SHORT TERM ELECTRIC LOAD FORCASTING USING ARTIFICIAL NEURAL NETWORK
AAMIR SIDDIQUI1, PROF NISHEET SONI2
1M.Tech (Power system), Department of Electrical Engineering, SRIT Jabalpur – India
2Assistant Professor, Department of Electrical Engineering, SRIT Jabalpur – India
Abstract:-This paper uses Neural Network Toolbox in Matlab for electric load forecasting. Artificial neural network is implemented for the purpose of accurate prediction of future load. The prediction model is trained by historical data from electric system utility. The method consists of graphical user interface form Matlab to easily solve the complex problem like electric load prediction. The output of model used is then compared to actual data for validation.
1. INTRODUCTION
Artificial intelligence is finding its application in various fields of engineering problems. Predicting the future load for an electric system utility is a similar kind of problem. Any electric system utility in the scenario of deregulated economy needs to accurately perform electric load forecasting for secure and economical operation of power system. There are three basic types of load forecasting classified according to the period of forecast: (1) short term load forecast for a period ranging from one hour to one week, (2) medium term load forecast for duration of one week to one year, (3) long term load forecast for period of one year to several years.There are various factors which affects the accurate load prediction like temperature and whether conditions. The model is build considering the above mentioned factors. Till now a number of techniques have been used for electric load prediction. These techniques are broadly classified as classic or modern. The most important methods are regression method, time series approach, Kalman filters, wavelet transform, particle swarm optimization, fourier series approach and support vector machines. In recent research artificial neural network has been used in wide applications related to load prediction. There are other methods called as hybrid methods are also used for modeling such as fuzzy logic and genetic algorithm. The artificial neural network,
fuzzy logic and genetic algorithm are intelligent method and able to learn and gain the changes in the circumstances.
These methods are called as artificial intelligence methods. Artificial neural network is based on training the network with past and current data as input and output respectively. Simulimk in Matlab is a graphical user interface which helps the designer to simulate the actual mathematical model with the use various components provided in Simulink library.
Short-term load prediction is used to provide utility company management with future information about electric load demand in order to assist them in running more economical and reliable day-to-day operations.The power load during the year followed the same the daily and weekly periods of electric load, which shows the daily and weekly cycles of human activities and behaviour patterns, with some cyclical and random changes. The impact of space cooling on the electric load is very obvious during the summer time.
When the temperature increases, the demand for electricity also increases.
On the other hand, during the winter time the reverse relationship between temperature and electric demand exists because of the need for space heating.
The system load is the sum of the entire consumer’s load at the same time. The objective of system STLF is to predict the future system load. A good understanding of the system characteristics helps to design reasonable prediction models and select
appropriate models operating in different situations. Various factors that influence the system load behaviour can be classified into the following major categories:
• Weather
• Time
• Economy
• Random disturbance
2. ELEMENTS OF ARTIFICIAL NEURAL NETWORK
An artificial neural network is chosen for prediction of electric load as it has the capability to approximate non- linear function present in load profile of electric system utilities. ANN refers to a class of models inspired by biological neuron system. It consists of a number of computing units working in parallel called as neurons.
A neuron (figure: 1) is a multiple input single output processing element consisting of a summation operation and activation function. Below is a brief description of the component.
Weighting functions (input links X1-Xn):
an adjustable representative of the input’s connection strength.
Summation function (∑): This component performs the weighted summation of the various inputs received by the neuron.
Figure: 1 Artificial Neuron.
Activation function: The output of the summation function is then taken by the transfer function. This transfer function transforms the summation output into a working output.
Transfer functions that may be used are:
(a) Hard limiter,
(b) Ramping function, and (c) Sigmoid functions.
Output function: Each neuron produces an output to many other neurons. In Figure1, it is represented by y.
Figure 2 shows the first layer of the three-layer feed forward network that is implemented in this paper.
Feedforward means that inputs are
progressed in a forward manner from layer to layer.
The number of input variables, number of hidden layers and the number of neurons in the hidden layers make the results change. That is why they need to be chosen very carefully.
Figure: 2, First layer of three layer feedforward network
However the training procedure of the neural network helps in choosing the appropriate network configuration.
Training is done to find the weights that minimize the error. The training stops when either the number of iteration has been reached or performance goal has been achieved.
The proposed ANN structure is shown in figure 3 below. There are four inputs given to the ANN in this case, however the number of inputs can be less or more according to the data available. For example we can add data related to type of weekday, whether it is a weekday or weekend. There is single output which is forecasted load.
Figure: 3, Artificial Neural Network Structure
3. PREDICTION PROCEDURE
In predicting of electric load we have used a three layer feedforward neural network, trained by using back propagation method of training. Table 1 shows the structure or topology used in the ANN model used in this paper.
As seen in the table, there are four numbers of input variables we are using
to forecast the profile of electric load in future. The data used in the training.
TABLE 1
DEFINITION OF ANN INPUTS AND OUTPUT
Inputs Description 1 Previous day load 2 Previous day temperature
3 Previous day Humidity
4 Time Outputs Description 1 Forecasted load A. Neural network model:
The below shown figure explains how the neural network process the given input and produce an output that is given to the network. As the network is given the input data which is pre-processed and the processed input is given to the network that is a black box trained by the historical data which gives the predicted value related to the input given. Output of the network is post processed and shown as a graph or table.
Figure: 4, Block diagram of Artificial Neural Network
B. Simulink toolbox in Matlab:
Matlab have a number of toolbox that can solve various engineering problems.
But Simulink toolbox is different in a way as it can provide a graphical user interface to the designer. This makes the designing process easy as compared to writing a program for the same problem. It also provides the facility to convert the GUI into codes so that if someone wants to work with the
programming the same will be available easily.
Below shown figure is the Simulink model of the ANN that is used for load forecasting. We have to give the inputs and when we run the program output can be seen by double click on the plot box (y1).
Figure: 5, Model of Artificial Neural Network in Matlab
4. RESULTS
The ANN works as a black box that is trained using the historical data and put for prediction of the future data. Here we have trained the network using load data taken from substation of local electric utility and day ahead prediction of load is generated.
Simulation results are shown in the below shown table and figure.
Table : load prediction
Figure: 6, Mean square error for the forecast
Monday Tuesday Wednusday Thusday Friday Saturday Sunday
No Target ForcastedNo Target ForcastedNo Target ForcastedNo Target ForcastedNo Target ForcastedNo Target ForcastedNo Target Forcasted 1 2700 2700 1 2700 2700 1 2700 2700 1 2700 2700 1 2700 2700 1 2700 2700 1 2700 2700 2 2590 2680 2 2590 2680 2 2590 2680 2 2590 2680 2 2590 2680 2 2590 2680 2 2590 2680 3 2570 2675 3 2570 2675 3 2570 2675 3 2570 2675 3 2570 2675 3 2570 2675 3 2570 2675 4 2598 2675 4 2598 2675 4 2598 2675 4 2598 2675 4 2598 2675 4 2598 2675 4 2598 2675 5 2600 2680 5 2600 2680 5 2600 2680 5 2600 2680 5 2600 2680 5 2600 2680 5 2600 2680 6 2810 2805 6 2810 2805 6 2810 2805 6 2810 2805 6 2810 2805 6 2810 2805 6 2810 2805 7 3010 2810 7 3010 2810 7 3010 2810 7 3010 2810 7 3010 2810 7 3010 2810 7 3010 2810
Figure: 7, Mean square error for the forecast
Figure: 8, Mean square error for the forecast
Figure: 9, Mean square error for the forecast
This work provides a unified approach that enables the “hourly”
resolution property for all of the mentioned forecast ranges. The
proposed method consists of a nested combination of two methods
for modelling and forecasting electric loads. The two methods are: neural network and linear regression models. The procedure of work could be summarized as a neural network was applied using electric load data from year 2002 to year 2006 to predict hourly load for one day, one week, one month and one year.
Figure 10 show the output of neural network training regression.
Figure 10. Neural network training regression plot.
RESULT
One of the primary tasks of an electric utility is to accurately predict load requirements at all times.
Results obtained from load forecasting process are used in planning and operation. Neural Network can learn to approximate any function just by using example data that is representative of the desired task. They are model free estimators, which are capable of solving complex problem based on the presentation of a large number of training data. Neural Networks estimate a function without mathematical description of how the outputs functionally depend on the inputs. They represent a good approach that is potentially robust and fault tolerant. In this work, an electric forecasting method based on neural network integrated with simple linear regression model was implemented using MATLAB. The system performs better results than some other systems. The accuracy can further be improved if we take more than one factor (calendar, temperature, humidity and random factors) as input, which is large enough to incorporate all the effects which can be quantified.
REFERENCES
[1] Haida, T. and Muto, S. 1994. Regression based peak load forecasting using a transformation technique.IEEE Trans. Power Syst., 9, 1788–94.
[2] Almeshaiei, E. and Soltan, H. 2011. A methodology for Electric Power Load Forecasting.Alexandria Engineering Journal, 50, 137–144.
[3] Filik, Ü. B., Gerek, Ö. N. and Kurban, M.
2011. A novel modeling approach for hourly forecasting of long-term electric energy demand.Energy Conversion and Management.
52, 199–211.
[4] Wang, J. Zhu, S. Zhang, W.and Lu, H.
2010. Combined modeling for electric load forecasting with adaptive particle swarm optimization.Energy,35, 1671–1678.
[5] Hill,T. O’Connor, M. and Remus, W. 1996.
Neural networks models for time series forecasts.Manage. Sci. 1082–92.
[6] Fan, J. Y. and McDonald, J. D. 1994. A real-time implementation of short-term load forecasting for distribution power systems.IEEE Trans. Power Syst. 9, 988– 94.
[7] Amjady, N. 2001. Short-term hourly load forecasting using time series modeling with peak load estimation capability.IEEE Trans.
Power Syst. 16, 798–805.
[8] Irisarri, G. D. and Widergren, S. E. and Yehsakul, P. D. 1982. On-line load forecasting for energy control center application.IEEE Trans. Power Appar. Syst. 101, 71–8.
[9] Rahman, S. and Hazim, O. 1996. Load forecasting for multiple sites: Development of an expert system-based technique.Electr.
Power Syst. Res. 39, 161–9.
[10] Kandil, N.,Wamkeue, R.,Saad, M. and Georges, S. 2006. An efficient approach for short-term load forecasting using artificial neural networks.Electr. Power Energy Syst.
28, 525–530.
[11] Santos, P.J.,Martins, A. G.and Pires, A. J.
2007. Designing the input vector to ANN- based models for short-term load forecast in electricity distribution systems.Electr. Power Energy Syst. 29, 338–47.
[12] Cai, Y.,Wang, J.-Z., Tang, Y. and Yang, Y.- C. 2011. An efficient approach for electric load forecasting using distributed ART (adaptive resonance theory) & HSARTMAP (Hyper- spherical ARTMAP network) neural network.Energy. 36, 1340-1350.
[13] Lin, W.-M., Gowa, H.-J. and Tsai, M.-T.
2010. An enhanced radial basis function network for short-term electricity price forecasting.Applied Energy. 87, 3226–3234.
[14] Lin, W.-M. , Gow, H.-J. and Tsai, M.-T.
2010. Electricity price forecasting using Enhanced Probability Neural Network.Energy Conversion and Management. 51, 2707– 2714.
[15] Che, J.,Wang, J. and Wang, G. 2011. An adaptive fuzzy combination model based on self-organizing map and support vector regression for electric load forecasting.(7th Biennial International Workshop “Advances in Energy Studies”), Energy.37, 1, 2012, 657- 664.
[16] Yang, X. Yuan, J. Yuan, J. and Mao, H.
2010. An improved WM method based on PSO for electric load forecasting.Expert Systems with Applications. 37, 8036– 8041.
[17] Bakirtzis, A. G., Theocharis, J. B., Kiartzis, S. J. and Satsois, K. J. 1995. Short term load forecasting using fuzzy neural networks.IEEE Trans. Power Syst. 10, 3, 1518–1524.
[18] Lauret, P.,Fock, E., Randrianarivony, R.
N. and Ramsamy,J.F. M.2008. Bayesian neural network approachto short time load
forecasting.Energy Convers. Manage.49, 5, 1156–1166.
