Elsevier

Neurocomputing

Volume 502, 1 September 2022, Pages 140-147
Neurocomputing

Predicting vehicle fuel consumption based on multi-view deep neural network

https://doi.org/10.1016/j.neucom.2022.06.047Get rights and content

Abstract

The problem of global warming is getting more serious, and vehicle emission is the main cause. In recent years, the number of locomotives in China has been increasing at a rate of more than 20% per year, and the problem of automobile pollution is becoming more serious. The transportation industry is the main source of fossil fuel combustion and environmental pollution. Therefore, in this paper, we propose a multi-view deep neural network (MVDNN) to analyze the key factors affecting the fuel consumption of automobiles. The experiments show that the introduction of human input improves the prediction accuracy and the root mean square error (RMSE) achieves 0.993. In addition, this paper also finds that for drivers, driving habits, driving frequency, and safety awareness are the most important factors affecting the fuel consumption of vehicles by combining Lasso regression with MVDNN. Finally, by comparing the prediction accuracy of different experiments, relevant policy suggestions are made.

Introduction

Humans emit greenhouse gases through the burning of fossil fuels, causing global warming problems. At present, the concentration of vehiclebon dioxide in the atmosphere is the highest in the past at least 2 million years [1]. In recent years, COVID-19 has exacerbated global warming. Rapid but uneven economic recovery from the 2020 COVID-induced recession is putting strains on parts of today’s energy system, causing the second-largest annual increase in CO2 emissions in history. Consistent with the landmark 2050 Net Zero Emissions Scenario (NZE), there is still a long way to go before stabilizing the global temperature rise at 1.5 °C and achieving other energy-related sustainable development goals [2]. The report “2050 Net Zero Emissions: A Roadmap for the Global Energy Sector” points out that there will be a net-zero emission route in the global energy industry by 2050. The key is to make unprecedented changes in the production, transportation, and use of global energy. Specific actions include that after 2035, no new internal combustion engine passenger vehicles will be sold (electric vehicles and other new energy vehicles will replace traditional gasoline vehicles) [3].

Tackling climate change has also become one of the core issues of various countries. During the 26th Conference of the Parties (COP26), more than 50 countries have pledged to meet net-zero emissions targets [2]. According to ECIU 2020, 127 countries have pledged to achieve vehiclebon neutrality, and most of them are planning to realize it by 2050, such as the European Union, Britain, Canada, Japan, New Zealand, South Africa, etc [4]. If these countries can achieve their goals on time and in full, the global emissions would bend to curve down, as modeled in the Announced Pledges Scenario (APS) by IEA (2021).

China is the world’s biggest source of CO2. In 2019, China's greenhouse gas emissions were 14 billion tons of vehiclebon dioxide equivalent, accounting for 26.7% of the total global emissions [5], [6]. In 2020, Chinese vehiclebon dioxide emissions accounted for approximately 29% of the world's total emissions [7]. The main reason is that China is the world’s factory with abundant coal resources and relies on a lot of fossil fuels. China has actively adopted various measures to reduce emissions. As the world's second-largest economy and the largest developing country, China has shouldered additional responsibilities for the international community by implementing Paris Agreement. Starting from the “Eleventh Five-Year Plan”, China has incorporated energy conservation and vehiclebon reduction into its national economic and social development plans. By actively promoting industrial structure adjustment, energy structure optimization, and energy efficiency improvement in key industries, significant results have been achieved in energy conservation and emission reduction, laying an empirical foundation for the realization of the “dual vehiclebon” goal. At the end of 2019, China's vehiclebon intensity had been reduced by about 48.1% compared with 2005, and non-fossil energy accounted for 15.3% of primary energy consumption, fulfilling ahead of schedule the voluntary emission reduction commitments made by the Chinese government at the Copenhagen Climate Change Conference [7]. China is seeking more sustainable, inclusive, and resilient economic growth, following the updated nationally determined contribution targets of the Paris Agreement. In response to climate change, the Chinese government has set a clear goal to reach vehiclebon peak by 2030 and achieve vehiclebon neutrality by 2060. Vehiclebon dioxide emissions per unit of GDP will drop by more than 65% compared to 2005. Non-fossil energy will account for about 25% of primary energy consumption. Wind power and solar power generation's total installed capacity will reach more than 1.2 billion kilowatts, achieving vehiclebon neutrality by 2060 [8].

The transportation industry is one of the most responsible agents for the depletion of fossil fuels and environmental pollution. Approximately, one-fourth of greenhouse gas (GHG) emissions are contributed by transport vehicles [9]. In 2014, the global transportation sector accounted for 28% of global energy demand and 23% of vehiclebon emissions. Every 5,000 miles of personal driving mileage reduced each year around the world will reduce more than one ton of vehiclebon [10]. In recent years, the number of motor vehicles in China's large cities has increased by more than 20% annually, and motor vehicle pollution has become increasingly serious [11]. The State Environmental Protection Administration (SEPA) of China has identified motor vehicle emissions as the main source of air pollution in Chinese cities [12]. In 2004, it was estimated that nitrogen oxides (NOx) emitted by road vehicles accounted for more than 50% of the emission inventory of large cities [12]. According to China’s Multi-resolution Emission Inventory (MEIC), the contribution of vehicles to China’s vehiclebon monoxide (CO), volatile organic compounds (VOCs), and nitrogen oxides (NOx) emissions in 2010 was 11%, 9%, and 17%, respectively. Compared with developed countries (e.g., 500–800 vehicles/1,000 people in Japan, Europe, and the United States), China’s vehicle ownership per thousand people is still very low (i.e., 88 vehicles/1,000 people in 2012), and many studies show that China’s vehicle ownership will continue to increase by 2030 [13], [14], [15], [16]. Therefore, air pollution caused by vehicle fuel will further aggravate if without policy control and type replacement [17].

Many researchers, including data scientists, have been working to determine how the transportation sector affects global warming. Data scientists can effectively contribute to research by designing predictive models of vehicle fuel consumption, which can further help accurately predict future changes in air pollution trends. Therefore, deep learning (DL) models can be used as an effective tool to help develop predictive models. In the past, many scholars have used back-propagation (BP) neural networks, general recurrent neural networks (GRNNs), and other neural networks to predict vehicle fuel consumption [18], [19], [20], [21], [22], [23]. In addition, multiple linear regression, support vector machines, and the K-means clustering method have also been applied to this field with their advantages [24], [25]. However, the existing methods usually adopt only a single type of feature for prediction [39]. In fact, there are multiple features can be used such as environmental features, vehicle features and driver features, and these features constitutes multi-view feature. The existing methods ignore the complementarity between different views, therefore their prediction ability is limited.

To address the above problems, we propose a multi-view deep neural network (MVDNN) model to better use the complementarity between the different views and predict the fuel consumption rate more accurately. We analyze the main factors that affect the fuel consumption of automobiles. Compared to the previous studies, we collected large survey data of driving behavior factors, and filled up the lack of behavior data in the field of fuel consumption prediction literature. Extensive experiments have been conducted and the results demonstrate the effectiveness of the proposed method. The contriubtions of this paper can be summarized as follows:

  • 1.

    Compared to the previous studies, we collect a large number of data with driving behavior features and fill up the behavior data in the field of fuel consumption rate prediction.

  • 2.

    We combine the environmental features, vehicle features and driver features to comprehensively leverage the features from multiple views.

  • 3.

    We propose MVDNN which integrates the environmental, vehicle and driving behavior to achieve more accurate predictions for vehicle fuel consumption.

The rest of the paper is structured as follows: Section 2 covers related work on predicting vehicle fuel consumption. Section 3 introduces the proposed method, models, and results are introduced in Section 4. Section 5 is the conclusion of the study.

Section snippets

Related work

To predict vehicle fuel consumption, scholars have adopted many methods which can be roughly divided into traditional methods and machine learning methods. Different methods have different application scenarios and advantages.

Since the 1990s, more and more studies have paid attention to fuel consumption prediction. The early methods are mainly regression models, time series models, and so on. Some researchers [26] use several hybrid regression models that predict hot stabilized vehicle fuel

Methodology

In this section, we introduce the MVDNN and the impact of drivers on prediction accuracy. The main features of the proposed model are as follows.

  • 1.

    To predict the metrics related to the fuel consumption rate of vehicles, we use MVDNN with environmental, driver and vehicle views. We train and compare the performance of the prediction models using MSE with 80% of the data as the training set and 20% of the data as the test set.

  • 2.

    We optimize the efficiency of MVDNN by reducing the number of neurons

Dataset

The real fuel consumption data in this paper is obtained from Bear Oil APP (https://www.xiaoxiongyouhao.com). Bear Oil APP contains the driving information of vehicle owners from 31 provinces in China, with a total of over 23 billion kilometers of recorded mileage and over 51 million kilometers of real fuel consumption data. The factors considered in this paper include driver factor, vehicle factor, and environmental factor. To better investigate the impact of driver factors on fuel

Conclusion

Global warming is getting more and more serious, and vehiclebon emission is the main cause. The transportation industry is the main source of fossil fuel combustion and environmental pollution. In this paper, we analyze the main factors that affect the fuel consumption of automobiles. MVDNN is proposed for vehicle fuel consumption prediction and more accurate prediction results are achieved. Compared to the previous studies, we collected large survey data of driving behavior factors, and filled

CRediT authorship contribution statement

Yawen Li: Conceptualization, Methodology, Writing – original draft. Isabella Yunfei Zeng: Formal analysis, Writing – original draft. Ziheng Niu: Conceptualization, Writing – review & editing. Jiahao Shi: Data curation, Writing – review & editing. Ziyang Wang: Conceptualization, Methodology, Supervision. Zeli Guan: Methodology, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work is supported in part by the National Natural Science Foundation of China (61902037, 72134002, 72173145), and in part by the Fundamental Research Funds for the Central Universities under Grant 500419804.

Yawen Li is an associate professor of School of Economics and Management, Beijing University of Posts and Telecommunications. She received her Ph.D. in Innovation, Entrepreneurship and Strategy from Tsinghua University in 2018. Her research interest focuses on the collaborative innovation, the development of science parks, and scientific productivity of firms. Her research papers have been published in or accepted by journals including IEEE Transactions on Knowledge and Data Engineering,

References (53)

  • Z. Xue et al.

    Deep low-rank subspace ensemble for multi-view clustering

    Inf. Sci.

    (2019)
  • J. Yuan et al.

    Ship energy consumption prediction with gaussian process metamodel

    Energy Procedia

    (2018)
  • R. Vazquez et al.

    Stochastic analysis of fuel consumption in aircraft cruise subject to along-track wind uncertainty

    Aerosp. Sci. Technol.

    (2017)
  • Intergovernmental Panel on Climate Change. Climate Change 2021: the Physical Science Basis,...
  • International Energy Agency. World Energy Outlook 2021,...
  • International Energy Agency. Net Zero by 2050: A Roadmap for the Global Energy Sector,...
  • The European Consortium of Innovative Universities. 2020: the natural year for climate action,...
  • BP. BP Statistical Review of World Energy 2019,...
  • United Nations Framework Convention on Climate Change. United Nations Climate Change Annual Report 2019,...
  • G. Zhuang

    Challenges and countermeasures for my country to realize the “double carbon” goal

    People's Forum

    (2021)
  • Y. Bai et al.

    The background, challenge, opportunity and realization path of the dual carbon goals

    China Econ. Rev.

    (2021)
  • Intergovernmental Panel on Climate Change. Outreach event on IPCC Special Report on 1.5 Degrees (SR15),...
  • State Environmental Protection Administration of China. 2004 Annual Plenary Meeting of Joint Research Network on...
  • M. W, H. Huo, L. Johnson, D. He. Projection of Chinese motor vehicle growth, oil demand, and CO2 emissions through...
  • H. Huo et al.

    Vehicular air pollutant emissions in China: evaluation of past control policies and future perspectives

    Mitig. Adapt. Strat. Glob. Change

    (2015)
  • H. Huang et al.

    Construction of a python-based automobile fuel consumption prediction model

    Electron. Meas. Technol.

    (2021)
  • Cited by (35)

    View all citing articles on Scopus

    Yawen Li is an associate professor of School of Economics and Management, Beijing University of Posts and Telecommunications. She received her Ph.D. in Innovation, Entrepreneurship and Strategy from Tsinghua University in 2018. Her research interest focuses on the collaborative innovation, the development of science parks, and scientific productivity of firms. Her research papers have been published in or accepted by journals including IEEE Transactions on Knowledge and Data Engineering, Neurocomputing, Asian Journal of Technology Innovation, Journal of Leadership and Organizational Studies.

    Isabella Yunfei Zeng is an undergraduate at Massachusetts Institute of Technology. She has focused on environmental engineering and green economics. Her research ranged from carbon tax and carbon policy, GHG emissions, fuel consumption, to green city and urban sustainable studies. Her research papers have been published in or accepted by journals including Sustainability, Energies.

    Ziheng Niu is an undergraduate student at the International College of China Agricultural University (CAU), currently majoring in Economics and minoring in Mathematics at CAU, and studying Economics at the College of Arts and Sciences in Denver, Colorado. His research interests span mainly Green Development, Financial Economy, and Macroeconomics.

    Jiahao Shi is currently an undergraduate at the School of Environmental Science and Engineering at Peking University and the National Institute of Development at Peking University. He has done research on Green Finance and Machine Learning. His research interests span mainly Green Development, Financial Economy, Machine Learning, Forecasting, and Time Series.

    Ziyang Wang is currently a postdoctoral fellow at the School of Statistics and Mathematics at Central University of Finance and Economics. He received her Ph.D. in Finance from Tsinghua University in 2020. He has done research on Company Valuation and Business Model. His research interests span mainly Organizational Innovation, Digital Economics, Business Model and Financial Economy. His research paper has been published in Management World.

    Zeli Guan is currently a Ph.D. candidate of School of Computer Science, Beijing University of Posts and Telecommunications. His mainly research directions for federated learning, graph neural network and machine learning. His research papers have been published in or accepted by International Journal of Intelligent Systems, Computational Intelligence and Neuroscience, and Visual Informatics.

    View full text