An agent-based modeling approach combined with deep learning method in simulating household energy consumption

Abstract

Household has been one of main energy consumption sectors in China, which brings about more challenges for the control of air pollutants and greenhouse gases. Due to the heterogeneities among household individuals, it is much more difficult to regulate the household energy consumption. So, it becomes necessary to understand and guide household energy consumption behaviors. In this paper, a multi-level agent-based model was constructed to aggregate the macro-level energy consumption (such as regions) from the micro-level entities (such as devices), and deep learning models were conducted and embedded in to predict the household energy consumption behaviors with factors as inputs and behavior parameters as outputs, meanwhile an energy substitution mechanism was designed. A series of scenarios were simulated based on actual survey samples, and the results indicated the great potential of our proposed method in the household energy simulation. The simulation fit the actual statistics well in the validation step. Under scenario analysis, the whole coal-to-electricity conversion was deeply determined by low-income individuals, and the effect pattern of certain factor was distinct and could be enhanced by others, while the marginal utility changed non-linearly with peak point effected by income growth rate. It is inspired that the local economic and socio-demographic situations should be comprehensively considered to achieve a trade-off between costs and expected results of regional energy strategies. Our research proposes a novelty study framework to simulate the household energy consumption from the micro individual perspective, and it is a valuable attempt to understand and quantify complex household energy consumption behaviors.