The researchers propose the use of a neural network training algorithm that is a hybrid of three data mining techniques namely clustering, classification, and backpropagation in order to forecast a 24-hour load curve. The former consists of applying the K-means algorithm to daily morning load feature of a 24-hour electrical load data set and using the clusters/seasons to draw data-types. A comparison was made between the k- Nearest Neighbors approach and Naïve Bayes theorem to determine which would yield a higher accuracy in classifying data types using a sample cluster. The results show the k-Nearest Neighbors is the better classifier together with K-means. The next phase is the artificial neural network builder or training stage (learning phase). This approach is proposed to improve the learning performance of neural networks in the context of seasonality by optimizing the data sets. This paper focuses on the results of the first stage and is a work in progress. Specifically, it is envisioned that by using the results of the first phase, the network builder will be able to perform faster and produce better results. It is said that good clustering applications and classifying techniques can be judged by their predictive power. For this case, the results show that the use of the daily load morning slope is effective with or without the use of temperature values for K-means clustering. The results also show that when clustering load data, a better approach can be the use of a feature such as the daily morning slope in order to accurately reflect load behavior. By doing this, it is envisioned that a more accurate forecast can be made.
This manuscript adapts the DNNs to short-term natural gas demandforecasting and evaluates DNNs’ performance as a forecaster. Little work has been done in the field of time series regression using DNNs, and almost no work has been done in the field of energy forecasting with DNNs. One notable example of literature on these subjects is Qui et al., who claim to be the first to use DNNs for regression and time series forecasting [ 13 ]. They show promising results on three electric loaddemand time series and several other time series using 20 DNNs ensembled with support vector regression. However, the DNNs they used were quite small; the largest architecture consists of two hidden layers of 20 neurons each. Because of their small networks, Qui et al. did not take full advantage of the DNN technology.
The aim of this research work is Short-TermLoadForecasting of Chhattisgarh Grid by using the data obtained from State Load Dispatch Centre (SLDC) of Chhattisgarh State Power Transmission Company Limited (CSPTCL). Artificial Neural Network (ANN) is used to forecast the one day ahead load-demand requirement for Chhattisgarh grid. A complete database of loaddemand ranging from 5 th March 2014 to 3 rd March 2015 on daily 24-hour format, along with the maximum and minimum
The first regression model used was based on neural networks. In fact, this study uses the Feed-Forward Neu- ral Network architecture because they can approximate any square-integrable function to any desired degree of accuracy provided a training set [7,8]. A simple FFNN contains an input layer and an output layer, separated by l layers (the set of l layers is known as hidden layer) or neuron units. Given an input sample clamped to the input layer, the neuron units of the network compute their pa- rameters according to the activity of previous layers. This research considers the particular neural topology where the input layer is fully connected to the first hidden layer, which is fully connected to the next layer until the output layer.
This project is about the forecasting of load in Peninsular Malaysia for shortterm period such as for a week ahead load based on demand for 7 weeks previous data. Because of the power supply company in the Peninsular Malaysia is Tenaga Nasional Berhad (TNB), thus the historical load data from this company is applied as inputs data for this project. The method that applied in this project is a Feed Forward Neural Network method which will be simulated by using the Matlab software.
influence the electricity demand. The energy conservation policy advocate people substituting their high-energy device to low-energy device, it also reduces electricity demand. Things like these will be recorded in the data table for text mining. It likes an abstract to make people remember what happened during a day, or a period. And with the development of information technology, these information can be easily got from html pages in the Web. When the structured text data is prepared, the text mining process will be used to exact the implicit knowledge and rules. Classification is one of the most important tasks in text mining. The main goal of classification is to find the knowledge from the condition attributes to decision attributes. For loadforecasting corrective, it mainly finds the if-then rules from the factors to the errors, and these rules can be integrated with the CNN’s result. And the final result should be more accurate.
on their local generation/consumption conditions. The introduction of these highly dynamic elements will lead to a substantial change in the curves of demanded energy. The aim of this paper is to apply Short-TermLoadForecasting (STLF) in microgrid environments with curves and similar behaviours, using two different data sets: the first one packing electricity consumption information during four years and six months in a microgrid along with calendar data, while the second one will be just four months of the previous parameters along with the solar radiation from the site. For the first set of data different STLF models will be discussed, studying the effect of each variable, in order to identify the best one. That model will be employed with the second set of data, in order to make a comparison with a new model that takes into account the solar radiation, since the photovoltaic installations of the microgrid will cause the power demand to fluctuate depending on the solar radiation.
Short team loadforecasting is the prediction of electrical loaddemand for a period varying from the next few minutes up to a week. Short team loadforecasting plays a vital role in system operation and is the main source of information for all daily and weekly operations concerning generation commitment and scheduling. Short-team load forecast is also important for the economic and reliable operation of the power system. In order to achieve high forecasting accuracy and speed, it is required to know the factor that affects the load. Some of this factor is: the type and time of day, the weather conditions of the forecasting area, the season, etc. since most days have different load profiles, it is necessary to have a day type. Time of the day is an important factor in short team loadforecasting. It is required to know the forecasting time of the day is different. Therefore, the relationships between this factors and the loaddemand need to be determined so that the forecast may be as accurate as possible..
management and is important for the electricity industry in the deregulated economy. It has many applications which includes energy purchasing, generation, load switching, contract evaluation and infrastructure development. A large variety of mathematical methods have been developed for loadforecasting . An accurate load forecast can be very helpful in developing a power supply strategy, finance planning, market research and electricity management . For every forecast, there are different factors to be put into consideration. These factors to a great deal determine how accurate the forecast will be, as well as determine the loaddemand and hence, affect the load curve. These factors include calendar effects, seasonal variations, weekday variations, weekend-days variation, weather and temperature amongst others. Calendar effects include the effects of working days or trading days and holidays. Growth in the economy, population, extreme weathers may also contribute to annual variations. These variations need to be considered in forecasting energy intended for domestic use .
Two Layer Neural Network Power Systems architecture was chosen for analysis. Backpropagation algorithm was implemented with and without training. Tan-sigmoid function as in has been chosen in the hidden layer and purelin (linear) transfer function in the output layer. This is a useful structure for function approximation problem. Forecasting was tested to the neural networks based on designed network architecture described in Table-IV.
Although the ARIMA model generally produces good results, it has a drawback. It as- sumes a linear relationship between present and future values of load and between weather variables and load. Based on this consideration, we need a nonlinear approach to achieve better performance. Since ARIMA models can capture non-stationary and linear factors, and since neural networks can capture nonlinear effects, we have decided to combine these two approaches to produce a periodic nonlinear ARIMA (PNARIMA) neural network. This is a new approach to shorttermloadforecasting with neural networks, because it considers not only the autoregressive component, but also the moving average component (the addi- tion of the moving average component requires that the network be trained with dynamic backpropagation, which is not needed for the purely autoregressive neural networks that have been used in the past). In addition, the PNARIMA network uses periodic tapped de- lay lines to account for the seasonality in the load process, and it also includes differencing, which allows the model to handle the non-stationarity in the load process.
Electric loadforecasting is used by power companies to anticipate the amount of power needed to supply the demand. In the last few years, various techniques for the STLF have been proposed and applied to power systems. Conventional methods based on time series analysis exploit the inherent relationship between the present hour load, weather variables and the past hour load. Auto regressive (AR) and moving average (MA) and mixed Auto regressive moving average (ARMA) models  are prominent in the time series approach. The main disadvantage is that these models require complex modeling techniques and heavy computational effort to produce reasonably accurate results . Basically, most of statistical methods are based on linear analysis. Since the electric load is non linear function of its input features, the behavior of electric load signal can not completely be captured by the statistical methods. So statistical methods are not adaptive to rapid load variations. Another difficulty lies in estimating and adjusting the model parameters, which are estimated from historical data that may not reveal shorttermload pattern change . The emergence of artificial intelligence (AI) techniques has led to their application in STLF as expert system type models. These methods are discrete and logical in nature. By simply learning the historical samples, these methods can map the input-output relations and then can be used for the prediction.
Power system scheduling, load flow analysis, day to day operation and efficiency are some of very interesting field that can be explored by loadforecasting . The estimation of loaddemand is very significant as it will help the production and distribution of electric power. By under-estimating the loaddemand, it has a negative consequence on demand response and hence on power installation. Also, this under-estimation of load results difficulty to manage the overload conditions. Similarly, the over-estimation affects the installation and hence the efficiency of system . For the purpose of loadforecasting, several techniques have already been applied during the last few decades [1- 6].
systems, including a simple forecasting model , a grey-box forecasting approach , a lifting scheme combined with ARIMA models , and functional clustering combined with linear regression . An efficient forecasting approach to energy demandforecasting based on semiparametric regression smoothing was proposed by . Other approaches include a general fixed district heating model structure that can be adapted for any particular district heating system and used in cost-optimization studies , a forecasting method for space heating in a single- family houses , and nonparametric regression model . Whereas linear ARX models have been successfully applied in load-forecasting applications , nonlinear ARX models based on neural networks have also been proposed .
Loadforecasting plays an important role in power system planning and operation. Basic operating functions such as unit commitment, economic dispatch, fuel scheduling and unit maintenance, can be performed efficiently with an accurate forecast .Loadforecasting is however a difficult task. First, because the load series is complex and exhibits several levels of seasonality. Second, the load at a given hour is dependent not only on the load at the previous day, but also on the load at the same hour on the previous day and previous week, and because there are many important exogenous variables that must be considered . Various statistical forecasting techniques have been applied to shorttermloadforecasting (STLF). Examples of such methods including, Time Series, Similar-day approach, Regression methods and expert systems. In general, these methods are basically linear models and the load pattern is usually a nonlinear function of the exogenous variables . This final year project presents STLF with feed forward neural network algorithm to forecast future half hourly loaddemand for 24 hours or one week ahead with minimum error.
This project presents a study of short-term hourly loadforecasting using Artificial Neural Networks (ANNs). To demonstrate the effectiveness of the proposed approach, publicly available data from the Australian national electricity market (NEMMCO) web site has been taken to forecast the hourly load for the Victorian power system. We predicted the hourly loaddemand for a full week with a high degree of accuracy. Historical load data of 2006 obtained from the NEMMCO web site was divided into several where half of them are used for training and the other half is used for testing the ANN.
Almost all short-termforecasting techniques use as independent variables certain weather condition information such as temperature, humidity or wind speed. After many processes of trial and error for daily gas consumption prediction, since some variables like day type (i.e. working day and holiday), wet bulb temperature and gas price haven’t any effect on the network performance thus they are deleted from the model input for the simplicity of the network and 29 desired network inputs are considered. At the input, meteorological parameters (i.e. daily effective temperature, cloudiness, rain rate and wind velocity) and also the gas consumption for the previous five days are fixed. The meteorological parameters for the prediction day are also considered as the network input. However, at the output, the gas consumption rate for the prediction date is estimated [23,24].
The electrical peak consumption is about 15,000 MW during a working day. This demand is highly volatile on a day- to-day basis and is being significantly affected by weather conditions. The weekly pattern is comprised of the daily shapes are Monday, Weekday (Tuesday through Friday), Sunday and Holiday, reflecting the main working activities. Figure 1 shows the shapes of the typical electrical load curve during a week. Generally the load pattern on normal weekdays, when work is already in full swing, remains almost constant with small random variations from varying industrial activities, weather conditions etc., Figure. 2 shows load on Mondays and Weekday is different from that on other weekdays due to pick-up loads on mornings when all business and industries just start work, and evening loads in weekdays, because of its proximity to the weekend. The load pattern on Saturday and Sunday are different from rest of the weekdays. The peak load also takes a dip on Saturday and Sunday, which is a rest day for most of the people. The shape of the load curve on Sundays is similar to that on holidays. The peak load decreases considerably before and after major public holidays.
Electric loaddemand is a function of weather variables and human social activities, industrial activities as well as community developmental level to mention a few [2-7]. Statistical techniques and Expert system techniques have failed to adequately address this issue [2-10]. The daily operation and planning activities of an electric utility requires the prediction of electricity demand of its customers. In general, the required load forecasts can be categorized into short-term, mid-term, and long-term forecasts. The short-term forecasts refer to hourly prediction of the load for a lead time ranging from one hour to several days out. The mid-term forecasts can either be hourly or peak load forecasts for a forecast horizon of one to several months ahead. Scheduling of fuel purchases, load flow studies or contingency analysis, and planning for energy, while the long-term forecasts refer to forecasts made for one to several years in the future. The quality of short-term hourly load forecasts has a significant impact on the economic operation of the electric utility since decisions such as economic scheduling of generating capacity, transactions such as ATC (Available Transmission Capacity) are based on these forecasts and they have significant economic consequences.
The electricity supply industry requires to forecast electricity demand with lead times that range from the shortterm (a few minutes, hours, or days ahead) to the long term (up to 20 years ahead). Short-term forecasts, in particular, have become increasingly important since the rise of the competitive energy markets. Shorttermforecasting of electrical load is important for optimum operation planning of power generation facilities as it affects both system reliability and fuel consumption. Accurate forecasting of electricity and power demand determines the utility to match its generationcapabilities to the expected requirements. Electric utility operators for the purpose of scheduling and dispatching generating units need short-term forecasts. A short-term forecast is important for “unit commitment, economic dispatch,hydrothermal co-ordination, load management, etc.” Ackerman  says that a short-run forecast plays an important role in the day-to-day operations of a utility, and it is typically used for optimizing system operation and scheduling of hydro units and other peaking plants, such as gas turbines. The objective of the operators is to minimize variable costs without jeopardizing the electric system to power failures. The short-term (one to twenty-four hour) load forecast is of importance in the daily operations of the utility. With the emergence of Load Management, the short-termload forecast has a broader role in utility operations; it is also required for the co-ordination of Load Management programs with conventional system resources. Since the effectiveness of Load Management programs is sensitive to the system load, this additional function places higher accuracy requirements on the shortterm forecast, also required for shortterm maintenance scheduling. Accurate forecasting of power demand is to determine the assessment of the dynamic behavior of the system during disturbances so that the proper preventive action can be taken up. Loadforecasting is however a difficult task. First, because the load series is complex and exhibits several levels of seasonality: the load at a given hour is dependent not only on the load at theprevious hour, but also on the load at the same houron the previous day, and on the load at the samehour on the day with the same denomination in theprevious week. Secondly, there are many importantexogenous variables that must be considered,especially weather-related variables. It is relativelyeasy to get forecast with about 10 % mean absoluteerror; however, the cost of error are so high thatresearch could help to reducing it in a few percentpoints would be amply justified. Most forecasting