpagatin n the steepest descent technique. 1. Dynamic Security Preliminary results were btained frm studies f the dynamic security f a 9 buses, 3 generatrs, and 14 transmissin lines test system. Our first study analyzed the relatinship between system stability and the utput pwer f generatr 3, the excitatin f generatr 3, the apparent pwer f lad 8, and the availability f lines 9 and 10. The results were reprted in [I]. Sme f these results are shwn in Figures 1 and 2. Basically, the accuracy f classificatin and the ability f the ANN t generalize were very gd. Hwever, we encuntered difficulties in extending the results t larger systems due t prp'erties f the prjectin algrithm. First, the number f hidden ndes must be at least as large as the number f training data pints. Secnd, as the number f hidden ndes grew, the prjectin algrithm became unstable due t the prblem f inverting an ill-cnditined matrix Our secnd examinatin f the dynamic security f this system used the errr backprpagatin training algrithm [2]. While this apprach requires fewer ndes in the ANN, the training time is lnger because it is iterative. This secnd study examined the security f the system with respect t the utputs P3 and 43 f generatr 3, the apparent pwer utput S2 f generatr 2, and the status f line LO. he ANN used had 4 input ndes, three representing

the quantities P3, Q3, and S2 and 1 fr a cnstant bias input. The hidden layer had 10 ndes and the utput had ne nde. f training.sets were used, each f 1000 pints m allwed dmain in the three dimensinal P3 x 43 x 52 space. One set was chsen at randm ver this dmain using -a unifrm distributin. The ther training set was selected t emphasize the security regin bundaries. Each training set was used separately t train an ANN. In bth cases, cnvergence required a few hundred iteratins thrugh the set as shwn in Figure 3. T assess the classificatin accuracy f each trained ANN, we perfrmed an exhaustive query and Figure 4 shws the perfrmance curves. Fr-bth cases the classificatin accuracy is seen t be very gd. The additinal accuracy that appears t be achieved by the bundary enhancing training set is prbably nt.imprtant. Rather, the significance f this result is that the ANN is able t give a gd representatin f the security bundary fr a cmplex relatinship, infrmatin that can help the dispatcher perate near the bundary with mre'cnfidence than is available at present. 2. Static security ~hisysectin prvides a brief review f the wrk reprted in [3].,The test system.was cmpsed f 8 buses, 4 generatrs, and 14 transmissin lines. The gal was t represent the security relatinship invlving the lads at buses 6 and 8 and transmissin line number 4. The ANN chsen had 3 input neurns, ne fr each f P6, Q6 and Sai- and ne cnstant value, r bias input neurn. There was ne hidden layer with 10 neurns and ne utput neurn. With-.line 4 perating, the netwrk was trained t mnitr the values f P6, 46 and S8 and t indicate when cnstraints wuld-be vilated if line 4 were t fail. That is, line 4 was the nly element in the cntingency list. Traieng was accmplished with the errr backprpagatin algrithm and tw different training sets. The first set was a tw-dimensinal case where S8 was fixed at 100% f its nminal value. Tw thusand training pints were selected randmly in the P6, 46 plane. Figures 5 and 6 indicates the relatinship between the true and predicted security regin. Abaut 2% f the 6561 pints ver which the ANN was tested were misclassified. The false secure and false insecure indicatins were abut equal in number. d training set explred all 3 input variables. The S8 was allwed t have the discrete values f O%, 0% f nminal 10 and P6 and 46 were again

selected randmly fr each f these S8 values. A ttal f 1469 training pints was used. When tested, the ANN was fund t generalize smthly between the fixed values f S8 used in training as shwn in Figures 7 and 8. In bth cases, the perfrmance f the ANN was judged very gd fr the cmplex relatinships being classified. 3, The Current Challenge: slving Full Scale Prblems The basic challenge we see at present fr develping useful tls is ne f scale. Pwer systems are typically cnsidered "large scale systems. Hence, an ANN apprach must ultimately accmmdate this prblem f scale in sme manner, The prblems that we think will need t be faced are t determine : hw large an ANN is required and what its architecture must be, hw much data is required fr training and if that training can be accmplished in reasnable time, whether the training data (prbably frm ff-line studies) can be generated with reasnable effrt, a methd fr testing the trained ANN t measure its classificatin accuracy and develp the cnfidance f the dispatcher, if warranted, and hw t update the ANN s it cntinues t be current. 4. References [I] M. E. Aggune, et al., "Preliminary Results n Using Artificial Neural Netwrks fr Security Assessment.ll PICA Cnference, pp. 252-258, May 1989. [2] M. A. El-Sharkawi, et al., "Dynamic Security Assessment f Pwer Systems Using Back Errr Prpagatin Artificial Neural Netwrks," Secnd Smpsium n Expert Systems ~pllicatin t Pwer Svstems, pp. 366-370, July 1989. [3] M. E, Aggune, et al., "Artificial Neural Netwrks fr Pwer System static security Assessment." IEEE Prceedinas f the Internatinal Svm~sium n Circuits and Systems, Vl. 1, pp. 490-494, May 1989,

'1.............I....... Y... '... '.. a....... a........... at"""'.." " " U.....a....