# Load Forecasting for Peak Power Station Using Artificial Intelligence

8453 words (34 pages) Full Dissertation in Full Dissertations

06/06/19 Full Dissertations Reference this

Disclaimer: This work has been submitted by a student. This is not an example of the work produced by our Dissertation Writing Service. You can view samples of our professional work here.

Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of UK Essays.

Abstract

# Chapter 1: Introduction

1. Research Background

Electric load forecasting has become a major research field in electrical engineering. In a smart grid system, forecasting of electricity demand is indispensable to help system operators to well-organize the schedule of the spinning reserve allocation. Besides that, the load forecasting helps in providing information for the energy interchange with others utilities.

Electric load forecasting, usually use time series analysis (Box et al., 2015) and regression analysis (Armstrong, 2011) to predict the future electricity. Time series analysis is the method of using the previous load pattern as a time series signal and predict the future load demand. The review from other research, say the time series method consists of two problems which is inaccurate prediction and numerical instability (D.C. et al., n.d.). The reason that occur inaccurate prediction and instability of numerical is due to the load pattern that lack of information for the various factor such as the temperature of the day, humidity of the area, weather and the holiday period. All these information are dependent and independent variables that can affect the outcome of the electricity demand. Therefore, regression analysis is use to recognize the load pattern as dependent on the variable and find the functional relationship between the variables and the load.

Neural network (Neural Networks, n.d.) are processing electronic model based on the neural structure of the brain. The neural network system will consist a lot of the artificial neuron to receive the information and response it after process the information. Electric load forecasting require a lot of load data, variable and information, therefore a systematic artificial neural network require to process all this information(Kalaitzakis et al., 2002).

1. Problem Statement/ Research Question

Load forecasting is define as an action to predict the future load at different time interval. Its plays an important roles in system planning, operation and control. Therefore, the accurate of the load forecasting become crucial. To increase the accuracy of the forecast, there are few factors need to be consider in the research such as weather factor, time and seasonal factor, and others factor(Kiartzis et al., 1997a).

1. Research Objectives

There are four important task in designing an artificial neural network. The four task are data pre-processing, neural network design, neural network implementation and validation(Demuth et al., 2014). Data pre-processing is the task that done before the data send to neural network system which can make the forecasting problem to be more controllable. The purpose of data pre-processing is to filter the data by removing the outliners, missing values and some irregularities because neural networks is very sensitive with defective data. In determine the shape of the load profile, one of the factor to be consider was the calendar date because the data will be influenced by week days, weekend, holidays, social customs and the working pattern. All these influences can be classify by the data pre-processing. The data pre-processing will create different classes to set each influence as a factor on determine the shape of the load profile.

In neural network design, selecting a suitable structure is very important. A suitable structure of neural network is normally decided by the number of hidden layer been use, input nodes, neuron per layer, and type of activation layer. However the number of output neurons, input neurons and hidden neurons can be decide by other factors. In electric load forecasting, number of output neuron are decide by three different method of forecasting which are iterative forecasting, multi-model forecasting and single-model multivariate forecasting. Basically, iterative forecasting require the least number of output and single-model multivariate forecasting require the most number of output. For the input neurons are decide by the factor that affect the output of the system.

After designing the neural network, the next task is to do the neural network implement and validation. Neural network will input the data from the past electric load, and algorithm will predict. When the neural network forecast the electric demand, it will tested for the validation. The test for validation is very important to track the error percentage from the neural network. The lower the error percentage, the more accurate of the neural network.

1. Research Scope

This research will focuses on designing the peak load forecasting. To achieve an accurate peak forecasting, various factors such as time analysis, regression analysis and others factor will include to the research. The relationship between the various factors and the load data will build utilizing artificial neural network.

1. Research Contribution/ Significance

As been defined, peak load forecasting is the forecast of predict future peak load demand by utilize artificial neural network. The data that been forecasted can be essential part for power system planning and operation. For example, by getting the long-term forecasting result (forecast results for one year ahead) can helps to predict the needs for expansion of the power station, getting equipment purchases and staff hiring. For the short-term forecasting result (forecast results for one day ahead) can supply information for system management of day-to-day operation and unit commitment (Kiartzis et al., 1997b).

Besides that, load forecasting provides many advantages such as enables utilities company to plan well, help to determine the requires resources, and maximum utilization of the power generating plan (Drezga and Rahman, 1998). Therefore, the accuracy of the load forecasting is necessary to power generating station.

1. Summary

In this research, the peak load forecasting will focus on designing the structure of artificial neural network. The structure of the network is mostly depending on the relationship of the various factor such as time factor, weather factor or season factor and the load data. A balance neural network structure is basically following four task of designing which. The four task are data pre-processing, neural network design, neural network implementation and validation.

# Chapter 2: Literature Review

2.1 Introduction

In this chapter, the contain will divided into three part. The first part is the concept and theory about load forecasting that will used in the research. In the second part literature review on the load forecasting will be discussed and last the research frame work will be discussed.

The first part of this chapter which is the concept and theory will discuss the method that will use to accomplish this research. Peak load forecasting consist of three type which divide to long-term load forecasting, medium-term load forecasting and short-term load forecasting. Each type of forecasting is divided by the period of the forecast and the method to forecast is different for long-term, medium-term and short-term forecasting. For long-term load forecasting, the forecast period is between one week to one year and medium-term load forecasting is forecast for few weeks. Lastly for short-term forecasting the casting period is with-in day-to-day ahead forecasting. The concept to short-term forecasting will discuss at this part.

Besides that, in the part of concept and theory will discuss on the method to forecast the load forecasting will be done. The method is data pre-processing, neural network design, neural network implementation and validation. This four step is to arrange the data in a form so the neural network can be read as data. Then the designing or programming the network with the relationship of the past load data and various factor. After that, implement of the network will be done and lastly validation is to track the percentage error and reducing it.

For the second part of this chapter, the literature review will be discuss on the research paper that have review based on this title. Beside, in this chapter, the factor such as temperature and time factor will be discuss here. For example, the load demand in night will be higher than afternoon or morning due to the application of the load. The neural network will discuss at here too.

Lastly in the third part of the chapter is about research framework. In this part will explain of the concepts and, together with the definitions and reference to relevant scholarly literature, existing theory that is used for my particular study.

2.2 Concept and Theory

A proper operation of electric utilities will requires short-term, medium and long-term forecasting of load demand. The load forecasting in this research will focus on short-term forecasting of load demand. A medium and long-term load forecasting with the forecasted times which long enough to one year ahead, this can helps to plan for long-term maintenance, construction scheduling for developing new generation facilities, purchasing of generating units, developing transmission and distribution systems. However, the accuracy of the medium and long-term load forecast will bring influence on developing future generation and distribution planning. In the medium and long-term electric load demand is depends on a complex factor such as daily and seasonal weather, national economic growth and social habits (Mbamalu and El-Hawary, 1993). These factor consist of high nonlinear characteristic and hard to achieve an accurate forecasting results for medium and long-term electric load demand.

Short-term load forecasting is forecast load demand for several hours to several weeks ahead using various factor such as temperature and based on the observed load time series data. For power generating station, load forecasting become an important aspect in develop any model for electrical planning and the past load and weather information are critical to predict the forecasting result. However, the various factor that influence the forecasting are too many. For example, the time factor, in the weekend the electric uses will be higher demand compare with weekdays and the special days like New Year, Chinese New Year, and Deepavali will have the power demand higher than any normal days. All these data, must let the neural network to train. Therefore, the percentage of error during forecasting will be higher (Zhang et al., 1998).

Other than that, the method of designing a neural network based of forecast should be include to be discuss in this chapter. As mention in the introduction of the proposal, there are four important task should be done in sequence, so that a good structure of neural network can be build. The four design task planning is

1. Data pre-processing
2. Neural Network design
3. Neural Network implementation
4. Validation

i) Data pre-processing

In the first task on designing a neural network structure is data pre-processing. The load data that obtain on the previous day which recording the data from power plant station may contain some missing data and noise. The missing data and noise will make the load data to be abnormal data. So, if the neural network process with these abnormal data, interfere in load forecasting will occur. Even, there are interfere occur during the load forecasting, neural network will still generate the forecast data but the data will be in the low compromise of the feature.

The review from the research paper (Anders and Korn, 1999), data pre-processing is require during the research due to it helps the neural network to learn the behavior of the load and it helps to provide an accurate and reliable forecasts. Review of research paper (Armstrong and Fildes, 1995) state the results obtain without pre-processing of the data consist of high error percentage. Therefore, the pre-processing is important step to filter the data, by removing the outliners, missing values or any irregularities due to the neural network system are sensitive to such data.

Pre-processing work by classifying the input data and then using separate neural networks to model data from each class. For example, in recognizing the pattern of the load profile the time will be one of the important factor. So, the time factor will be classified to three group which is morning, afternoon and night. Besides that, there research paper which using the same concept to solve the holiday pose a special problem (Song et al., 2005)

ii) Neural Network designing

The first step on designing a neural network based forecasting system is to select an appropriate architecture. Selecting a neural network architecture involve several decision making in the type, size and number of neural network used. There are many types of architecture have been used in the research of load forecasting system. Most of the research will use multilayer feed-forward neural network due to the good performance.

To design a multilayer feed-forward network, it needs to choose the number of output neurons, input nodes, and hidden neurons. There are three ways on selecting the number of output neurons; a) iterative forecasting, b) multi-model forecasting, c) single-model multivariate forecasting.

1. Iterative Forecasting

Iterative forecasting is to forecast of one hourly load at a time and then aggregating it to the series, so that, the forecast for the later hours will refer back to the forecast for the earlier ones. In the reviewed paper, there are one paper which using this method. It conclude that, this method is suitable for short-term load forecasting in a deregulated environment (Hill et al., 1994).

1. Multi-model forecasting

This method is common for load forecasting with regression models which using 24 different models and one model for each hour of the day. The advantage of this method is that the individual networks are relatively small, and they will not likely to be over fitted.

1. Single model forecasting

Single model forecasting is the most popular method, this method using a multivariate method to forecast all the loads at once, so that each profile is represented by a 24-dimensional vector.

In many research, the structure of the neural networks used for load forecasting is mostly determined by how the way of a particular task that needs to perform by a single network. One of the method, which commonly used, is to apply one large network to forecast the entire day load. For example, a network will have a 24 outputs and appropriately large number of inputs. On the other hand, is to use a set of small neural networks, one for each time lead or for each hour. A set of 24 networks is typically used and every networks will have only one output and few inputs. The second method is much easier to design and train a set of 24 small networks than one large network. Using this method can have high accuracy and it is more suitable for on-line applications.

Besides selecting the number of layers and output neurons required, it needs to select the number of input nodes. There are few ways to help to select the number of input, one of a priori knowledge about the behavior of the system under study. To determine the number of neuron in the hidden layer, it may be can compared to that of choosing the number of harmonics to be included in a Fourier model to approximate a function. If they are too few, the model will not flexible enough to model the data well but if they are too many, the model will overloaded the data (Shimakura et al., 1993)

iii) Neural network implement

After designing the neural network, the neural network will be trained. Training in neural network is refer to estimate the parameters, where it will select a “training algorithm”. The most common algorithm is the back propagation algorithm, which based on a steepest-descent method that performs stochastic gradient decent on the error surface (Drezga and Rahman, 1998). Neural network system are “data-driven” methods, which mean it require large samples in training.

iv) Neural network validation

The last stage of forecasting system is make validation. This task is important to predict the actual performance of a method, so the researchers could test their models by examining their errors in sample other that the one used for parameter estimation.

2.3 Literature Review

The review form most research paper on short-term load forecasting will only focus at temperature or weather as their factor that will influence the forecasting. However, in this research other variables such as time factor may include as a variable, in order to achieve high accuracy of forecast.

i) Weather factor

The most important variable for load forecasting is the weather. The effect of weather can alter the load profile of industrial consumer. To minimize the operational cost, load forecasting model will use of weather forecast and other factors to predict the future load. However, the sea breeze, after moon thunderstorms, back door fronts is some of the environmental factors that can decrease the temperature and causing the forecasting will be overestimated load forecast (Usman Fahad and Arbab, 2014). Therefore, the power generator will generate more power than required. This situation is same as the forecasting during dry season can cause the load demand to increase more than expected.

Besides that, temperature can also effect the conductivity of the transmission line and the carrying capability of the transmission lines. High temperature will increase the resistance of the transmission lines, and alter the reactance of line, due to temperature induced expansion of the length of transmission line. From the weather factor there are including some other factors like temperature, humidity, precipitation and wind speed and wind chill.

1. Temperature

Temperature is the measure of average kinetic energy of the atoms or molecules of an object or it can defined as “the measure of degree of hotness or coldness of a body”. From the review of research on influence of temperature on short-term load forecasting (Paravan et al., n.d.), shows that there is high positive correlation between temperature and load during summer season and there is a negative correlation between temperature and load during winter. This prove that during summer season, increase in temperature will affect the load to increase and when the temperature decrease the average daily load will decrease as well as the peak demand will be low. The situation during summer season will be the contrast for the winter season. The reason of the situation is due to the temperature affect, where during summer increase in temperature will results the people feeling of comfort and consumer will use electricity for cooling purpose, where in winter electricity is used for heating purposes.

1. Humidity

Humidity is defined as the amount of water vapors in air. Humid air is refer as the mixture of a water vapors and other constituents of air. In practical world the water contents of this mixture is call absolute humidity or relative humidity and expressed in percentage. The humidity in certain area can increase apparent temperature while it has no effect on the real temperature. For example, when the temperature of the area is 30 degree celcius but due to the humidity the surrounding felt like 35 degree celcius. Humans are sensitive with humidity due to the mechanism used to regulate the body temperature is evaporative cooling. When there is high humid atmosphere, the rate of evaporation through skin is lower than normal humid. Human are perceives rate of transfer of heat rather than temperature itself, so we can feel warmer at high humid condition. Therefore, humidity can increase the feeling of temperature increase, so it will cause people to use more cooling appliances. This will cause the peak load demand to higher value during humid day. Even though, humidity has no direct effect on real temperature but it can manipulate the severity of hot climate. As a conclusion, the prediction of daily load of domestic consumer it must consider apparent temperature instead of real temperature.

1. Precipitation

Precipitation can directly and indirectly effect load consumption. For the direct effect example, during heavy rain can make people to stay home and can cause luminous intensity to decrease. So because to the rain, people will be forced to stay in door and they will consume more electricity for lighting purpose and entertainment appliances. For indirect effect example, during heavy rain can decrease the temperature thus may have positive or negative effect on load consumption. In raining season effect of precipitation can be positive because temperature will decrease and so less AC and other cooling appliances used will reduce the load demand. In monsoon season the effect of precipitation is negative because it can decrease the temperature and can further intensify the severity of cold weather. Therefore, due to over use of electric heater during rain a sudden peak will arise (Tong et al., 2017). Rain in monsoon season will decrease temperature and also make people to stay in door causing more power used for heating purpose and lighting the room. Therefore, short-term load forecasting must consider for precipitation factor to accurately predict the load and if not considered the predicted load may be have higher percentage of error.

1. Wind speed and wind chill

Wind speed can be defined as the motion of air respect to the surface of the earth covering a unit distance in unit time. In weather forecast, wind speed is one of the manipulation and it measured with anemometer. Wind direction and speed is affected by three main factors. First is the fractional forces, such as the wind crossing forest, mountains and buildings. Second is temperature gradient which sometimes referred as pressure gradient and lastly is earth rotation known as Coriolis Effect. On load consumption, low humidity rates the speed of wind will lowers the apparent temperature and increases the rate of evaporation of perspiration from the human body. Therefore it gives the cooling effect. So, during a windy day of summer the consumption of electricity will be lower because lesser cooling appliances will be used.

Besides, if during a windy day renewable energy resources generator including wind power generators then the output power of wind generators will be high. By forecasting the load demand of generating stations during this situation then we must consider wind speed as important factor because instead of over generation by the thermal generators, the short fall can be compensated by the increase in power generation of wind turbines, because electricity generated by wind turbines is economical then the same generated by thermal or even by hydropower plants. The wind speed can manipulate the apparent temperature. Therefore during a windy day, load demand will be low, so load forecasting model employed even for distribution feeders.

ii) Time factor

Time factor is one of the main factor in short term load forecasting because its impact on consumer load is highest.

Graph 1: Total power generate on Monday 22th MAY 2017 generation of power station.

From the observation of daily load curve of any grid station it can be seen that load variation follows certain rules with the “time point” of the day. In graph hourly load curves is shown for 24 hours of a day. The interval of each point is one hour, so there are 24 intervals in the figure. It can be seen in figure that the load is low and stable from 0 am to 8 am (0am means 12 o clock at night), the load start rising at 9 o clock till 2 pm.

After that it descends till 4 pm (16:00 in graph) and after 6 pm it start rising till 8 pm (20:00), after 8 pm load gradually decrease again until the end of the day. It can be seen in figure 1.1 that maximum demand occurs at 8pm, and minimum load demand occurs after mid night. So if we closely observe this load curve it can be seen that load demand reflects the consumer’s daily life style. At mid night everyone is sleeping so there is no need of lighting in the house, so load becomes least. Similarly at 8 pm everyone is at home watching TV, doing homework, so load is highest at that part of the day.

From the graph we can know the people’s daily life activities can be classified in to three parts

1.  Working Time
2.  Leisure Time
3. Sleeping Time

In graph 1 the load curve has a flat point in the working time and one peak load in the leisure time, while at sleeping time load curve shows least value. The rule of load variation discussed above is not the only rule. There are certainly other rules of load variation with time.

Graph 2: Total power generate on Sunday 21th MAY 2017 generation of power station.

For example is observed on graph 2, weekend load to be high than the week day loads due to the increase of leisure time.

Artificial Neural network

Other than review on factor that affect the load forecasting, learning artificial neural network also important for the peak load forecasting. Artificial neural network are mathematical tool that mimic the way of human brain processes information. Neural network consist neurons where it is to enable data receiving through a number of input nodes, process it and put out the responses based on the input data given. The neuron is divided in two stages process which the first state is to combine the input values and the result will used as the argument of a non-linear activation function. For example, given the input set of data of the temperature is X and load demand is Y, this will give a relationship between the two variable. When the input of one nodes is depend on the output of the previous nodes this method is known as feed-forward and the layer between the input nodes and the output nodes is known as hidden layer. Neurons that be organized in multiple layers which known as multilayer perceptron (MLP). However, all the layer that been arranged will not connect with each other (Neural Networks, n.d.).

Figure 1: Multilayer perceptron with input layer, output layer and hidden layer.(“A Quick Introduction to Neural Networks,” n.d.)

Multilayer perceptron have the ability to estimate the parameter which known as “training” of the network. Training is use the data which consist of the input and associated output vectors. During training the multilayer perceptron is repeatedly presented with the training data and the weights in the network are adjusted until the desired input—output mapping occurs. Multilayer perceptron learn in a supervised manner. During training the output from the multilayer perceptron, for a given input vector, may not equal the desired output. An error signal is defined as the difference between the desired and actual output. Training uses the magnitude of this error signal to determine to what degree the weights in the network should be adjusted so that the overall error of the multilayer perceptron is reduced. There are many algorithms that can be used to train a multilayer perceptron.

Training a multilayer perceptron is the procedure by which the values for the individual weights are determined such that the relationship the network is modelling is accurately resolved. At this point we will consider a simple multilayer perceptron that contains only two weights. For any combination of weights the network error for a given pattern can be defined. By varying the weights through all possible values, and by plotting the errors in three-dimensional space, we end up with a plot like the one shown in figure 1. Such a surface is known as an error surface. The objective of training is to find the combination of weights which result in the smallest error. In practice, it is not possible to plot such a surface due to the multitude of weights. What is required is a method to find the minimum point of the error surface.

Peak load means the time of high demand and the peaking demands are always require for only shorter durations. Peak demand could be explain as the difference between the base demand and the highest demand. For examples, the household loads such as microwave oven, electrical cooking apparatus, and television will switch on for a short period of time. Peaking power plants are power plants that will run only when there is a high demand for electricity. Due to peak power plant supply power only for certain period, the power supplied commands a much higher price per kilowatt hour than base load power. Peak load power plants will combination with base load power plants, which supply a dependable and consistent amount of electricity to meeting the minimum demand. The peak load power plants are generally use the natural gas as the power source but there are few burn diesel oil and petroleum.

Power plant

Power plant, also refer as power station or power house, is an industrial facility for the generation of electric power. Some of the power plant contain more than one generators and machine that converts mechanical power into electrical power. There are many type of power plant where some will use relative motion between magnetic field and a conductor creates an electrical power. Most of the power station will use natural resource to generate power such as coal, natural gas and petroleum but there is an increasing use of renewable source such as solar, wind, wave and hydroelectric.

2.4 Research Framework

There are many research paper in the development of a short-term load forecasting model based on Artificial Neural Network. In few months earlier, there are research about short-term residential load forecasting based on resident behavior learning. The research aim for a Long-Short Term Memory based deep learning forecasting framework with appliance consumption sequences is proposed to address such volatile problem(Kong et al., 2017). The forecasting result shown that the forecasting accuracy can be notably improved by including appliance measurements in the training data. The effectiveness of the proposed method is validated through extensive comparison studies on a real-world dataset. At the end of the research, the proposed LSTM recurrent neural network based forecasting framework performs better in establishing meaningful temporal relationships between consumptions across time intervals. Besides, when there are consumption sequences of major appliances are available, they can further improve the meter-level forecasting accuracy under the proposed LSTM framework. This research paper have include many contextual variables such temperature, humidity, the day of the week and special events to increase the accuracy.

Besides that, the research on “artificial neural network based monthly load curves forecasting” have review others method besides artificial neural network(Barbulescu et al., 2016). There have make a comparison between the ANN and ANFIS (adaptive neuro-based fuzzy inference system) for short term load forecasting is performed in other research paper. Real data have been used, being represented by the load and temperature data in Turkey. Backpropagation algorithm has been implemented. The comparison is provide more idea and method for the load forecasting.

2.5 Summary

As discuss earlier, load forecasting is important for future planning and controlling of a power station. While peak load forecasts are important for planning but in particular, it more for securing adequate generation, transmission and distribution capacities. In more specific, peak forecasts are important in decision making capabilities in capital expenditures and improve reliabilities of the system. Therefore, low accuracy of peak load forecasting will lead to bad planning and inefficient operation of power plant station.

# Chapter 3: Proposed Methodology

In this chapter, the planning of the project will be proposed and discussed. The management planning of this project will be listed out.

Table 1: Plan activity of the project for first semester.

 ACTIVITY PLAN START (week) PLAN DURATION (week) ACTUAL START (Weeks) ACTUAL DURATION (per week) Meet for discussion on final year title 1 1 1 1 Review power station 2 1 2 1 Review load forecasting 3 1 3 1 Review neural network 4 1 4 1 Write proposal form 5 1 5 1 Write abstract 6 1 6 1 Write Introduction 7 1 7 1 Write literature review 8 1 8 1 Write methodology 9 1 9 1 Submit proposal and do stimulation for load forecasting 10 1 10 1 Prepare for presentation 11 1 11 1 Do stimulation for load forecasting 12 1 12 1 Presentation for proposal defense 13 1 13 1 Do stimulation for load forecasting 14 1 14 1

Flow chart : Plan for task designing

Get the power load data and daily weather.

Extract the data such as peak power and temperature for every hours

Design the structure of the neural network.

Implement the neural network by training with new data.

Test for neural network

Neneural

Reset training data

Low error of percentage

Flow chart 1: The plan for designing neural network work for the load forecasting.

Research Planning (Gantt Chart)

Gantt Chart 1: The work plan for first semester

 ACTIVITY PERIODS 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Meet for discussion on final year title Review peak power station Review load forecasting Review neural network Write proposal form Write abstract Write Introduction Write literature review Write methodology Submit proposal and do stimulation for load forecasting Prepare for presentation Do stimulation for load forecasting Presentation for proposal defense Do stimulation for load forecasting

Gantt Chart 2: The work plan for second semester

 ACTIVITY PERIODS 1 2 3 4 5 6 7 8 9 10 11 12 13 Build prototype Neural Network design Neural Network Implement and validation Start writing thesis Pass up thesis Prepare for presentation Presentation for project

# Chapter 4: Conclusion

As the conclusion, the peak load forecasting is design for forecast the peak load on the power generating station. Peak power plant is not operate for the whole day, it only operate for the peak hours. Therefore, the load forecasting need to be accurate. In order to achieve the high accuracy, the various factor such as temperature, peak hours and weather will include as the data for implementation. The high accuracy forecasting is important for planning and operate the power plant station to the maximum efficiency.

# References

A Quick Introduction to Neural Networks [WWW Document], n.d. URL http://www.kdnuggets.com/2016/11/quick-introduction-neural-networks.html/2 (accessed 8.1.17).

Anders, U., Korn, O., 1999. Model selection in neural networks. Neural Netw. 12, 309–323. doi:10.1016/S0893-6080(98)00117-8

Armstrong, J.S., 2011. Illusions in Regression Analysis (SSRN Scholarly Paper No. ID 1969740). Social Science Research Network, Rochester, NY.

Armstrong, J.S., Fildes, R., 1995. Correspondence on the selection of error measures for comparisons among forecasting methods. J. Forecast. 14, 67–71. doi:10.1002/for.3980140106

Barbulescu, C., Kilyeni, S., Deacu, A., Turi, G.M., Moga, M., 2016. Artificial neural network based monthly load curves forecasting, in: 2016 IEEE 11th International Symposium on Applied Computational Intelligence and Informatics (SACI). Presented at the 2016 IEEE 11th International Symposium on Applied Computational Intelligence and Informatics (SACI), pp. 237–242. doi:10.1109/SACI.2016.7507378

Box, G.E.P., Jenkins, G.M., Reinsel, G.C., Ljung, G.M., 2015. Time Series Analysis: Forecasting and Control. John Wiley & Sons.

D.C., P., M.A., E.-S., R.J., M., n.d. Electric load forecasting using an artificial neural network – IEEE Xplore Document [WWW Document]. URL http://ieeexplore.ieee.org/abstract/document/76685/ (accessed 8.1.17).

Demuth, H.B., Beale, M.H., De Jess, O., Hagan, M.T., 2014. Neural Network Design, 2nd ed. Martin Hagan, USA.

Drezga, I., Rahman, S., 1998. Input variable selection for ANN-based short-term load forecasting. IEEE Trans. Power Syst. 13, 1238–1244. doi:10.1109/59.736244

Hill, T., Marquez, L., O’Connor, M., Remus, W., 1994. Artificial neural network models for forecasting and decision making. Int. J. Forecast. 10, 5–15. doi:10.1016/0169-2070(94)90045-0

Kalaitzakis, K., Stavrakakis, G.S., Anagnostakis, E.M., 2002. Short-term load forecasting based on artificial neural networks parallel implementation. Electr. Power Syst. Res. 63, 185–196. doi:10.1016/S0378-7796(02)00123-2

Kiartzis, S.J., Zoumas, C.E., Theocharis, J.B., Bakirtzis, A.G., Petridis, V., 1997a. Short-term load forecasting in an autonomous power system using artificial neural networks. IEEE Trans. Power Syst. Vol. 12.

Kiartzis, S.J., Zoumas, C.E., Theocharis, J.B., Bakirtzis, A.G., Petridis, V., 1997b. Short-term load forecasting in an autonomous power system using artificial neural networks. IEEE Trans. Power Syst. 12, 1591–1596. doi:10.1109/59.627863

Kong, W., Dong, Z.Y., Hill, D.J., Luo, F., Xu, Y., 2017. Short-Term Residential Load Forecasting based on Resident Behaviour Learning. IEEE Trans. Power Syst. PP, 1–1. doi:10.1109/TPWRS.2017.2688178

Mbamalu, G., El-Hawary, M., 1993. Load forecasting via suboptimal seasonal autoregressive models and iteratively reweighted least squares estimation [WWW Document]. ResearchGate. doi:http://dx.doi.org/10.1109/59.221222

Neural Networks, n.d.

Paravan, D., Debs, A., Hansen, C., Becker, D., Hirsch, P., Golob, R., n.d. Influence of Temperature on Short-Term Load Forecasting Using the EPRI-ANNSTLF.

Shimakura, Y., Fujisawa, Y., Maeda, Y., Makino, R., Kishi, Y., Ono, M., Fann, J.Y., Fukusima, N., 1993. Short-term load forecasting using an artificial neural network, in: [1993] Proceedings of the Second International Forum on Applications of Neural Networks to Power Systems. Presented at the [1993] Proceedings of the Second International Forum on Applications of Neural Networks to Power Systems, pp. 233–238. doi:10.1109/ANN.1993.264285

Song, K.-B., Baek, Y.-S., Hong, D.H., Jang, G., 2005. Short-term load forecasting for the holidays using fuzzy linear regression method. IEEE Trans. Power Syst. 20, 96–101. doi:10.1109/TPWRS.2004.835632

Tong, C., Li, J., Lang, C., Kong, F., Niu, J., Rodrigues, J.J.P.C., 2017. An efficient deep model for day-ahead electricity load forecasting with stacked denoising auto-encoders. J. Parallel Distrib. Comput. doi:10.1016/j.jpdc.2017.06.007

Usman Fahad, M., Arbab, N., 2014. Factor Affecting Short Term Load Forecasting. J. Clean Energy Technol. Vol. 2.

Zhang, G., Eddy Patuwo, B., Y. Hu, M., 1998. Forecasting with artificial neural networks: Int. J. Forecast. 14, 35–62. doi:10.1016/S0169-2070(97)00044-7

## Related Services

View all

### DMCA / Removal Request

If you are the original writer of this essay and no longer wish to have the essay published on the UK Essays website then please:

Prices from
###### £124

Undergraduate 2:2 • 1000 words • 7 day delivery

Delivered on-time or your money back

(30 Reviews)