Rural household energy consumption behavior with neural network approach: A case study

Sheng Cheng, Xiongbiao Xie, Gideon Kwaku Minua Ampofo, Jiping ChuInternational Journal of Heat and Technology 2018 · DOI

Abstract

Rural household energy consumption is an important component of national energy consumption and plays a vital role in rural economic and ecological environment development. In this paper, the energy consumption of 1,126 households in four typical agricultural provinces was investigated. In addition, a multilayer perceptron (MLP) neural network is adopted to conduct variable selection. Moreover, Logit, Tobit, and multiple regression models are used to reveal the impact of different factors on rural resident choice and quantity behavior. The results show that family size, market distance, education level, farmland area, and occupation affect farmers' energy choices and quantity consumption behavior to varying degrees. The differences in energy availability and convenience caused by geographical features are important factors affecting consumption behavior. Income level is a key determinant of the transition to cleaner energy, though coal consumption remains relatively insensitive to income changes. Improving the rural energy structure in China requires boosting rural income levels and integrating rural energy into the national strategy framework. The government should increase investment in renewable energy research and use economic measures like taxes and subsidies to reduce costs and accelerate the upgrading of the rural energy structure.

Introduction

For decades, China’s energy strategy focused heavily on urban centers and heavy industry, leaving rural energy systems in a state of relative isolation. However, as rural economies grow, energy demand is skyrocketing and shifting from traditional biomass (straw and wood) to commercial sources like coal and electricity. The challenge for policymakers is that rural energy behavior is incredibly heterogeneous. Factors like geography, market access, and family size create a "dual structure" that makes a one-size-fits-all policy ineffective. This study identifies specific factors driving fuel choice and quantity to provide a data-driven roadmap for green transformation in the countryside.

Methodology

The authors employ a two-stage analytical framework bridging machine learning and classical statistics. To avoid subjective bias in choosing which factors to study, the researchers utilized a three-layer Multilayer Perceptron (MLP) neural network for variable selection. By training the network on survey data and applying an improved model selection criterion (AICc), they filtered out insignificant variables, leaving only the most impactful "effective variables" for final analysis.

The neural network structure used for identifying effective variables in rural energy consumption.

Once variables were selected, three distinct econometric models were used:

  • Logit Model: Analyzes binary choice behavior (e.g., whether a household chooses LPG).
  • Tobit Model: Analyzes quantity behavior for fuels like biomass and LPG, addressing "zero-consumption" data points from households that do not use those specific fuels.
  • Multiple Regression (GLM): Used for electricity analysis, as it is universally adopted across the surveyed households.

Data and Variables

The study is based on a field survey of 1,126 valid household samples across four typical agricultural provinces: Shandong, Hubei, Hunan, and Sichuan. These regions represent a spectrum of economic development and agricultural output. The dataset includes metrics on per capita income, arable land area, market distance, family demographics, and geographical features (plains vs. mountains).

Key Insights

  • The "Coal Trap": Unlike other fuels, coal consumption is not sensitive to income. Because coal is easy to transport, store, and buy in bulk, it remains the preferable choice for poorer families far from markets, regardless of their desire for cleaner options.
  • The Land-Biomass Link: There is a direct positive correlation between arable land ownership and biomass consumption. More land provides more "free" fuel (straw and stalks), disincentivizing the switch to paid commercial energy.
  • Education as a Catalyst: Higher education levels in the household head significantly correlate with a shift toward clean energy, suggesting the transition is driven by environmental and health awareness as much as wealth.
  • Geography Dictates Availability: Households in mountainous regions rely more heavily on biomass and are less likely to adopt LPG due to high transportation costs and lack of infrastructure.

Key Results

The study confirmed that rural households do not simply "switch" fuels; they "stack" them. The most common pattern, observed in 54.33% of households, is a mix of coal and electricity, followed by a mix of straw, coal, and electricity (42.67%). While 100% of surveyed households now use electricity, it is primarily used for lighting and appliances rather than primary cooking or heating.

Prevalence of different energy 'stacking' combinations in the surveyed households.

Regarding income elasticity, per capita income has a significant negative effect on biomass choice (-0.2408) and a strong positive effect on LPG choice (0.7102). Market distance also plays a critical role: for every unit increase in distance from a market, coal consumption tends to increase while LPG adoption drops.

Conclusion and Policy Recommendations

While income is the primary driver for energy transition, the "income-insensitivity" of coal suggests that wealth alone will not eliminate dirty fuel use. The authors recommend that the government use economic levers, such as taxes on high-emission fuels and subsidies for clean technology, to lower the initial installed costs of renewable systems. Furthermore, because rural energy is highly dependent on local environmental endowments, national policies must allow for regional flexibility to be effective.

Reference: Cheng, S., Xie, X., Ampofo, G. K. M., & Chu, J. (2018). Rural household energy consumption behavior with neural network approach: A case study. International Journal of Heat and Technology, 36(4), 1482-1492.

Summary updated

Abstract

Rural household energy consumption is an important component of national energy consumption and plays a vital role in rural economic and ecological environment development. In this paper, the energy consumption of 1126 households in four typical agricultural provinces was investigated. In addition, multilayer perceptron neural network is adopted to conduct variables selection. Moreover, Logit, Tobit and multiple regression models are used to reveal the impact of different factors on rural resident choice and quantity behavior. The results show that family size, market distance, education level, farmland area and occupation affect farmers' energy choices and quantity consumption behavior to varying degrees. The differences of energy availability and convenience caused by the difference of geographical features are important factors that affect the energy consumption behavior of farmers. Income level is a key determinant of transition of energy consumption behavior, which however is not sensitive to coal. We hold the opinion that the key to improve rural energy structure in China is to boost the income level of rural residents. Meanwhile, rural energy consumption should be included in the national energy strategy framework. The government should increase the investment on the research and implementation of the renewable energy and use economic measures such as taxes and subsidies to reduce initial installed costs and operating costs so as to speed up the of upgrading rural energy consumption structure.
BibTeX
@article{cheng2018rural,
  author = {Sheng Cheng and Xiongbiao Xie and Gideon Kwaku Minua Ampofo and Jiping Chu},
  title = {{Rural household energy consumption behavior with neural network approach: A case study}},
  journal = {International Journal of Heat and Technology},
  year = {2018},
  doi = {10.18280/ijht.360442},
}
Figure