Natural gas consumption forecasting model based on coal-to-gas project in China

1 Introduction

Oil and natural gas are the most important sources of energy worldwide.Governments rely upon accurate energy demand to formulate energy policies and adjust industrial structures.Internationally,energy security is an integral part of every country’s national security because natural resources could create dependencies that could hinder a nation’s economic development.For instance,the demand for gas is increasing because power companies in the United States of America are increasingly implementing coal-togas projects (CGPs) [1].China’s energy consumption is expected to increase by 53 percent within 2017-2027 [2]; this issue may constrain the economic and social development of China.China relies strongly on energy imports for its continual rapid economic development.According to the 2018 edition of the British Petroleum world energy statistics yearbook,the global energy demand exceeded its ten-year average growth rate.Even with the promotion of CGPs,China is still the world’s largest energy consumer,accounting for 23.2% of the global energy consumption and 33.6% of the global energy consumption growth.

Following the 2017 Air Pollution Prevention and Control Work Plan in the Beijing-Tianjin-Hebei (BTH) region and the surrounding region issued by the Ministry of Environmental Protection of China,a CGP was created to improve the energy structure and increase environmental protection.With the development of national policies,the CGP in Hebei Province has accelerated since 2017 owing to the promotion of the local governments.The actual number of households finishing CGP in Hebei Province is larger than the planned number in the winter of 2017.

Several experts view CGPs as a green technology that produces extremely less sulfur dioxide and nitrogen oxide as a solution for urban pollution.China’s energy consumption is expected to increase by 53 percent within 2017-2027 [2]; this issue may constrain the economic and social development of China.The shortage of gas supply in northern part of China directly leads to the rising gas price nationwide.Domestic gas supply is not sufficient,and the foreign natural gas supplied to China is monopolized.As a subcompany of China National Petroleum Corporation,PetroChina Natural Gas Marketing North Company (PCNGMN) has been focusing on natural gas resource allocation while considering numerous issues such as ensuring the residential gas consumption in winter and the performance of economic indicators for an entire calendar year in the northern part of China.

As energy productivity is an important index of local government [3],local state-owned enterprises(SOEs) may pursue more sustainable financing objectives similar to a few foreign oil and gas companies that have been focusing on sustainable supply chain management [4-7].The CGP policy not only focuses on improving the air quality index but also provides a robust opportunity for citizens,businesses,and the government to collaborate and learn how to make CGPs sustainable through emphasis on the safety and well-being of society and continual energy poverty management.Energy poverty becomes apparent in the energy consumption per capita,affordability,energy consumption structure,and access to modern clean energy [8].Energy safety is vital for the well-being and health of society,and it is closely related to how energy poverty management is governed.Due to the insufficiency of gas consumption forecasting methods and evaluation,the CGP policy has been put forward without the full demonstration.The negative impact on natural gas and energy market still exists,such as the price fluctuation of natural gas,the imbalance between supply and demand.Thus,it is extremely important for the government and energy industry to forecast gas consumption with the implementation of CGPs.

The prediction of natural gas consumption is a classical study area.There are many studies and ways have focused on developing different forecasting approaches.Ma [9]established forecasting models for annual natural gas consumption in China.Kizilasl [10]built forecasting models for monthly natural gas consumption.Brabec [11]built daily natural gas forecasting models.From the view of forecasting-tools,curve model,grey prediction model,statistical models,artificial neural networks,mathematical model,conditional demand model are used to forecast the natural gas consumption.

The most recent studies for natural gas consumption methods are concerning various economic factors and horizons.Zhang [12]presents a long-term natural gas load forecasting model based on gray neural network.A tenyear gas consumption example shows the validity of the combined forecasting model.Meanwhile,Zhu [13]presents a support vector regression-based approach to predict short-term natural gas demand.It has been used to predict the daily gas consumption in a week and successfully applied to operations for National Grid in the United Kingdom.Ioannis [14]even gives the day-ahead gas demand forecasting with a model combination of the Wavelet Transform,Genetic Algorithm,Adaptive Neuro-Fuzzy Inference System and Feed-Forward Neural Network.Shaikh [15]presents a logistic model to forecast the natural gas demand in China.Karimi [16]uses an artificial neural network-based genetic algorithm to predict natural gas consumption.These studies purely use the historical data to get better fitted curves.The related factors such as housing structure,gas stove,snowing and raining weather,and etc.are considered as implicit effects to the predictions.

The main objective of these models was to investigate the underlying relationship among various variables that drive energy demand to estimate future energy demand.Using the statistical data based on regression analysis,a number of scholars have examined the effects of temperature and solar radiation on the heating activities of residential buildings to predict the heating energy consumption of the buildings.Numerous factors influence the daily gas consumption by users.It is necessary to determine the different use of natural gas for predicting gas consumption accurately.Thus,it may be helpful to compare outliers and spikes within raw data analyses.

According to the statistical data provided in this paper,the gas consumption before and after the CGP is considerably different.This makes the typical methods for gas consumption forecasting invalid after the CGP.Thus,a new method is proposed for the prediction considering the CGP process.The rest of the paper is organized as follows.The characteristics of the gas consumption and gas consumption structures is analyzed with the gas consumption data from users.From the analysis,one can see that the impact of CGP policy is obvious.Thus,it is necessary to get the gas consumption method based on the historical data and prediction method.Following the gas consumption characteristics,the gas consumption can be divided into two parts,i.e.the CGP and none-CGP parts.The gas consumption forecasting method is built for the two parts.Finally,the effectiveness of the method is validated by the gas consumption in 2018.

2 Analysis of sales data

The original sales data of company users from PCNGMN are confidential.Thus,the detailed data related to actual company users are decrypted by technical method.

In Northern China,the CGP is implemented in the BTH region,in which Hebei had the highest gas consumption growth in 2017,as shown in Fig.1.In 2015 and 2017,the annual growth rate of the total gas consumption was 7% and 2% in Beijing,7% and 9% in Tianjin,and 8% and 32% in Hebei,respectively.The increase in natural gas consumption in Hebei Province between 2015 and 2017 is quite different from that in Beijing and Tianjin.Hence,it is necessary to analyze the growth rate of Hebei Province in 2017.

Fig.1 Annual gas consumption in BTH region from 2015-2017

The analysis of daily gas consumption in the BTH area shows that there is a similar trend between Beijing and Tianjin.The daily gas consumption in both cities increases in winter and summer,where the increase in summer is relatively flat.The daily gas consumption in Hebei Province increases only in winter,and the increase in winter in 2017 is considerably larger compared to 2016.This is caused by the influence of the CGP.The relative increase of natural gas consumption in summer shows that both Beijing and Tianjin use natural gas to generate electricity in summer,while there is a small part of natural gas to generate electricity in Hebei Province.investigation and Therefore,the variation in daily gas consumption in Hebei Province is different from that in Beijing and Tianjin.Hence,it is necessary to analyze these three areas separately.

Fig.2 Daily gas consumption in BTH region from 2015-2017

The daily gas consumption in BTH region shows that the gas consumption in BTH region in three years has different characteristics as in Fig.3-5.

Fig.3 Daily gas consumption in Beijing from 2015-2017

Fig.4 Daily gas consumption in Tianjin from 2015-2017

Fig.5 Daily gas consumption in Hebei from 2015-2017

Through the intuitive comparison of the natural gas consumption data of the BTH area,the change in the natural gas consumption in Hebei Province caused by the CGP shows distinct characteristics.Therefore,as Hebei Province is the most affected by the CGP,the natural gas consumption forecasting model for this region is important for the sustainable development of natural gas marking in China.

3 Natural gas consumption forecasting model with CGP process

Natural gas consumption is typically influenced by temperature,population,GDP,and other factors.Natural gas consumption forecasting is the estimation and prediction of future data based on known historical data while considering the relevant factors that affect natural gas demand.To obtain accurate prediction results,we should analyze the influencing factors in different regions in different periods.The effect and degree of different factors on the prediction object is as in Table 1.

Table 1 Pearson product-moment correlation coefficient

Factor Correlation coefficient Average temperature -0.95 Heating condition 0.89 GDP 0.73 Resident income -0.7 Population -0.68 GDP of tertiary industry 0.65 GDP of primary industry 0.48 Air quality index -0.43 GDP of secondary industry 0.43 Urbanization rate 0.41 Wind speed -0.13 Average precipitation -0.1 Average humidity -0.08

Natural gas consumption is not only affected by stable factors but also by policies to a large extent.The influence of policies is difficult to predict using conventional forecasting methods.This is because conventional forecasting methods predict natural gas consumption based on historical data.However,CGP policies are implemented abruptly.The effect of such policies is not included in historical data.Hence,the natural gas consumption with the CGP can be divided into two parts,i.e.,the basic amount and the increment caused by the CGP.In the process of forecasting,appropriate methods should be adopted to find and utilize the law of change implied in historical data and the mechanism of the influence of the CGP.The influence of the CGP is mostly observed in the heating season.Hence,this study focuses on the natural gas consumption in the heating season in Hebei Province.

To predict the gas consumption in Hebei Province considering the CGP process,this study divides the final model into two submodels,namely,the model for predicting the increment in gas consumption due to the CGP and the basic consumption prediction model (excluding the CGP effect),as shown in Fig.6.The basic consumption part uses the support vector machine (SVM) method,which is a kind of generalized linear classifier for binary classification of data by supervised learning method.SVM method is used as a black-box model to process the historical data.The incremental model due to CGP uses the characteristics of end users (not company users) as the basic data,and calculates the incremental consumption part.

Fig.6 Gas Prediction Model

3.1 The basic consumption part prediction

The original consumption data are processed using the small wavelet method,and the basic consumption is predicted by utilizing the SVM method.The sample data used are the daily gas consumption amount in Hebei Province from January 2015 to July 2017.With the average temperature,GDP,heating conditions and etc.as the input parameters to the SVM model of Hebei Province,the gas consumption prediction result from Angust 2017 to May 2018 is as shown in Fig.7.Since the price is the result of CGP and gas consumption,it is not considered in the consumption forcasting process.Here,Pearson productmoment correlation coefficient method is used,that is,

where rx,y—simple correlation coefficient of x,y；

yi—ith observation of variable y；

xi—ith observation of variable x；

yave—average value of variable y；

xave—average value of variable x；

n— number of observed measurements of variables x and y.

Fig.7 Gas consumption prediction by SVM method

The actual gas consumption is illustrated by black line and the prediction result is marked by red line.The predicted result is from the historical data and follow the characters of past years.The predicted result does not show the influence of CGP,that is,the basic consumption part.So,there is a large gap between the black line and red line.The difference between this two year is considered as the actual increment of CGP in 2017.Other methods such as Exponential Smoothing,Grey Prediction,Logic Model and Artificial Neural Network get nearly the same results.It is not repeated here.

3.2 The incremental consumption part prediction

As the CGP primarily consists of newly added rural users,the amount of the historical gas consumption data of these users is not as much as the data for previous years.It is impossible to establish fitted models using typical methods.Based on the historical data of smart meters,we obtained the daily gas consumption data of 1189 households in December 2018.The statistics of the average daily gas consumption and frequency diagram is as in Fig.8.The distribution of the average daily gas consumption is best fitted by a Beta distribution.

Fig.8 The average daily gas consumption and frequency diagram

Each household’s daily gas consumption can be obtained by sampling the distribution of the average daily gas consumption shown above.The total increment in gas consumption can be calculated as follow:

where a is the number of CGP households,q is each household’s daily gas consumption and s is the number of days in heating season.

The public data online shows that the number of CGP households 2.7 million and 1.4 million in 2017 and 2018,respectively,and the number of heating days is 120.According to the characteristics of end-user gas consumption,the sample shows that the increment in gas consumption due to the CGP in 2017 should be:

According to an online report,only 75% of the users actually ventilated.Hence,the actual increment due to the CGP in 2017 should be:

The CGP that was completed but not utilized in 2017 was used next year.Hence,this surplus should be 0.7 billion cubic meters in 2018.

The newly completely CGP in 2018 is also utilized in 2018.The sample shows that the increment in gas consumption due to the CGP in 2018 should be:

3.3 Prediction of total gas consumption

Finally,the total gas consumption in winter of 2018 is,Q2018=P1+P2+P3+P4 ,which consists of these 4 parts as in Fig.9.The total gas consumption in winter of 2018 increases by 40% over last year.Comparing with the accounting data of PCNGMN,the prediction error is less than 10%.

Fig.9 Total gas consumption in winter of 2018

4 Conclusions

The sudden gas consumption increasement caused by the implementation of the CGP makes it more difficult for natural gas suppliers.This study establishes a model to forecast natural gas consumption based on the CGP in China to better understand the consumption of natural gas in the process of the CGP and make gas consumption sustainable.The total natural gas consumption after the implementation of the CGP is divided into two parts,i.e.,the basic part and the increment due to the CGP.The SVM method is used to predict the basic part.

The increment due to the CGP is predicted using the data obtained from end users’ smart meters.The average daily gas consumption is described as a probability distribution,and the probability distribution sampling method is used to predict the increment due to the CGP.Finally,the prediction result is obtained,and prediction error is within the acceptable range.

At present,China’s natural gas supply facilities are still in the stage of development.However,the demand of natural gas users is growing rapidly.Hence,it is necessary to study natural gas consumption characteristics.The accurate prediction of natural gas consumption and preparation for sales in advance can lead to the sustainable development of natural gas as a clean energy source.

New pipelines for imported natural gas will be available in future to support natural gas supply,and the quantity of imported natural gas will increase.The relationship between natural gas consumption,policies,and gas supply facilities may change over time.Therefore,future studies should consider gas supply facilities and update the potential relationship between natural gas consumption,policies,and gas supply facilities.

Acknowledgements

This research was supported by the National Natural Science Foundation of China (No.51504271).