A SHAP machine learning-based study of factors influencing urban residents' electricity consumption - evidence from chinese provincial data

The surge in urban household electricity consumption (HEUR) has posed a challenge to China's energy conservation efforts. In this paper, from the new perspective of SHAP interpretable machine learning, we address the "black box" drawbacks of machine learning in HEUR prediction research, study the influencing factors and mechanisms of the surge of urban household electricity consumption, and explore the interactions of critical factors, such as the aging of the population, the per capita income (PCI), and the electricity consumption rate of a single-person household (SPHR). The findings are as follows: (1) The key influencing factors of HEUR in China (number of general higher education graduates, per capita living space, urbanization rate, number of older people, electricity consumption per square meter of urban housing, and per capita income). (2) PCI is not the developed region's HEUR's dominant factor; on the contrary, PCI significantly affects less developed regions. (3) The influence of key influencing factors on HEUR is a left-low-right-high U-shape nonlinear effect, while AS and FR show inverted U-shape nonlinearity. (4) Higher education has not increased people's awareness and action to consume HEUR; in the future, SPHR will become the main force of HEUR consumption. In addition, compared to men, women with lower incomes save more HEUR. It provides a set of scientific research bases and suggestions for energy conservation, emission reduction and sustainable energy development in China and other countries.