Short-term Electric Power Forecast in the Nigerian Power System Using Artificial Neural Network
Chapter One
Objectives of the Study
Although the objectives of the study can be inferred from the background to the study outlined in the previous section, it can still be clearly and concisely stated that the objectives of the study are:
- To model an artificial neural network which can forecast electric power supply for one day in advance (Short Term Load Forecasting);
- To train the model (using back propagation algorithm) with pre-historical load data obtained from a sample of the Nigerian power company so that each input produces a desiredoutput;
- To Test the model to get the values of future power supplies in the Nigerian power system ;and
- In the light of the above, make necessary recommendations and suggestions for further
CHAPTER TWO
LITERATURE REVIEW
Definition of Load Forecasting
Forecasting according to Sarangi et al, [22] is a phenomenon of knowing what may happen to a system in the next coming time periods. Chakrabarti and Halder, [23] also defined load forecasting as a method to estimate the load for a future time point from the available past data.
Vadhera, [1] seem, however, to have given a more comprehensive and acceptable definition of load forecasting when he notes that load forecast is no more than an intelligent projection of past and present demand patterns to determine future ones with sufficient reliability. The term load according to Vadhera, [1] is a device or conglomeration of devices that taps energy from the power system network. Load is a general term meaning either demand or energy, where demand is time rate of change of energy.
Importance of Load Forecasting
To cast the importance of load forecasting, Vadhera, [1] swiftly noted that good could not be emphasized enough in forecasting future requirements.
Still on the justification of the need for load forecasting, Alfares and Nazeerudin, [4] contended that load forecasting is a central and integral process in the planning and operation of electric utilities. Alfares and Nazeerudin, [4] went further to note that load forecasting involves the accurate prediction of both the magnitude and geographical locations of electric load over the different periods (usually hours) of the planning horizon. They went further to add that accurate load forecasting holds a great saving potential for electric utility corporations. According to Bunn and Farmer, [3], these savings are realized when load forecasting is used to control operations and decisions such as dispatch, unit commitment, fuel allocation and off-line network analysis. Adepoju et al, [24] shared the same view as Bunn and Farmer, [3] when they noted that load forecasting being very essential to the operation of electricity companies enhances the energy-efficient and reliable operation of a power system.
According to Adepoju et al, [24], the operation and planning of a power utility company requires an adequate model for electric power load forecasting. Load forecasting plays a key role in helping an electric utility to make important decisions on power, load switching, voltage control, network reconfiguration, and infrastructure development [24]. Feinberg et al, [25] was of the view that accurate models for electric power load forecasting are essential to the operation and planning of a utility company. Load forecasting helps an electric utility company to make important decisions including decisions on purchasing and generating electric power, load switching, and infrastructure development [25]. Those who can benefit from the knowledge of Load forecast include energy suppliers, ISOs, financial institutions, and other participant in electric energy generation, transmission, distribution, and markets [25].
Load forecasts can be divided into three categories: short-term forecasts which are usually from one hour to one week, medium-term forecasts which are usually from a week to a year, and long-term forecasts which are longer than a year [25], [4], [26], and [24].
Problems of Load Forecasting
Load forecasting, however, is not an easy thing. This is because load is affected by many physical factors such as weather, national economic health, popular TV programs, public holidays, etc. [27]. This actually makes load forecasting a complex process demanding experience and high analytical ability using probabilistic techniques including neural networks.
CHAPTER THREE
RESEARCH DESIGN/METHODOLOGY
Introduction
This chapter is focused on the simulation design which includes (a) Research Data
(b) Data pre-processing (c) Construction of Network Architecture (d) Requirement of minimum number of patterns, (e) Selection of input variables (f) model training and simulation, and (g) generalization problem.
Research data
The data that was used in training and testing of the model proposed in this research are the daily Electric power supplied to Enugu state of Nigeria as contained in the National Electric power authority New Haven, Enugu 132/33kV Transmission station daily hourly load reading sheets for the months of February and March 2011. A month’s data could do for the purpose of short-term electric forecasting [24, 22] and hence for this research the load profile for the month of March 2011 was actually used. The essence of collecting two months data is stated in the following subsection.
Data Pre – Processing
Although in theory all ANNs have arbitrary mapping capacities between sets of variables, it is convenient to normalize the data before carrying out the training to compensate for the inevitable scaling and variability differences between the variables [8]. Neural network training can be made more efficient if certain preprocessing steps are performed on the network inputs and targets [151].
Data pre-processing was in two stages: the first action on the load data was to find replacement for missing load values. The case of missing load values arose either due to system collapse, earth fault, transformers being on soak, feeder opened for the purpose of maintenance operation etc,. In cases like the above, load information for the same hour, weekday, and week of the preceding month would always be used to refill the missing gap. Calendar, however, showed that the months of February and March 2011 were the most compatible months for the purpose of this data doctoring. This is because the 1st days of both months were coincidentally on the same day of the week, namely Tuesday. The second operation on the data was performed to put the input values in the same scale. The approach here was to normalize the mean and standard deviation of the training set. This would be implemented using the matlab code ‘prestd’. This code normalizes the network inputs and targets so that they will have zero mean and unity standard deviation.
CHAPTER FOUR
EXPERIMENTAL RESULTS AND DISCUSSIONS
Introduction
Forecast results and statistical properties obtained from the application of the developed Short Term Load Forecasting (STLF) ANN model on the load data of New Haven Enugu transmission station, a typical Nigerian power system are presented and discussed in this chapter.
The STLF results for the utility of New Haven Enugu, Nigeria produced by the ANN structure proposed by this research were analyzed on the basis of the well-known statistical index, mean absolute percentage error (MAPE) stated in equation 3.1 and repeated below:
MAPE(%) = 1 ΣN
CHAPTER FIVE
CONCLUSION AND SUGGESTIONS FOR FURTHER RESEARCH
Conclusion
This work and its results show that the ANN represents a powerful tool for decision making in electric power utility companies.
The result of the feed-forward time-delay (NewFFTD) network model used for one day ahead short term load forecast for New Haven Enugu transmission station, a typical Nigerian Power System, shows that NewFFTD, which is a multilayer feed forward network with time delay, has a good performance, and reasonable prediction accuracy was achieved for this model.
Its forecasting reliabilities were evaluated by computing the mean absolute error between the exact and predicted values. The results suggest that ANN model with the developed structure can perform good prediction with least error and finally, this neural network could be an important tool for short term load forecasting. Our experimental results also show that a simple ANN-based prediction model appropriately tuned can outperform other more complex models.
We conclude therefore by saying that this research is a novel attempt to deal with load forecasting in the Nigerian power system by means of Artificial Neural Network with emphasis on ANN model simplicity, input data homogeneity without compromising forecasting accuracy.
Suggestions for Further Research
In spite of the delimitations of this work and the observations made on the course of the experiment proper, we make the following suggestions for further studies:
- The neural network typically shows higher error in the days when people have specific start-up activities such as Friday (for example on day 1 of the test set in table 4.6), or variant activities such as during Sundays which are like holidays in the Eastern part of the country (for example, on day 3 of the test set in table 4.6). In order to have more accurate results, one may need to have more sophisticated topology for the neural network which can discriminate start-up days from other days. In other words, a model with special holiday encoding may perform this task better.
- Determination of values of weights and biases that prompts fast convergence of training algorithm is still an issue in load forecasting by means of neural networks. So, a hybrid approach may be necessary to this effect. Use of genetic algorithms or swarm optimization techniques for the determination of weights in a back propagation network for short-term load forecasting may be of help in improving neural network performance in load
- Due to time constraint and financial limitations we narrowed our work to New- Haven, Enugu, Nigeria transmission station load profile. It may be reasonable if the model is tested on load data obtained from a larger part of the Nigerian Power system.
- Finally, since the effects of exogenous variables on models accuracy is still in contention today, development of a neural network model that in addition to pre- historical load data, can take as input some other exogenous variables as input data may be necessary for improved model forecasting
REFERENCES
- Vadhera, S.S., Power System Analysis and Stability, Khana Publishers, NaiSarak, Delhi,
- The Nigerian Dailies, Daily Sun Newspaper, p45, Monday January 10,
- Bunn, D.W, and Farmer, E. D., Review of Short-term Forecasting Methods in the Electric Power Industry, New York: Wiley, pp13-30,
- Alfares, H.K., and Nazeeruddin, M., “Electric Load Forecasting” Literature survey and classification of methods”, International Journal of System Science, Vol. 33(1), pp. 23-34,
- Haida, T., and Muto, S., to Electric Regression based Peak Load Forecasting Using a Transformation Technique, IEEE Transactions on Power Systems, Vol.9, pp.1788- 1794,
- Srinivasan, D., and Lee, M. A., Survey of Hybrid Fuzzy Neural Approaches to Electric Load Forecasting, Proceedings of IEEE International Conference on Systems, Man and Cybernetics, Part 5, Vancouver, BC, pp. 4004-4008,
- Kalaitzakis, K., Stavrakakis, G. S., Anagnostakis, E. M. “short-term load forecasting based on artificial neural networks parallel implementation”, Electric Power Systems Research 63, pp.185-196,
- Bassi, D., and Olivares, O., “Medium Term Electric Load Forecasting Using TLFN Neural Networks”, International Journal of Computers, Communications and Control Vol. 1 N0. 2. pp. 23-32.,
- Arroyo, D. O., Skov, M.K., and Huynh, Q., “Accurate Electricity Load Forecasting with Artificial Neural Networks”, Proceedings of the 2005 International Conference on Computational Intelligence for modeling, control and Automation, and International conference on Intelligent Agents, Web Technologies and Internet Commerce,
