Mustafa Akpinar

Forecasting natural gas consumption with hybrid neural networks — Artificial bee colony

Natural gas distribution companies have different consumer types including manufacturing industry, organized industrial zones, food and beverage industry, household and other low consuming enterprises, etc. Leading two categories of these consumers are household and low consuming enterprises as they have high consumption in winter whereas low in summer. The paper studies consumption demand forecasting for certain consumption group using artificial neural network (ANN). Prepared consumption data is divided into two groups. First three years daily consumption data is kept for training while forth year data is kept for testing. For consumption forecasting its own historical data is used. The research is completed by applying two different model types having eleven different sub-models each. Sub-models have different numbers of neurons and three hidden layers at most. Estimations are done with twenty-two different scenarios in total. In two distinct models, ANN weights are trained with backpropagation (BP) and artificial bee colony (ABC) algorithms. After training stage, network structures are tested by test datasets. As a result, it is concluded that ABC model with two hidden layered scenarios gives better results in demand forecasting than the others.

Forecasting household natural gas consumption with ARIMA model: A case study of removing cycle

Forecasting natural gas consumption in Turkey is very important at energy sector. For this purpose kindly prediction methods are used. In this study autoregressive integrated moving average (ARIMA) method is used and main idea in this study is removing cycling component in time series. For removing cycling, time series divided monthly data and merged co-exhibiting behavior months. Same months and different years data is merged and called as “Model” and 6 Models are prepared. Last model; Model 7 is a general model that includes all consumption data. ARIMA models are applied and mean absolute percent errors (MAPE) are found. Selected minimum MAPE and values of (p, d, q) predictions for Models. For 2012, predictive values of models and Model 7 are compared with actual consumptions. Model that removed cycling (Merged Model) 2.2% better than Model 7.

Day-ahead natural gas forecasting using nonseasonal exponential smoothing methods

Natural gas is one of the most commonly used energy sources. In real life, natural gas consumption values and the amount of natural gas extracted are required to be equal. Thus, problems with respect to supply and demand are reduced. Problems in the supply side arise from the fact that the demand can not be determined correctly. Therefore, the imbalance in the system should be reduced by correctly determining the demand. In this study, day ahead demand forecast for the natural gas sector is examined. In the day ahead approach, demand estimations are performed using over four years of daily data and applying simple, double, linear, damped trend exponential smoothing methods at different data sizes. The effect of using different sizes of dataset on the demand estimation is tried to be identified. While the results showed that the simple exponential smoothing method gave the best result, the estimations made with the 6-week and extended datasets forecasted more accurate results. In addition, it is observed that the increase in the number of data in the day ahead demand forecast, allows prediction where exponential smoothing methods are used, with a lower error. In this research, the lowest mean absolute percent error (MAPE) for four years is determined as 14.1%, while the coefficient of determination (R2) is 0.917 with the SES method.

Estimating household natural gas consumption with multiple regression: Effect of cycle

Estimating natural gas consumption is important at energy sector. With this purpose different prediction methods are used. One of these methods is multiple linear regression method which is used in a wide range of applications In this study, multiple linear regression method was used for estimation of gas consumption and effects of cycle were examined. To examine the cycle, the data were divided into 6 pieces. Each piece of data was called as “Model”. The data which is included in these models were gathered from the same months of successive years. In addition a new model named, Model 7 was created by using all the data. Multicollinearity was removed when creating the models. Multiple linear regression method was applied to the models and the gathered results were compared by using mean absolute percent error. The Merged Model which was created by using predictive values of models for 2012 and Model 7 were compared. Mean absolute percent errors of Merged Model and Model 7 were 14.38% and 55.10% respectively.

Determination of the Gas Density in Binary Gas Mixtures Using Multivariate Data Analysis

Some solvents in commercial products may have harmful effects on human health. It is important to determine the percentage of this certain solvent in a product to detect any possible health hazards. In this paper, three different solvents, acetone, methanol, and chloroform, are used to form binary gas mixtures in a laboratory environment. Nine quartz-crystal microbalance sensors are used, and gas data are obtained through the responses of these sensors. First, the data set divided 11 times randomly for validation sensitivity of the results. For each of the binary gas mixtures, insignificant sensors are removed, considering multivariate analysis of variance analysis, and sensor data sets are obtained. The statistical multivariate linear regression (MvLR) method is used to determine the ratio of individual gasses in each binary gas mixture. Flexible models are created by removing insignificant sensor data from the equations in the MvLR. Prediction performances of 11 data sets reveal and validate that statistical methods can be used to detect the ratio of a certain gas within a gas mixture, and reliable results can be achieved.

Time series forecasting using artificial bee colony based neural networks

Artificial neural networks (ANN) are among the nonlinear prediction techniques popular in the last two decades. Recent studies show that ANN can be modeled with different training techniques. ANN is usually trained by the backpropagation method (BP). In this study, ANN structures were trained by using artificial bee colony algorithm (ABC) and, weight and bias values were tried to be determined. ABC training (ANN-ABC) was tested over three different datasets and compared with the BP training (ANN-BP) results. In addition to use ABC in modeling, different error types such as mean square error (MSE), mean absolute percent error (MAPE) and adjusted coefficient of determination (R) have been used in the training. The results on popular time series datasets have shown that ABC based ANN training yields successful results in forecasting.

Carrier-grade NAT — is it really secure for customers? A test on a Turkish service provider

Dramatic rise in the user amount yields increase in the number of internet accessed devices within the last decade. Since most of the devices have internet connection, IPv4 space becomes inadequate. In order to avoid this situation, internet service providers focus on using their IPs within their IP pool, optimally. The most preferred approach to handle this problem is called Carrier Grade Network Address Translation (CGN). In this technique, a city, a neighborhood or a group of users could be configured as if they are in the same Local Area Network (LAN) and they have IPv4 Network Address Translation (NAT) connections for Wide Area Network (WAN) accesses. By applying this approach, IP costs are reduced and number of IPs in the pool is optimized. However, implementations done in recent systems could cause vulnerabilities as well. This work focuses on examining a part of CGN applied network that acts as LAN, by scanning, exploring users, devices and vulnerabilities for a specific neighborhood in Turkey. Users and devices are determined and since they are considered in the same LAN most of them are easily gained access and the insecurity of the system is proved. Also it is also observed that a user could stop or slow down the traffic by Denial of Service (DoS) or Distributed DoS attacks.