PRESENTATION ON

LOAD FORECASTING USING HYBRID CONTROLLERS

Submitted by: Pradip Niranjan M. Tech. 6th Semester Roll No.: 6004520004
S.R. No.: 809/10

Under the Guidance of: Mr. C. N. Singh Associate professor

DEPARTMENT OF ELECTRICAL ENGINEERING H.B.T.I. KANPUR

Motivation of Study
Load forecasting is enable to take utility unit commitment

decision. The precise forecasting is the basis of electrical energy trade and spot price establishment for the system to gain the minimum electricity purchasing cost. In the real-time dispatch operation, forecasting error causes more purchasing electricity cost or breaking-contract penalty cost to keep the electricity supply and consumption balance. Contd..

Motivation of Study
STLF is Carried out based on historical data and different mathematical techniques are used each each may have its own accuracy at a particular time point.
It is necessary to use the combination of different methods and computational i.e hybrid intelligence algorithms to result forecast with high accuracy.

Objective
The objective of project work is to develop new practical models with computational intelligence algorithm which have high in load for cast The research tries to bring forth the advantages of different computational intelligence methods and develop a comprehensive method of selection to fulfill the goal

Artificial Neural Network(ANN)

A neural network is a massively parallel distributed processor made up of

simple processing units, which has a natural propensity for storing experimental knowledge and making it available for use. ANN acquire the information from environment through a learning process. Neural networks having remarkable ability to derive meaning from complicated or imprecise data. A neural network is a machine that is designed to model the way in which the brain perform a particular task. A neural network has remarkable ability to derive meaning from complicated or imprecise data and extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques.

Mathematical Model of ANN

X1,x2,xp are input from environment to neurons in input Layer .

Wk1,wk2,...wkp synaptic weights

Model of an ANN

Network Architectures
Single-layer Feed Multilayer Feed

forward Networks

forward Networks

Learning Process
learning is process to modify the synaptic weights of network in an orderly fashion to attain a desired design objective.
Supervised learning
Learning is accomplished under

Unsupervised learning
There are no target outputs

supervision As the inputs are applied to the network, the network outputs are compared to the targets. The learning rule is used to adjust the weights and biases of the network in order to move the network outputs closer to the targets.

available The weights and biases are modified in response to network inputs only

Introduction to load forecasting

Load forecasting phenomenon of estimating

measures in the coming future . Electrical load forecasting is concern with estimation of maximum demand. Electrical load forecasting is carried out with historical load data and whether conditions i.e temperature , humidity and wind speed.

Need of load forecasting

Ensures secure operation of electrical utility grid. Play important role in generation scheduling ,unit

commitment Hydrothermal coordination. To improve reliability of A.C power line network and optimal load scheduling .

Electrical load forecasting techniques

Multiple regression
Exponential smoothing Iterative reweighted least-squares

Adaptive load forecasting

Stochastic time series Neural network Fuzzy logic

Multiple regression
Multiple regression analysis for load forecasting uses the technique of

weighted least-square s estimation. Based on this analysis, the statistical relationship between total load and weather conditions as well as the day type influences can be calculated. The model used in multiple regression is given as Y(t)=v(t)a(t) +e(t) Where, t= Sampling time Y(t)=Measured system load V(t)= Vector adaptive variables time temperature etc A(t)=Transposed vector of regression coefficient e(t)= Model error at time t . The regression coefficient is computed by equally or exponentially least square estimation

Exponential Smoothing
One of classical method used for load forecasting. This approach is used to model the load this model is

used to estimate the future load . Y(t)=(t)T f(t) + e(t) Where, Y(t)= Load at time t sec. f(t)= fitting function vector of the process (t)= coefficient vector e(t)= White noise T= Transpose operator

Iterative reweighted least square

The method uses an operator that controls one variable at a time. An optimal starting point is determined using the operator This method utilizes the autocorrelation function and the partial

autocorrelation function of the resulting differenced past load data in identifying a sub optimal model of the load dynamics. The weighting function, the tuning constants and the weighted sum of the squared residuals form a three-way decision variable in identifying an optimal model and the subsequent parameter estimates.
Consider the parameter estimation problem involving linear

measurement equation Y=X +

Y(nx1 )= vector of observation : (px1)=matrix of un known parameters X(nxp)= matrix of known coefficient :(nx1)= Vector of random error

Adaptive load forecasting

In adaptive load forecasting the model parameter are

automatically corrected to keep track of the changing load conditions Regression analysis based on Kalman filter theory is used. The kalman filter normally uses the current prediction error and the current weather data acquisition programs to estimate the next state vector . The state vector is determined by analyzing the total historical data not only recent measured load and weather data.

Stochastic time series

Using the time-series approach, a model is first

developed based on the previous data, then future load is predicted based on this model. Some of time series model is listed below(i) auto regressive model (ii) Auto regressive moving average model (iii) Auto regressive integrated moving average model

ARMAX Model based on genetic algorithm

The genetic algorithm (GA) approach is used to identify the

autoregressive moving average with exogenous variable (ARMAX) for load demand forecast. This algorithm offers the capability of converging towards the global extremum of a complex error surface. The model can be given by: A(t)Y(t)=B(t)U(t)+C(q)e(t): Where Y(t)=Load at time t U(t)=exogenous temperature input at time t e(t)=whit noise at time t and q= Back shift operator and A(q),B(q) and C(q) are parameters of auto regressive(AR),exogenous(X),and moving average(MA) parts respectively.

ANNs controller
ANNs needs selection of variable as network input.
Develop the relationship between input variables and

predicted load based on training process. Back propagation learning algorithm is used to train ANNs forecasting time series.

Selection of parameters for ANN

STLF requires the following parameter to be selected as input to ANN
Whether Time Economy Random disturbance Class of customers

ANN Controllercontd

Fuzzy Controller
This approach is applied to forecasting where the

advantage of an operators expert knowledge is used. The fuzzy decision system require detailed analysis of data and fuzzy rule base has to be developed heuristically for each session The problem with fuzzy system is that the rules fixed in fuzzy rule base may not always yield the best forecast.

Need of hybrid controller

Reliance on large historical database with possible

obsolete and irrelevant data, assumptions about static load shape and parameters. Neural network is one approach which is able to deal with nonlinear and adaption.

Hybrid Controller
The Hybrid controller consist of artificial neural

networks (ANNs) and fuzzy network . The functionality of fuzzy system and learning capabilities of neural network can be merged to yield a forecasting system more power then either of system alone. Hybrid System has fuzzy neural network(FNN) used in pre forecasting and fuzzy expert system used in final forecaster .

Fuzzy neural network (FNN) for initial forecast

Fuzzy neural network model is based on multilayer perception

using gradient descent based on back propagation algorithm by incorporating fuzzy sets at various stage . Fuzzy sets are defined for load , temperature, humidity parameter. The input to FNN is comprise the membership value to overlapping partition of linguistic property of small, medium and large corresponding to each input feature past load temperature , humidity ..etc
The Out put layer consist of

the membership value to the overlapping partition linguistic properties small medium and large corresponding to load forecasted.

Fuzzy Neural network (FNN)

FNN is based on the multilayer perception using

back propagation algorithm . The fuzzified input vector consist of the member ship values of past load and weather parameters i.e temperature, humidity and wind speed. The output vector is defined in terms of fuzzy class membership values of the forecast load.

Fuzzy neural network(FNN)

Input pattern: P(i-1)= average load on (i-1)th day
min(i-1)=minimum temperature (i-1)th day

max(i-1)= maximum temperature (i-1)th day

min(i)= minimum temperature of ith day

max(i)= maximum temperature of ith day

Output pattern:
P(i) = average load on ith day

Fuzzy expert system (FES)

The input to fuzzy expert system consist of differences

in whether parameters between present and forecasted instant . The output of the FES gives the load correction which when added to the initial forecast, yields the final forecast. Fuzzy network is used to calculate minimum ,maximum mean and variance of the estimated load.

Fuzzy expert system for final forecast

Neural fuzzy network(NFN) produces initial forecast with a

set of initial load, temperature and humidity expressed in terms of fuggy membership value. Let initial forecasted load by (NFN) be P(i) The actual load be A(i) As we know FES works on deference between predicted and actual load P(i)= A(i)-P(i) Similarly temperature and humidity error is given= (i)- (i-1) H=H(i)-H(i-1)

Fuzzy expert system(FES)

The errors in the weather parameters and

load-correction values are fuzzified using six fuzzy sets such as SP(small positive), MP (medium positive), LP (largepositive), SN (small negative), NM (medium negative) and LN (large negative).

Fuzzy expert system

load correction

output sets have six member and use linear fuzzification principle for obtaining membership grade . the member ship value of load correction is given as

Where Cmax is slop of load error correction

Load variation curve of Delhi load dispatch center for (01 to 10 Jan 2013)

Whether variation curve of Delhi (from 07 to 13 Jan 2013)

Conclusion
Hybrid approaches show a general way of integrating neural

networks and other methodologies to form a more efficient model of forecasting. The hybrid neuro-fuzzy and neuro-symbolic approaches have been used to forecast loads with better accuracy than the conventional ANN approaches when used in a standalone mode. Neuro-fuzzy provides a general method for combining available numerical information and human linguistic information in a common framework. The Neuro-Genetic approach is also another potential application of hybrid approach which offers its optimization feature to be used for the convergence of ANN weights.

References
[1] S. H. Ling, Student Member, IEEE, Frank H. F. Leung, Senior Member, IEEE, H. K. Lam, Member, IEEE, and Peter K. S. Tam, Member, IEEE Short-Term Electric Load Forecasting Based on a Neural Fuzzy Network IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 50, NO. 6, DECEMBER 2003. [2] Changhao Xia, Jian Wang & keren McMenemy Short, medium and long term load forecasting model and virtual load forecaster based on radial basis function neural networks. Electrical Power and Energy Systems 32 (2010) 743750. [3] D K Chaturvedi, P S Satsangi and P K Kalra Fuzzified neural network approach for load forecasting. engineering intelligent systems vol 8 no 1 march 2001. [4] Gaddam.Mallesham A fine load forecasting using neural networks and fuggy neural networks Proceedings of the 5th WSEAS Int. Conf. on Power Systems and Electromagnetic Compatibility, Corfu, Greece, August 23-25, 2005 (pp224-229).

Contind..
[5] Santos, P.J., Martins, A.G., Pires, A.J., Martins, J. F., Mendes, R.V Shortterm load forecast using trend information and process reconstruction [6] Mrs. J. P. Rothe ,Dr. A. K. Wadhwani and Dr. S. Wadhwani Hybrid and integrated approach to short term load forecasting. International Journal of Engineering Science and Technology Vol. 2(12), 2010, 7127-7132 [7] Hesham K. Alfares and Mohammad Nazeeruddin Electric load forecasting: literature survey and classication of Methods. International Journal of Systems Science, 2002, volume 33, number 1, pages 23-34

Contind..
[8] MING-GUANG ZHANG Short-term load forcsting based on support vector machines regression.Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005. [09] C. F. Juang, J. Y. Lin, and C. T. Lin, Genetic reinforcement learning through symbiotic evolution for fuzzy controller design, IEEE Trans. Syst., Man, Cybern. B, vol. 30, no. 2, pp. 290302, Apr. 2000.
[10] G. X. Yao and Y. Liu, Evolutionary programming made faster, IEE Trans. Evol. Comput., vol. 3, pp. 82102, July 1999.

Thank You