Elsevier

Energy

Volume 180, 1 August 2019, Pages 149-162
Energy

A method for natural gas forecasting and preliminary allocation based on unique standard natural gas consumption profiles

https://doi.org/10.1016/j.energy.2019.05.084Get rights and content

Highlights

  • Unique standard gas consumption profiles are developed.

  • A method for usage of the profiles for preliminary allocation is developed.

  • We provide a way to fairly and in a standard way allocate gas consumption.

  • The data needed to set up the consumption profiles is attached.

Abstract

The paper reports on the development of unique standard gas consumption profiles for the end gas consumers and the preparation of a method for the implementation of the developed profiles for forecasting and preliminary gas allocation. Four years of gas consumption and temperature measurements were used to develop eight types of consumption profiles for 17 gas consumer groups, which were grouped according to their professional activity. As an alternative to the exponential, Gompertz or logistic model functions, frequently used in gas consumption model developments, the sigmoid model function is implemented and model constants for the eight types of profiles are developed based on the knowledge of the temperature independent portion of the gas consumption and separate treatment of workdays/weekends. Based on these profiles, a method was developed for the preliminary allocation of the gas consumption. The developed profiles and the gas consumption allocation method were validated on the available set of gas consumption data for the Slovenian gas market, proving the sigmoid model function based gas consumption allocation as an accurate and viable means of gas consumption forecasting.

Introduction

Natural gas is one of the main energy sources used in modern society. As the majority of natural gas is supplied to the end consumers through a pipeline based distribution system, there is a need to forecast the gas demands in the future as accurately as possible. In the case of a pipeline system, the main challenge is to balance the supply and demands for the end consumers on a daily basis, with the need to forecast the gas consumption within a day and for the next day accurately. In addition to forecasting, temporary allocation of consumed gas among consumers is an important part of the daily operation of the gas transmission system. (see Fig. 12)

In order to predict the behaviour of a large number of energy consumers, several methods have been used in the past. Statistical approaches were employed to analyse gas consumption quite some time ago by Herbert [1]. More recently, several algorithms were used to make forecasts, such as the grey model, statistical models, econometric models, neural network models, genetic algorithms, mathematical models, and their combinations [2]. The Hubbert curve model has also been extensively used for supply forecasts.

Zeng and Li [3] used a self-adapting intelligent grey model for forecasting the natural gas demand in China. They focus on annual estimates and predict China's gas demands until 2020. Ma and Liu [4] also predicted the growth of annual natural gas consumption in China until 2020. They showed, that predicting the behaviour of very large groups of consumers using the grey model, is accurate. In contrast, prediction of gas used by consumers of a single gas supplier for a single day, which is the purpose of this paper, is more challenging. Similarly, natural gas consumption was studied by Xie and Li [5], who proposed the usage of the grey model in combination with a genetic algorithm for prediction of annual gas consumption totals. Also, Azadeh et al. [6] proposed an adaptive network-based fuzzy inference system-stochastic frontier analysis algorithm for long-term natural gas consumption forecasting and applied it to Bahrain, Saudi Arabia, Syria, and the UAE.

Wadud et al. [7] used a econometric model to study natural gas demand in Bangladesh. They include gas price, GDP and number of consumers to fit a relationship giving an estimate of total gas usage. Such a method also works for large scale only, and it is useful for developing energy polices and not appropriate for day-to-day allocation and prediction of gas demands. Neural networks were used by Farzaneh-Gord and Rahbari [8] to perform unsteady simulations of response of natural gas distribution pipeline networks to ambient temperature variation. They are able to estimate daily gas demands. They successfully demonstrate that neural networks are capable of predicting the response of gas users to temperature variations. In this work, we develop a similar algorithm, based of fitting a sigmoid curve. The main advantage of our approach is it simplicity, which makes it easy for gas supply companies to implement in their work flow, after the coefficients, presented in the appendix of the paper, are published within the legislative framework of a country they operate in.

Several other researchers focused on China. They follow the same agenda, obtain measurements of gas usage for a chosen country and implement a numerical method to forecast future usage. A wide rage of methods is used, for example, Zeng [9] used the grey model for modelling of natural gas demand in China. Similarly, Shaikh et al. [10] used optimized nonlinear grey models for forecasting China's natural gas demand. Rui et al. [11] used a genetic algorithm to improve the least squares approach and study natural gas consumption in China. Shaikh et al. [12] also studied the gas demand in China using logistic modelling analysis, while Zhang and Yang [13] used the Bayesian model averaging. A model for short-term natural gas prediction using support vector regression was presented by Zhu et al. [14]. Bai and Li [15] proposed a structure-calibrated support vector regression approach to forecast the daily natural gas consumption. Zhang et al. [16] considered the impact of natural gas supply on infrastructure development in China. Some of this studies optimise their model for predicting annual averages, some focus on short term forecasting.

In this paper, we develop a model, which is aimed at day-ahead and within-day forecasting of gas demand. When implemented on national scale, it will enable a fair preliminary allocation of gas consumption between gas supply companies operating in the same market. Day-ahead forecasting was the focus of research conducted by Panapakidis and Dagoumas [17] who proposed a model using a combination of wavelet transform, genetic algorithm and neural network techniques. They confirm that accurate forecasts of natural gas demand can be essential for utilities, energy traders, regulatory authorities and decision makers. They worked on data from Greece and predict gas usage at distribution points. In our work, presented in this paper, we make predictions at consumer level, giving gas supply companies a possibility to predicts the behaviour of all customers in their portfolio. Short term forecasting was the focus of Yu and Xu [18], who used a combination of optimized genetic algorithm and neural networks to develop a short-term load forecasting model of natural gas. Similarly, the use of genetic algorithms to predict short-term usage of natural gas in houses was proposed by Aras [19], however they focused on monthly averages only. The split the gas consumption to temperature dependent and independent parts. This approach is also taken in this paper.

Natural gas demand in Turkey was studied by Erdogdu [20]. Their objective was to estimate short and long run price and income elasticities of sectoral natural gas demand in Turkey and to forecast future growth in this demand using ARIMA modelling and compare the results with official projections Fagiani et al. [21] reviewed several forecasting techniques aimed at developing smart natural gas and water grids.

Pelikan and Simunek [22] used a genetic algorithm as an optimising tool in risk management of the natural gas consumption. Karadede et al. [23] developed a breeder hybrid algorithm for natural gas demand forecasting and used it in Turkey to forecast natural gas demand from 2001 to 2014. They claim that the breeder hybrid model is superior to other models and can be used as general natural gas demand forecasting tool with daily, monthly and annually data with error close to zero.

Apart from natural gas forecasting, solar [24] and wind energy [25] demands have been studied extensively. Chen et al. [26] proposed a novel method based on nonlinear-learning ensemble of deep learning time series prediction based on long short term memory neural networks, support vector regression machine and extremal optimization algorithms. Thaler et al. [27] have developed an empirical model for prediction of energy consumption in a distribution system. They used a normalised radial basis function neural network to find an economically optimal energy distribution. Day-ahead forecasting of the electricity market was studied by Koltsaklis et al. [28]. Li et al. [29] studied the relationship between gas demand and electricity production for gas-to-power systems. Qiao et al. [30] built a comprehensive system model of a natural gas and electricity coupled network. Oliver et al. [31] developed a method for peak-day gas consumption for gas transmission system operators. Baldacci et al. [32] used natural gas forecasting in order to detect anomalies on the gas distribution network.

An econometric model for forecasting both the short and long-term dynamics of natural gas consumption in Pakistan was used by Khan [33]. They provide estimate until 2020. Taspinar et al. [34] used artificial neural networks to forecast natural gas consumption based on four years of data in Turkey. The ant colony algorithm was employed by Toksari et al. [35]. A degree-day concept was used by Gumrah et al. [36] to study the gas demands of Ankara city. Ervural et al. [37] also considered gas demand forecasting in Turkey using autoregressive moving average, while Akpinar et al. [38] used ABC-based neural networks and the sliding window approach for day-ahead natural gas demand forecasting. In comparison to using the sigmoid model, as proposed in this paper, the authors admit that the coding of the neural network algorithm is difficult, however claim that the use for companies should be easy. They stress that the decision makers can use the natural gas demand forecasting results obtained from forecasting models as decision support systems. In this work, we extent this hypothesis and claim that an well defined and easy implementable model can serve as a basis for preliminary allocation of gas consumption between supply companies on a national level.

Nonlinear programming and genetic algorithms were employed in Iran by Forouzanfar et al. [39] to forecast natural gas usage. They introduce the sigmoid curve, which is a special case of the logistic equation, as a good model for many different contexts such as demographics, T biology, economics, etc. The curves are used because of their ability to describe these processes and display typical phases of, among others, gas consumption: the low gas usage in summer, the logarithmic growth in the winter months and saturation when extreme cold limits gas consumption. In this work we also employ the sigmoid curve to model gas consumption.

In Poland, Siemek et al. [40] have used an adaptation of the Hubbert model to derive a model of future gas demand. They used the Newton-Gauss algorithm to determine the model constant. In our work, we use the Levenberg-Marquardt algorithm for the same task. Liu et al. [41] used a time-series approach to model gas consumption in Taiwan. Vondracek et al. [42] used a nonlinear regression model with individual customer effect, typical time-dynamics part and the temperature correction for natural gas usage estimation in the Czech Republic. Sabo et al. [43] investigated the natural gas usage in Croatia on the basis of hourly temperature and gas usage measurements. A review paper on the topic of forecasting natural gas consumption was prepared by Soldo [2], who summarised the approaches used by researchers based on the forecasting area (world, national, individual consumer), forecasting horizon (hourly, daily, monthly), gas data measurements used and the model applied. An overview of forecasting methods for energy demand was prepared by Ghalehkhondabi et al. [44]. Potočnik et al. [45] investigated risks associated with forecasting models and exposed the Slovenian economic incentive model that motivates natural gas distributors to forecast their future consumption. Potočink and Govekar [46] presented practical results of forecasting for the natural gas market where they stress the importance of incorporating the proper influential variables into the model, and by properly understanding the underlying principles of energy consumption. Potočnik et al. [47] also considered short-term residential natural gas forecasting in Croatia. The same research group studied the influence of solar radiation on the gas forecasting models ([48]). Artificial neural networks combined with the Levenberg-Marquardt training algorithm were used to produce short-term natural gas consumption forecasts in Serbia by Ivezic [49]. Gascon and Sancez-Ubeda [50] used generalised additive models for short-term natural gas demand forecasting. Aguilera and Ripple [51] employed the Variable Shape Distribution model to estimate the total endowment of conventional gas in Asia Pacific. Azadeh et al. [52] propose to use an integrated emotional learning fuzzy inference approach for optimum training of forecasting models for natural gas demand.

Apart from modelling gas consumption at the consumer level, models have been prepared that attempt modelling of the global natural gas supply (Crow et al. [53]). These types of models represent the natural gas market as a system of nodes and connections with prices and flows, and seek an equilibrium solution in the whole network (Busch [54]). Several mathematical approaches are utilised: Linear programming, nonlinear programming, or agent based economic modelling. Bianco et al. explore the relationship between the residential [55] and nonresidential [56] gas consumption and the Gross Domestic Product of Italy. They find strong correlation between the two. A similar result was reported by Apergis et al. [57], who considered economic growth in 67 countries, and by Karasalihovic et al. [58] for Croatia. Chavez-Rodriguez et al. [59] modelled long-term natural gas dynamics in the south of Latin America. They employed a combination of a simulation model LEAP (Long range Energy Alternatives Planning System), and the TIMES model was used to optimise the natural gas supply. Horschig et al. [60] prepared a biomethane market simulation model, which was used in Germany to explore the future relationship between the natural gas market and the biomethane market. In order to find the most efficient way of distributing gas, optimization of transmission networks has been considered as well. Uster et al. [61] developed an integrated large-scale mixed-integer nonlinear optimization model for design and operation of natural gas transmission networks. Rios-Mercado et al. [62] give a state-of-the-art review of optimization problems in natural gas transportation systems. On a smaller scale, Lo Cascio et al. [63] developed a multi-objective optimization model for urban integrated electrical, thermal and gas grids.

In this paper, we propose a method, which, when coupled with a mathematical forecasting model, enables gas transmission system operators to forecast or preliminarily allocate consumed gas fairly and uniquely. The development of the proposed model was initiated by the initiative of The Energy Agency of the Republic of Slovenia, with a clear goal to derive a model that would be accurate, yet easy to use, whether used by the gas distributors for forecasting the consumers demands or system operators for balancing purposes, and therefore would be widely accepted when set into the legislative framework, that had to be used by all members in the Slovenian gas market. The developed model and its method of use is simple to implement, and can be applied at the level of balance group leaders, members of a balance group, the operator of the gas market, the operator of the transport system, as well as at distribution system operators. The proposed method is not geographically restricted and can also be implemented in other countries, if the modelling assumptions are met. When adopted at a national or a regional level, it may serve as a legislative tool, which enables fair preliminary allocation of consumed gas and helps to avoid conflict between balance group leaders or transmission system operators. The underlying mathematical forecasting model should, naturally, be prepared separately for each region, since specific local, regional and climate characteristics must be taken into account.

The consumption of natural gas in a specific time period depends on many factors (for example temperature of the surroundings, other environmental elements, the purpose of usage, etc). Certain groups of users exhibit similar usage characteristics, which make it possible to predict the future usage based on the known or estimated environmental parameters. Such estimates are expected to work well only when used for a large number of users, so those single individual users who do not follow the statistical behaviour of the group completely, do not influence the estimate to a large extent.

In order to make it possible for the natural gas suppliers to have a unified, fair, and dependable model for estimating the natural gas usage, the Unique Standard Consumption Profiles (USCP) for individual groups of users are developed in this paper. The USCPs are developed in the form of a mathematical model, which includes several model constants which are fitted to the usage data of individual gas users. In this paper, we present the development of gas consumption profiles based on the influence of the outside temperature on the natural gas usage. The developed USCP can be used in combination with a method for preliminary allocation of consumed gas, which gives a unique and fair way of preliminary allocation. Gas usage data was obtained from measuring consumption of end users from individual characteristic groups in the time frame of 2009–2013, provided by the Energy Agency of the Republic of Slovenia. The USCPs in this work are prepared using the sigmoid model function, which presents an alternative to the exponential, Gompertz (Gutierrez et al. [64]), linear (Spoladore et al. [65]), support vector regression (Bai et al. [15]), and logistic model functions (Melikoglu [66]) used by other authors (Sabo et al. [43]).

The basic requirement for the models and methods prepared in this paper is to provide a simple algorithm for allocation of gas between gas suppliers on a short-term scale, i.e. within a day, or for the next day (Cui et al. [67]) based on forecasted climate conditions. The models and method are meant to be published in a legislative framework, and serve as a fair and standard method of performing preliminary allocation of gas consumption among gas suppliers. Since official temperature forecast is available, and since the number of gas consumers to which the model will be applied is expected to be large (i.e. all consumers within a country/region), we chose the sigmoid regression model. It is simple, quick and works well when used to model a large number of consumers. It can be defined uniquely by publishing only four parameters. We were able to establish, that the sigmoid model performs better than other regression models. Furthermore, as time passes and new measurements become available, the techniques developed in this paper can be used to update the forecasting models and, thus, provide a fairer and more accurate allocation of consumed gas among gas suppliers.

The developed method for preliminary gas allocation is applicable worldwide regardless of the forecasting models. The gas consumption forecast models we present in this paper have been developed based on four year long measurements of gas consumers all around the country. Slovenia has very diverse climate conditions (the Mediterranean in the south-west, alpine in the north, and continental in the east). Thus, the developed models (published in the Appendix), can be used in similar climate conditions around the world. Alternatively, if a region possesses a several years long gas consumption measurement dataset, only the method and the model derivation can be employed to recalibrate the model parameters for a specific region. The developed preliminary allocation method, in combination with the presented forecast model, can be used immediately after the adoption into the legislative framework. Implementation within either a spreadsheet software or a website is trivial, and can be done with minimal effort. The method requires data on average annual gas consumption of each consumer. This dataset (or at least an estimate) is usually readily available at gas providers.

The original contribution of this paper is twofold. Firstly, we present a method for preliminary (day-ahead) and within-day gas consumption allocation, which enables a fair and standardised distribution among gas suppliers. The developed method is applicable worldwide, and is not limited to a single region. Secondly, we propose the use of a sigmoid model function (as opposed to the exponential, Gompertz or logistic model functions proposed by other authors) in support of the allocation method.

The paper is organised as follows. Section 2 presents the measurements of gas consumption and temperature. Section 3 describes the development of consumption profiles and the method for their use. Section 4 the presents the validation of the developed profiles and the analysis of their applicability and accuracy. The paper ends with the conclusions and an acknowledgement.

Section snippets

The experimental dataset

The Energy Agency of the Republic of Slovenia performed hourly gas consumption measurements at 260 consumers (end users) in Slovenia. The measurements were taken during the period 1.9.2009–31.5.2013. At the same time, 18 meteorological stations recorded climate conditions.

Unique standard consumption profiles

In this section development of unique standard usage profiles, forecasting method and preliminary allocation method are described.

Results and discussion

Accurate predictions of the daily consumption for the distribution system on the national level could be obtained by the adoption of the proposed preliminary allocation method through each of the gas market operators, providing that the latter have a detailed database of their users within the consumer groups. The accuracy of these predictions depends strongly on the correct use of the developed USCPs. In the following, the analysis of the performance of the forecasts made by USCPs is made, and

Conclusions

Unique standard gas consumption profiles were developed for several Consumer Groups based on the measurement of gas consumption and air temperature. The sigmoid model function was implemented as the basis for the consumption profiles’ derivation. Eight different types of consumption profiles were developed, based on the separate treatment of workdays/weekends, and based on a priori knowledge of the temperature dependent portion of the gas consumption. Furthermore, a method was developed which

Acknowledgements

The authors would like to express their gratitude to The Energy Agency of the Republic of Slovenia for the data and financial support in the framework of the Contract No. 2525/201314/336.

References (69)

  • Y. Bai et al.

    Daily natural gas consumption forecasting based on a structure-calibrated support vector regression approach

    Energy Build

    (2016)
  • Q. Zhang et al.

    Study on the impacts of natural gas supply cost on gas flow and infrastructure deployment in China

    Appl Energy

    (2016)
  • I.P. Panapakidis et al.

    Day-ahead natural gas demand forecasting based on the combination of wavelet transform and ANFIS/genetic algorithm/neural network model

    Energy

    (2017)
  • F. Yu et al.

    A short-term load forecasting model of natural gas based on optimized genetic algorithm and improved BP neural network

    Appl Energy

    (2014)
  • E. Erdogdu

    Natural gas demand in Turkey

    Appl Energy

    (2010)
  • M. Fagiani et al.

    A review of datasets and load forecasting techniques for smart natural gas and water grids: analysis and experiments

    Neurocomputing

    (2015)
  • Y. Karadede et al.

    Breeder hybrid algorithm approach for natural gas demand forecasting model

    Energy

    (2017)
  • S. Sobri et al.

    Solar photovoltaic generation forecasting methods: a review

    Energy Convers Manag

    (2018)
  • H. Liu et al.

    Big multi-step wind speed forecasting model based on secondary decomposition, ensemble method and error correction algorithm

    Energy Convers Manag

    (2018)
  • J. Chen et al.

    Wind speed forecasting using nonlinear-learning ensemble of deep learning time series prediction and extremal optimization

    Energy Convers Manag

    (2018)
  • M. Thaler et al.

    Prediction of energy consumption and risk of excess demand in a distribution system

    Phys Stat Mech Appl

    (2005)
  • N.E. Koltsaklis et al.

    A mid-term, market-based power systems planning model

    Appl Energy

    (2016)
  • G. Li et al.

    Security-constrained bi-level economic dispatch model for integrated natural gas and electricity systems considering wind power and power-to-gas process

    Appl Energy

    (2017)
  • Z. Qiao et al.

    An interval gas flow analysis in natural gas and electricity coupled networks considering the uncertainty of wind power

    Appl Energy

    (2017)
  • R. Oliver et al.

    Forecasting peak-day consumption for year-ahead management of natural gas networks

    Util Pol

    (2017)
  • L. Baldacci et al.

    Natural gas consumption forecasting for anomaly detection

    Expert Syst Appl

    (2016)
  • M.A. Khan

    Modelling and forecasting the demand for natural gas in Pakistan

    Renew Sustain Energy Rev

    (2015)
  • F. Taspinar et al.

    Forecasting of daily natural gas consumption on regional basis in Turkey using various computational methods

    Energy Build

    (2013)
  • B.C. Ervural et al.

    Model estimation of ARMA using genetic algorithms: a case study of forecasting natural gas consumption

    Procedia Soc Behav Sci

    (2016)
  • M. Forouzanfar et al.

    Modeling and estimation of the natural gas consumption for residential and commercial sectors in Iran

    Appl Energy

    (2010)
  • J. Siemek et al.

    Estimation of natural-gas consumption in Poland based on the logistic-curve interpretation

    Appl Energy

    (2003)
  • L.-M. Liu et al.

    Forecasting residential consumption of natural gas using monthly and quarterly time series

    Int J Forecast

    (1991)
  • J. Vondráček et al.

    A statistical model for the estimation of natural gas consumption

    Appl Energy

    (2008)
  • K. Sabo et al.

    Mathematical models of natural gas consumption

    Energy Convers Manag

    (2011)
  • Cited by (21)

    • A sigmoid regression and artificial neural network models for day-ahead natural gas usage forecasting

      2021, Cleaner and Responsible Consumption
      Citation Excerpt :

      Karasalihovic et al. (2003) investigated the Croatian gas market and found a strong correlation with the Gross Domestic Product. Ravnik and Hriberšek (2019) proposed a methodology for short-term forecasts in Slovenia based on the sigmoid-type model. Hribar et al. (2019) focused on the gas use of the Slovenian capital Ljubljana by looking at different machine learning models, such as linear regression, kernel-machine and artificial neural networks.

    View all citing articles on Scopus
    View full text