In , fuzzy rules were used to approximate continuous load profile on a compact set to good accuracy. Hidden nodes in this model represented the fuzzy rules which are a set of if-else statements defined by an expert during the Fuzzy Neural Network training . The authors then compared the forecasts of ANN with Fuzzy Networks and found that maximum percentage errors dropped from 4.08% to 1.65% . This forecasting technique is effective when the demand fluctuates a lot and when other models fail to capture the holiday or off season effect. The disadvantage of this technique is that the tuning of fuzzy rules is complex and an expert is required to handle that kind of task. In , a Support Vector Machine (SVM) model was used along with a stepwise algorithm for feature selection, to model the electricload demands. This modeling technique helped create an adaptive model with limited user interaction. The performance of the SVM model was compared with neural networks and was found to give better results due to its lower susceptibility to local minima and higher immunity to model complexity .
Short-term loadforecasting is important for performing many power utility functions, including generator unit commitment, hydro-thermal coordination, short-term maintenance, fuel allocation, power interchange, transaction evaluation, as well as network analysis functions, security and load flow studies, contingency planning, load shedding, and load security strategies .
Abstract—The modeling of the relationships between the power loads and the variables that influence the power loads especially in the abnormal days is the key point to improve the performance of short-term loadforecasting systems. To integrate the advantages of several forecasting models for improving the forecasting accuracy, based on data mining and artificial neural network techniques, an ensemble decision tree and FLANN combining short-term loadforecasting system is proposed to mainly settle the weather- sensitive factors’ influence on the power load. In the proposed strategy, an ensemble decision tree with abnormal pattern modification algorithm and a FLANN algorithm are used respectively to obtain the initial predicting results of the power loads first, a BP-based combination of the above two results are used to get a better prediction afterwards. Corresponding forecasting system is developed for practical use. The statistical analysis showed that the accuracy of the proposed short time loadforecasting of abnormal days has increased greatly. Meanwhile, the actual forecast results of Anhui Province’s electric power load have validated the effectiveness and the superiority of the system.
owadays electric power loadforecasting techniques such as neural network and fuzzy theory are now actively being used to reduce the uncertainty and the nonlinear behavior of load. Peakload without previously estimated, will influence the operation such as scheduling and allocation of units to back up and can disrupt the electric power system’s reliability. High accuracy of the loadforecasting could improve the security of the power system and reduces the costs [1, 2]. Fuzzy set was first introduced by L.A. Zadeh in 1965 to manipulate the unprobabilistic and uncertainty of data and information. Fuzzy sets and fuzzy logic are the basis of the fuzzy system aiming to mimic how human brain works in manipulating the non-exact information.
A study was conducted at a campus in Australia that utilized a fuzzy decision tree model (FDT) to forecast energy demand with solar panel present (Detyniecki et al., 2012). The group utilized the national weather forecast services to apply to its fuzzy decision tree model. Data was classified into 5 different categories of days and was collected over 77 consecutive days. The categories were fair, partly cloudy, mostly cloudy, cloudy, and showers. The team used the weather forecast at sunrise to predict the energy output of the solar panels for the entire day using the FDT. The FDT was composed of 38 decision paths with an average of 5.1 decision nodes per path and a max of 7 nodes on one path. The initial model was only 60% accurate in predicting what category the day would fall under, this was corrected using manual input for the day type. This process had an MAE of 12% from the baseline of energy produced each day present (Deyniecki et al., 2012). This type of model assumes that the solar panels will generate relatively close to the same amount of energy per type of day for each time that type of day occurs.
Electrical energy usages continue to increase from year to year. Number of customers Indonesia State Electricity Company (ISEC) from 2011 to 2015 has increased more than 33 percent. Based on the increasing number of customers, ISEC is required to meet all the demand for electrical energy continuously. Continuous distribution of electrical energy is the right of ISEC customers to be prioritized by ISEC as the main provider of electrical energy in Indonesia –. To be able to meet the needs of electrical energy continuously, it is a necessity to balance supply and demand sides. Thus, the power generated must always be equal to the power consumed by the electric power consumer –. Therefore, it is also necessary to estimate short-term electrical loads that can predict the need for the electrical usage. The short-term electrical load forecast (SELF) aims to predict electricity requirements in minutes, hours, days, and weekly hours , , , –. The SELF-brings out to forecast load requirements at a load center switchyard (LCS) and play an important role in the real-time control and security functions of the energy management. An accurate estimation of the electrical load can save operational costs and safe conditions, both can be done by the supply and demand side
hand industrial and agricultural loads are highly inductive and start up and shut down of such type of load induce huge spikes to the load curve. These spikes are called the random disturbance because start up and shut down of these huge loads is quite random in nature and there is no way to predict the occurrence of these spikes. CNG station load also lies in this category. If we add these spikes in the training data of ANN model then the average error of the model becomes very high. Special events such as religious or cultural celebrations also are another source of random disturbance. Diwali, Eid day, Christmas and other religious events .
This paper proposes a novel method of adaptive ARIMA compared with conventional adaptive ARIMA. The experimental results show the advanced performance of proposed method. Because there is little fluctuation in daily electricload during the same quarter, selecting fixed observation for each quarter is acceptable. Moreover, the parameters of ARIMA model could be updated with the forecasting of each points.
Time factor: Time factor in case of STLF is most concerned thing for precise loadforecasting because SLTF is done on hourly basis. A load demand curve is published in a report by Sri Lanka Government  as shown in Figure (1).Curve showing the peak demand at 18HRS. So a uniform analysis for loadforecasting is not enough. Close monitoring of load in hourly basis will give good forecast. Also load at same time in summer and winter varies with a large margin. Certain changes in the load pattern occur gradually in response to seasonal variations such as the number of daylight hours and the changes in temperature. Figure (2) shows the system peak occurs with a steep increase from 18:00 to 19:00 and depreciates with a slow rate of decreasing which takes about 3 hours. (From 19:00 to 22:00) This feature is common for all three curves. General opinion on the night peak is that it is predominately governed by domestic activities and lighting. Morning peak of weekdays (recorded at 06:00) is rather symmetrical, which consists with rapid increase and a rapid decrease. However, when considering only Sundays, the curve does not show a significant peak demand, but just a slight increase, which is recorded at 06:30.
Utilization of electrical energy should be set as well. This setting starts from generation planning and load sharing, operation, setup and other matters . Planning is done on the load side for generation planning. This plan is also known as forecasting. The electrical loadforecasting is divided into three, i.e. long-term loadforecasting, short- term loadforecasting and extremely short-term loadforecasting [6, 7].
Forecasting ATM cash demands is a challenging research task. When the forecasting results are too high compared to the real demand, this will cause excessive cash at bank’s ATMs and the cost of lost interest. On the other hand, if the forecast is too low, this will result in dissatisfaction of bank customers because of cash-outs. Although recent studies focused on new computational intelligence techniques for cash demand forecasting, this paper advocates the enhancement of the dataset to improve the prediction performance of forecasting models. In this study, 19 special days in the UK have been considered and NN5 competition dataset, which includes 735 daily withdrawal amounts from 111 ATMs in UK, was updated with these calendar days. After preprocessing step and application of exponential smoothing method, we achieved 21.57 % average SMAPE for 56 daysforecasting horizon. This study shows that good forecasting results can be reached by improving the data even if we do not apply complex computational intelligence techniques.
Electricloadforecasting 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 electricload is non linear function of its input features, the behavior of electricload 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 short term load 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.
nomial, and exponential power). These models are ge- neric and can be used in medium-term loadforecasting for any power system. Results showed that the perform- ance of the linear and polynomial models perform was close, when applied to the hourly loads of the Jordanian power system for different years. The exponential power model performs close to the linear model, however, due to its more complex nature; it is only applied to a time span of one year. The incurred forecasting errors for the investigated three models is about 10% while the abso- lute error (APE) ranges between 4.6% (exponential)-to 6.7% (polynomial).
Over the years many techniques have been already implemented for LoadForecasting. Some of them are: artificial neural networks (ANN) , Fuzzy Logic Approach , Bayesian Network Approach , various hybrid approaches, and many others, including classical statistical approaches like multiple liner regression  and automatic regressive moving average (ARMA) , Autoregressive Integrated Moving Average (ARIMA) . During the last decade or so, neural network approaches combined with other methods (such as evolutionary or fuzzy methods) are most frequently used , . However, there are certain anomalies in prediction when neural network is applied on “real time” datasets, mostly because of the two effects related to neural networks called “overfitting”  (i.e., when a model describes random error or noise instead of the underlying relationship) and “curse of dimensionality” (problem caused by the exponential increase in complexity associated with adding extra dimensions to the ANN input). In such conditions the forecasting system can yield poor results. To eliminate above mentioned short comings, we have taken an approach based on Interval type-II Fuzzy inference system for electrical load prediction.
service class load. When the average annual growth is calculated, commercial peak demand of approximately 5.7% per year outpaced commercial consumption or load growth of 4.9% per year during the 1990s. In the meantime, industrial peakload growth was similar to the sector’s growth in consumption, both at around 2.5% per year. While industrial consumption grew by the smallest percentage of the major sectors within Utah, it still grew at a higher rate than nationally. Manufacturing production grew for most of the 1990s, but it began to shrink in 1999 and continues to do so. Figure 26 compares manufacturing and transportation/warehousing employment and income to the retail sector for the Salt Lake MSA. Manufacturing and warehousing firms comprise the majority of industrial customers in Salt Lake City, while retail represents a portion of commercial customers. These indicators help show the relative importance of these sectors to the economy and how each has grown or declined in relation to the other. Retail employment surpassed manufacturing employment in 1996. It continues to grow as a portion of overall employment, while manufacturing declines. Transportation and warehousing, the other category of industrial users, has held steady since 1999, with 6.5% of the jobs in the Metro Area. Jobs cannot be looked at alone, since manufacturing has been increasing mechanization, and as a result, the amount of electricity used. Retail, however, remains dependent on labor. The second graph in Figure 26 attempts to account for these differences by looking at the amount of income each sector brings to the Metro Area. As the graph shows, while manufacturing income has been steadily declining since 1990, it makes up a larger portion of personal income than does retail, while transportation and warehousing remained flat.
Abstract- In a remote areas supply of energy from national grid is insufficient for a sustainable development. Integration and optimization of local alternative renewable energy sources is an optional solution of the problem. The needs of rural electricity is met by conventional approaches is limited. In economic perspective, non-conventional forms of rural electrification may least-cost, particularly where villages are some distance from each others. In this paper the renewable energy reduces the burden on electricity supply shortfalls and the urgency of costly grid extension. In this paper we will bring in some new concepts associates with renewable energy forecasting.
Although decomposition methods were not developed for the primary purpose of prediction, their entrance and application is very appealing. In Theodosiou (2010) “disaggregating the various components in the data and predicting each one individually can be viewed as a process of isolating smaller parts of the overall process which are governed by a strong and persistent element, and therefore separating them from ‘noise’ and inconsistent variability”. And by this we can learn more from these processes and possibly obtain more accurate forecast.
In this project, the load data in 3 month will be used in training and load data for the next 3 month will be used in testing. The input load data will be load on Saturday, Monday, Wednesday, and Friday. Meanwhile, the target load data will be load on Sunday, Tuesday, Thursday, and Saturday. The parameter that used in this project are learning rate, momentum rate, maximum number of iteration (epoch) and the minimum error goal. The training performance must be 100 percent. If the 100 percent performance was not meet, then the parameter will be adjusted to get the 100 percent performance during training.
The electric power demand of Universiti Tun Hussein Onn Malaysia (UTHM) has steadily increased in the past seven years. This trend is certain to continue in future. The electrical load is the power that an electrical utility needs to supply in order to meet the demands of its customers. Electricity loadforecasting is thus an important topic, since accurate forecasts can avoid wasting energy and prevent system failure . It is no need to generate power above a certain predictable level and the latter when normal operation is unable to withstand a predictable heavy load. For example, short-term planning of electricity load generation allows the determination of which devices shall operate and which shall not in a given period, in order to achieve the demanded load at the lowest cost. It also helps to schedule generator maintenance routines. The system operator is responsible for the scheduling and aims foremost at balancing power production and demand. After this requirement is satisfied, it aims at minimizing production costs, including those of starting and stopping power generating devices, taking into account technical restrictions of electricity centrals. Finally, there must always be a production surplus, so that local failures do not affect dramatically the whole system .