Machine Learning Applications for Load, Price and

Wind Power Prediction in Power Systems

Michael Negnevitsky, Senior Member, IEEE, Paras Mandal, Member, IEEE, and Anurag K. Srivastava, Member,
IEEE

Abstract— This paper reviews main forecasting techniques describe very short-term wind power prediction utilizing a
used for power system applications. Available forecasting tech- case study from Tasmania, Australia. Wind power presently
niques have been discussed with focus on electricity load and is the fastest growing power generation sector in the world.
price forecasting as well as wind power prediction. Forecasting
problems have been classified based on time frame, application However, wind power is intermittent. To be able to trade
specific area and forecasting techniques. Appropriate examples efficiently, make the best use of transmission line capability,
based on data pertaining to the Victorian electricity market, and address concerns with system frequency in a re-regulated
Australia and the PJM electricity market, U.S.A. are used system, accurate wind power forecasting in very short-term are
to demonstrate the functioning of the developed neural net- essential. This paper discusses the application of an adaptive
work (NN) method based on similar days approach to predict
hourly electricity load and price, respectively. The other impor- neuro-fuzzy inference system (ANFIS) to forecasting a wind
tant problem faced by power system utilities are the variability time series [4].
and non-schedulable nature of wind farm power generation. The objective of this paper is to provide a brief overview of
These inherent characteristics of wind power have both technical advancement in machine learning applications for forecasting
and commercial implications for efficient planning and operation in power systems with focus on short-term electric load
of power systems. To address the wind power issues, this paper
presents the application of an Adaptive Neural Fuzzy Inference and price forecasting as well as wind power prediction. The
System (ANFIS) to very short-term wind forecasting utilizing a problems and various available techniques in these areas of
case study from Tasmania, Australia. forecasting are discussed.
Index Terms— Adaptive neuro-fuzzy inference system (AN- The rest of the paper is organized as follows. Section II
FIS), neural network (NN), short-term load forecasting, short- provides classification of forecasting problems and methods
term price forecasting, very short-term wind power prediction. associated with wind power, electric load and energy price
based on time frame and applications. Case study results of
short-term load and price forecasting, and very short-term
I. I NTRODUCTION wind power prediction are presented in Section III. Section IV
presents the conclusion of this paper.

F ORECASTING is a vital part of business planning in

today’s competitive environment. With increased penetra-
tion of renewable energy sources and introduction of deregula-
II. C LASSIFICATION OF FORECASTING PROBLEMS AND
TECHNIQUES
tion in power industry, many challenges have been encountered
In this section, we discuss about different forecasting prob-
by the participants of the electricity market. Forecasting of
lems and techniques in the area of electric load forecasting,
wind power, electric loads and energy price have become a
energy price forecasting and wind power prediction.
major issue in power systems. Following needs of the market,
various techniques are used to forecast the wind power, energy A. Load forecasting
price and power demand.
Load forecasting is a process to predict load for a future
This paper presents machine learning applications to short-
period. Application of load forecasting falls into different
term load forecasting utilizing a case study of the Victorian
time horizons: long-term forecasting (from one year to ten
electricity market, Australia to demonstrate the functioning of
years), medium-term forecasting (from several months to one-
the developed neural network (NN) model based on similar
year), short-term forecasting (from one-hour to one-week)
days (SD) method [1]. Load forecasting results obtained from
and real-time or very short-term forecasting (in minutes).
Okinawa Electric Power Co., Japan is also reported where
Long-term forecasts influence the decisions on generation
combination of fuzzy and NN approach is proposed [2].
and transmission planning, which is used for determining the
Integration of NN and SD approach has also been applied
economical location, type, and size of the future power plants.
to predict day-ahead electricity price in the PJM electricity
Medium-term load forecasts are necessary for generation and
market [3]. In addition, this paper focuses on machine learning
transmission maintenance, and also for fuel scheduling. Accu-
application to wind power prediction in which the authors
rate short-term load forecasts are necessary for unit commit-
M. Negnevitsky and P. Mandal are with the Centre for Renewable Energy ment an economic dispatch. Very short-term load forecasting
and Power Systems, University of Tasmania, Hobart, Tasmania 7001, Australia is for minutes ahead and is used for automatic generation
(e-mail: Michael.Negnevitsky@utas.edu.au, Paras.Mandal@utas.edu.au). control (AGC) [2],[5],[6].
A. K. Srivastava is with the Department of Electrical and Computer
Engineering, Mississippi State University, Mississippi State, MS 39762 USA Good confidence level has been achieved in load forecasting
(e-mail: srivastava@ece.msstate.edu). by researchers. Load forecasting have been challenge in the

978-1-4244-5098-5/09/$26.00 ©2009 IEEE

past due to following reasons. First, because the load series is unexpected incidents such as transmission congestion, trans-
complex and exhibits several levels of seasonality: the load at mission contingency and generation contingencies. It can also
a given hour is dependent not only on the load at the previous be influenced by other factors such as fuel prices, generation
hour, but also on the load at the same hour on the previous unit operation costs, weather conditions, and probably the most
day, and on the load at the same hour on the day with the theoretically significant factor, the balance between overall
same denomination in the previous week. Secondly, because system supply and demand [3],[6],[9].
there are many important exogenous variables that must be Applications of electricity price forecasting fall into differ-
considered, specially weather related variables [2],[6]. ent time horizons: short-term forecasting, medium-term fore-
Most forecasting models and methods have already been casting and long-term forecasting. Market participants need to
tried out on load forecasting, with varying degrees of success. forecast short-term (mainly one day-ahead) prices to maximize
Some of the models reported in literatures are multiplicative their profits in spot markets. Accurate medium term price
autoregressive models, dynamic linear or nonlinear models, forecasts are necessary for successful negotiations of bilateral
autoregressive models, and methods based on Kalman filter- contracts between suppliers and consumer. Long-term price
ing, Box and Jenkins transfer functions ARMAX models, forecasts influence the decisions on transmission expansion
optimization techniques, nonparametric regression. Several and enhancement, generation augmentation, distribution plan-
research works have been carried out on the application of ning and regional energy exchange [6].
artificial intelligence (AI) techniques, also termed as machine Electricity price forecasting models include statistical and
learning techniques, to the load forecasting problem. Various non-statistical models. Time-series models, econometric mod-
machine learning techniques reported in literatures are expert els and intelligent system methods are the three main cat-
systems, fuzzy inference, fuzzy-neural models, artificial neural egories of statistical methods. Nogales et al. [10] proposed
network (ANN), data mining, etc. Wu and Lu [7] proposed a time-series model for forecasting next-day electricity prices.
machine learning technique by using data mining approach to The time-series techniques are successful in the areas where
forecast load. Fan et al. [8] discussed on integration of two the frequency of the data is low, such as weekly patterns,
machine learning techniques: Bayesian clustering by dynam- but they can be problematic when there are rapid variations
ics (BCD) and support vector regression (SVR) for day-ahead and high-frequency changes of the target signal. Non-statistical
electricity load forecasting. Among the different techniques methods include equilibrium analysis and simulation methods.
on load forecasting, application of NN technology for load Methods based on machine learning techniques, especially the
forecasting in power system has received much attention in NNs [3],[9],[11],[12] are more common for electricity price
recent years. The main reason of NN becoming so popular lies forecasting due to their capability of learning complex and
in its ability to learn complex and nonlinear relationships that non-linear relationships that are difficult to model with conven-
are difficult to model with conventional techniques [2],[5],[6]. tional techniques. NN approach combined with similar days
method has been discussed by Mandal et al. [3] to forecast
day-ahead prices in the PJM market. Amjadi and Keynia [9]
B. Price forecasting
proposed a machine learning technique by using mutual infor-
The main objective of electricity market is to maximize mation technique and cascaded neuro-evolutionary algorithm
profits. Forecasting loads and prices in electricity markets are for day-ahead price forecasting. Other machine learning tech-
mutually intertwined activities, and error in load forecasting nique such as hybrid network of self-organised map (SOM)
will propagate to price forecasting. Electricity price has its and support-vector machine (SVM) has been proposed by Fan
special characteristics. The main features that make it so et al. [13] to forecast next-day hourly prices.
specific are at least three. One of them is its non-storability of
power, which implies that prices are strongly dependent on the C. Wind power prediction
power demand. Another characteristic is the seasonal behavior One of the fundamental problems faced by power system
of the electricity price at different level (daily, weekly and operators is the variability and non-schedulable nature of wind
annual seasonality) and the third one is related to the question farm power generation. These inherent characteristics of wind
of its transportability [3],[6]. Furthermore, electricity price can power have both technical and commercial implications for
rise by tens of or even hundreds of times of its normal value efficient planning and operation of power systems. Wind power
showing one of the greatest volatilities among all commodities. prediction systems provide the information how much wind
Electricity cannot be transported from one region to another power can be expected at which point of time in the next
one because of existing bottle-necks or limited transportation few days. Wind power forecasting is one of the most critical
capacity. Application of forecasting methods common in other aspects in wind power integration and operation. It is needed to
commodity markets, can have a large error in forecasting the estimate the long, medium, and short-term power production.
price of electricity [6]. The long-term forecast is required during the planning stage,
In most competitive electricity markets, the hourly price while the medium and short-term forecasts are needed in the
series have the characteristics such as volatility, non-stationary generation commitment and market operation. Long-term wind
properties, multiple seasonality, spikes and high frequency. power forecasting is based on long term wind patterns, while
These characteristics are due to events that may occur al- medium and short-term forecasts are generally for a few days
ternatively in a market. For instance, a price spike that is a (depends on the market operation, generally between one and
randomized event can be caused by market power, and also by three days), and hours to a few minutes, respectively [4].
 When considering the problem of wind power forecasting, accurate technique for short-term forecasting. However, in
a number of factors need to be considered. The first consid- general, statistical, NN methods, or several advanced hybrid
eration is the prediction horizon or forecasting time frame methods based on observations perform more accurately over
of interest. The broad timescales can be considered as [14]: the very short-term forecast range.
(i) Turbulence (1 second−30 minutes); (ii) Synoptic Scale in In general, forecasting techniques for very short-term wind
the Spectral Gap (30 minutes−6 hours); (iii) Synoptic Scale (6 power prediction use recent historical data as inputs to suit-
hours−7 days); and (iv) Climatic Scale (months to years). A ably structured models. These techniques include the simple
wind forecasting time scale of less than 30 minutes is referred persistence approach, classical linear statistical models such
to as the turbulence time scale or more generally as the very as Moving Average (MA), Auto-Regressive Moving Aver-
short-term time scale. age (ARMA) and the Box-Jenkins approach based on Auto-
Various studies have been conducted on power fluctuations Regressive Integrated Moving Average (ARIMA) or seasonally
from large wind farms and their effects on power system adjusted ARIMA models, also known as SARIMA mod-
operations. Large ramp variations in the very short-term time els [4],[15].
scale can have significant impacts on both power system The statistical time series and NN methods are mostly
security and (depending on the bidding and dispatch inter- aimed at short-term predictions. Typical time series models
vals involved) on trading in deregulated electricity markets. are developed based on historical values. They are easy to
The importance of achieving the best possible wind power model and capable to provide timely prediction. In several
forecasting accuracy in the very short-time scale is thus of predictions, they use the difference between the predicted
paramount importance during significant large ramp events or and actual wind speeds in the immediate past to tune the
periods of high volatility in wind farm power output [14]. model parameters. The advantage of the NN is to learn the
In addition, wind power − electric power generated from relationship between inputs and outputs by a non-statistical
wind kinetic energy − possesses unique characteristics and approach.
attributes that differentiate it from fossil-based electric power.
Unlike fossil- based power, where the rate of energy through- III. C ASE S TUDY R ESULTS
put is completely controllable, wind power is intermittent, The case study results presented in this paper mainly
variable, and non-dispatchable. Indeed, wind generation could focus on application of machine learning techniques for short-
vary according to diurnal heating and cooling patterns or term load and price forecasting as well as very short-term
suddenly increase with a storm front. Without an energy wind power prediction. The results are based on the authors’
storage system, wind energy being converted into electric published works.
power has to be consumed immediately. As a result, the
economic value of wind generation is dependent on the relative A. Load forecasting results
synchronized timing of the wind and load patterns. During the The previous works done by authors in the area of short-
on-peak period of the day, the production of wind generation term load forecasting are found in [1], [2] where the authors
commands a high value. While during off-peak periods, wind have proposed machine learning technique by using NN com-
generation may provide very little value or could even be cur- bined with SD method. SD method adopts the information of
tailed when there is no load to serve. Furthermore, wind power the days being similar to that of the forecast day. Accord-
generation does not lend itself readily to participate in the ing to similar days method, selection of similar load days
traditional generation scheduling process where controllable corresponding to forecast day is performed by adopting the
generators are scheduled to meet a variable load. Generation Euclidean norm equation [2]. Euclidean norm with weighted
scheduling process with wind power must take into account the factors is used to evaluate the similarity between a forecast
variability and intermittency characteristics. Therefore, certain day and searched previous days. The smaller the Euclidean,
power system resources must be allocated separately to hedge the better the evaluation of similar days.
against unavailability and variability of wind power during a Short-term load forecasting was carried out based on dif-
low or no wind condition. These additional reserves increase ferent time frames, e.g., reference [1] deals with several-
the overall generation costs [14],[15]. hour-ahead electric load forecasting in which authors have
Several techniques have been identified for wind forecast- presented one-to-six hour ahead load forecasting using NNs
ing. These techniques can be categorized into numeric weather using the case study of Victorian electricity market. The actual
prediction (NWP) methods, statistical methods, methods based and forecast load curves for six-hour-ahead load forecasting
upon NNs, and hybrid approaches. Statistical analysis of wind are presented in Fig. 1 where we can observe an improved
power forecast error was discussed in reference [16] where the accuracy in forecasting as the mean absolute percentage error
authors showed the usefulness of the forecast error probability (MAPE) by using NN (1.30%) is better compared to SD
density function (pdf) for finding the optimum rated energy approach (1.74%). It is also to be noted that one to six hour
storage system (ESS) power. Reference [17] discussed NWPs ahead load forecast errors (MAPE) range from 0.56% to 1.30%
estimator models based on machine learning technique using in reference [1].
fuzzy logic and wind power forecasting models using NNs Another work carried out by the authors on next-day load
combination. NWP based techniques are well established curve forecasting was presented in reference [2] where the
for wind parameter forecasting with a prediction horizon of authors have proposed load forecasting using hybrid correction
several hours or more. NWP methods could be the most method, which is a combination of NN and fuzzy logic using
 80

MAPE: 6.93 % (ANN) Forecasted (ANN):

13.90 % (SD) Actual:
70 Forecasted (SD):

LMP ($/MWh)
60

50

40

30

04:00 08:00 12:00 16:00 20:00 24:00

Fig. 1. Six hour ahead load forecasting (Monday, September 01−Sunday,

Hour
September 07, 2003). Fig. 3. Actual and forecast day-ahead PJM electricity price (Friday, January
1100 20, 2006).
Load [MW]

1000
100
MAPE: 7.66% (ANN) Forecasted
900 90 12.80% (SD) Actual

800 80

LMP ($/MWh)
700 70
: 1. Actual
600 : 2. Similar days 60
: 3. NN
500 : 4. NN+Fuzzy 50

400 40
1 3 6 9 12 15 18 21 24
Hour 30
(a)
20
24 48 72 96 120 144 168
Forecasted Error [%]

20
: 1. Similar days
Hour
15 (4.70%)
Fig. 4. Actual and forecast day-ahead PJM electricity price during low
: 2. NN
(2.24%)
demand week (Wednesday, February 1 to Tuesday, February 7, 2006).
10 : 3. NN+Fuzzy
(1.71%)
5 using NN model integrated with SD method [3]. In the
developed price prediction method, day-ahead electricity price
0
is obtained from the neural network that modifies the price
-5
curves obtained by averaging selected number of similar price
days corresponding to forecast day, i.e., two procedures were
-10 analyzed: 1) prediction based on averaging prices of similar
1 3 6 9 12 15 18 21 24 days and 2) prediction based on averaging prices of similar
Hour
(b) days plus neural network refinement. The developed multi-
Fig. 2. Forecasting results (Monday, July 21, 1997: Substitute holiday). layer feed-forward NN model was trained and tested using the
(a) Forecasted load curves [MW], (b) forecast error [%]. data (from January 1, 2004 to May 31, 2006) derived from the
PJM electricity market. Sets of data include hourly price and
the case study of Okinawa Electric Power Co., Japan. Refer- load data.
ence [2] contributes to machine learning application to day- Figure 3 shows the day-ahead price forecasting results for
ahead load forecast methodology, especially as it shows how January 20, 2006. The forecast results obtained from the
to reduce neural network forecast error over the test period by proposed NN are quite close to the actual LMP values. The
23% through the application of fuzzy logic correction, which prediction behavior of the proposed NN technique for this day
can be observed in Fig. 2(a) and (b). is very appropriate with a daily MAPE of only 6.93%, which
is much lower than that obtained using SD approach (13.90%).
Figure 4 shows the weekly price forecasts for a typical low
B. Price forecasting results demand week. These forecasts are based on day-ahead and
The price forecasting result presented in this paper is based have been represented for a week. As it can be seen from Fig. 4
on the authors previous work on machine learning application that the forecast results obtained from the proposed NN are
to forecast day-ahead electricity price in the PJM market close to the actual LMP values. Also, it can be observed that
when price spikes appear, our model does not forecast price
jumps as in last three days of this week. The weekly MAPE
obtained from the NN approach is 7.66%, which is much lower
than that obtained using the similar days method (12.80%).
Daily and weekly MAPE values obtained from the selected
day and week show that the NN technique outperforms the
direct use of similar days method.

C. Wind power prediction results

The case study presented in [4] used a wind site in Tasmania
as the data set, providing a 21-month time series in steps
of 2.5 min. This was to be the forecast period as well. The
application of a particular type of hybrid intelligent system
called an Adaptive Neuro-Fuzzy Inference System (ANFIS)
to very short term wind power prediction was investigated. Fig. 5. Chart of prediction errors for various systems, tested over a period
In order to optimize the performance of the ANFIS system, of eight months on a wind site in Tasmania, Australia.
multiple architectures should be evaluated using the same
data set [4]. A persistence model was also developed for
comparison. Persistence is presently an industry benchmark for
very short-term wind forecasting and so is the most indicative
assessment. The ANFIS model was developed in several
different formats. The ANFIS model design is flexible and
capable of handling rapidly fluctuating data patterns [15], [18].
This was to highlight the usefulness of intermediary splines
through data for very short-term forecasting. The results are
shown in Fig. 5. The chart also includes the results from the
persistence model. A useful comparison is available through
considering the persistence results and the ANFIS model with
no spline. The ANFIS model shows some improvement, in the
order of 5%.
In the case study as reported in [4], [15], a suitable ANFIS
model configuration was found to provide an overall perfor-
mance improvement of more than 8% over the industry stan-
Fig. 6. Snapshot of ANFIS model performance for 5 minute ahead wind
dard persistence approach for five minute ahead predictions. power prediction at a US wind farm.
Fig. 6 shows the snapshot of the ANFIS model performance
compared to persistence. Apart from the overall performance
improvement for the entire test period, it was observed that
R EFERENCES
the ANFIS model provided significantly better predictions than
persistence for certain periods in the time series when the wind [1] P. Mandal, T. Senjyu, and T. Funabashi,“Neural networks approach to
farm power output was rapidly increasing or decreasing. forecast several hour ahead electricity prices and loads in deregulated
market,” Energy Convers. and Manage., vol. 47, pp. 2128−2142, 2006.
[2] T. Senjyu, P. Mandal, K. Uezato, and T. Funabashi,“Next day load curve
IV. C ONCLUSIONS forecasting using hybrid correction method,” IEEE Trans. on Power Syst.,
vol. 20, no. 1, pp. 102−109, Feb. 2005.
This paper presented applications of machine learning tech- [3] P. Mandal, T. Senjyu, N. Urasaki, T. Funabashi, and A. K. Srivastava, “A
niques for forecasting short-term loads and prices as well novel approach to forecast electricity price for PJM using neural network
as very short-term wind power prediction in power systems. and similar days method,” IEEE Trans. on Power Syst., vol. 22, no. 4,
pp. 2058−2065, Nov. 2007.
A brief overview and general background of different ap- [4] C. W. Potter and M. Negnevitsky, “Very short-term wind forecasting for
proaches and developments were discussed. The paper also Tasmanian Power Generation,” IEEE Trans. on Power Syst., vol. 21,
underlined major challenges related to wind power prediction. no. 2, pp. 965−972, May 2006.
[5] H. S. Hippert, C. E. Pedreira, and R. C. Souza, “Neural networks for
The importance of forecasting loads and prices including short-term load forecasting: A review and evaluation,” IEEE Trans. on
their characteristics was discussed. The results of several case Power Syst., vol. 16, no. 1, pp. 44−55, Feb. 2001.
studies were presented and discussed. It was demonstrated that [6] M. Shahidehpour, H. Yamin, and Z. Li, Market operations in electric
machine learning techniques could be applied to forecasting power systems: forecasting, scheduling, and risk assessment. John Wiley
& Sons, Inc., New York, 2002.
problems in power systems. [7] H-C Wu and C-N Lu, “A data mining approach for spatial modeling in
small area load forecast,” IEEE Trans. on Power Syst., vol. 17, no. 2,
pp. 516−521, May 2002.
[8] S. Fan, L. Chen, and W-J Lee, “Machine learning based switching model
for electricity load forecasting,” Energy Convers. and Manage., vol. 49,
no. 2, pp. 1331−1344, June 2008.
[9] N. Amjadi and F. Keynia, “Day-ahead price forecasting of electricity mar- Anurag K. Srivastava (SM’01-M’05) received his Ph.D. degree from Illinois
kets by mutual information technique and cascaded neuro-evolutionary Institute of Technology (IIT), Chicago, in 2005, M. Tech. from Institute
algorithm,” IEEE Trans. on Power Syst., vol. 24, no. 1, pp. 306−318, of Technology, India in 1999 and B. Tech. in Electrical Engineering from
Feb. 2009. Harcourt Butler Technological Institute, India in 1997. He is working as an
[10] F. J. Nogales, J. Contreras, A. J. Conejo, and R. Espinola, “Forecasting Assistant research professor at Mississippi State University since September
next-day electricity prices by Time Series Model,” IEEE Trans. on Power 2005. Before that, he worked as a research assistant and teaching assistant at
Syst., vol. 17, no. 2, pp. 342−348, May 2002. IIT, Chicago, and as a Senior Research Associate in Electrical Engineering
[11] P. Mandal, A. K. Srivastava, M. Negnevitsky, and J.-W. Park, “An Department at the Indian Institute of Technology, Kanpur, India as well as a
effort to optimize similar days parameters for ANN based electricity Research Fellow at Asian Institute of Technology, Bangkok, Thailand.
forecasting,” IEEE Trans. on Ind. App. (in press). His research interest includes power system security, power system dereg-
[12] P. Mandal, A. K. Srivastava, M. Negnevitsky, and J.-W. Park, “Sensitivity ulation, power system modeling, and artificial intelligent application in power
analysis of neural network parameters to improve the performance of system. Dr. Srivastava is a member of IEEE, IET, IEEE Power and Energy
electricity price forecasting,” Int. J. of Energy Research, vol. 33, no. 1, Society, ASEE, Sigma Xi and Eta Kappa Nu. He is a recipient of several
pp. 38−51, Jan. 2009. awards and serves as a reviewer for IEEE Transactions on Power Systems,
[13] S. Fan, C. Mao, and L. Chen, “Next-day electricity-price forecasting international journals and conferences.
using a hybrid network,” Proc. Inst. Eng. Tech., Gen., Transm., Distrib.,
vol. 1, no. 1, pp. 176−182, Jan. 2007.
[14] M. Negnevitsky and P. Johnson, “Very short term wind power prediction:
A data mining approach,” in Proc. IEEE Power and Energy Soc. General
Meeting, pp. 1−3, Pittsburgh, PA, USA, 2008.
[15] P. Johnson, M. Negnevitsky, and K. M. Muttaqi, “Short-term wind
power forecasting using adaptive neuro-fuzzy inference systems,” in Proc.
Australian Universities Power Engineering Conference (AUPEC), Perth,
Western Australia, 2007.
[16] H. Bludszuweit, J. A. Dominguez-Navarro, and A. Llombart, “Statistical
analysis of wind power forecast error,” IEEE Trans. on Power Syst.,
vol. 23, no. 3, pp. 983−991, Aug. 2008.
[17] G. Sideratos and N. Hatziargyriou, “Using radial basis neural networks
to estimate wind power production,” in Proc. IEEE Power and Energy
Soc. General Meeting, pp. 1−7, Tampa, FL, USA, 2007.
[18] M. Negnevitsky, Artificial Intelligence: A Guide to Intelligent Systems.
Second Edition, Addison Wesley, 2005.

V. B IOGRAPHIES

Michael Negnevitsky (M’95-SM’07) received the B.S.E.E. (Hons.) and Ph.D.

degrees from the Byelorussian University of Technology, Minsk, Belarus,
in 1978 and 1983, respectively. Currently, he is Chair Professor in Power
Engineering and Computational Intelligence and Director of the Centre for
Renewable Energy and Power Systems at the University of Tasmania, Hobart,
Australia. From 1984 to 1991, he was a Senior Research Fellow and Senior
Lecturer in the Department of Electrical Engineering, Byelorussian University
of Technology. After arriving in Australia, he was with Monash University,
Melbourne, Australia.
His interests are power system analysis, power quality, and intelligent
systems applications in power systems. Dr. Negnevitsky is a Chartered
Professional Engineer, Fellow of the Institution of Engineers Australia. He is
also Member of CIGRE AP C4 (System Technical Performance) and CIGRE
AP C6 (Distribution Systems and Dispersed Generation), Australian Technical
Committees, and CIGRE Working Group JWG C1/C2/C6.18 (Coping with
Limits for Very High Penetrations of Renewable Energy), International
Technical Committee.

Paras Mandal (SM’05-M’06) received his B.E. degree in Electrical and

Electronics Engineering from Kuvempu University (now under Visveswariah
Technological University), India in 1998; M.E. degree in the field of Energy,
Economics and Planning from Asian Institute of Technology, Thailand in
2002; and Ph.D. degree from the University of the Ryukyus, Japan in 2005.
He is presently a research fellow at the Centre for Renewable Energy and
Power Systems, University of Tasmania, Hobart, Australia. From Nov. 2005
to July 2007, he was a JSPS postdoctoral fellow at the University of the
Ryukyus. Then, he worked in the School of Engineering, Yonsei University,
Seoul, South Korea as a Research Professor till January 2008.
His research interests include application of artificial intelligence tech-
niques to electricity load and price forecasting, power system modeling and
power system deregulation. Dr. Mandal is a member of IEEE Power and
Energy Society. Dr Mandal is a recipient of several awards and serves as a
reviewer for IEEE Transactions on Power Systems, international journals and
conferences.