A novel simulation and supervised machine learning-based prediction framework to predict the total transportation and energy costs for single-family households

Abstract

This paper focuses on predicting the total transportation and energy costs (TTEC) for single-family households. A system boundary consisting of grid-powered electricity (GE) and solar-powered electricity (SE) as energy inputs and transportation vehicles that include Gasoline Vehicles (GV) and Electric Vehicles (EV) as transportation methods for energy outputs is studied. A novel three-stage evaluation framework is proposed to predict the TTEC under varying single-family household parameters. In the first stage, an energy balance simulation model is proposed to estimate the TTEC for an individual household. In the second stage, the simulation model is run several times under varying parameters to develop synthetic data that is used as input for the third stage supervised machine learning (SML) models. In the third stage, numerous SML models are trained and tested to determine the best SML model that enables us to predict the TTEC with high accuracy. This best SML model can be used as a substitute for simulation model, thereby reducing the computation burden of running the simulation model for each new single-family household. A case study of single-family households in Central Texas in the US is used as an application of the framework. The results indicate that regression SML models are best in predicting the total costs with an adjusted R-squared of 99.13% and 98.88% on training and testing datasets, respectively. In addition, the parameter analysis of regression SML models suggests that the house size, number of GVs, number of EVs, EV and GV ownership costs, and solar implementation at households are the most important parameters to predict TTEC for single-family households. Counterintuitively, number of residents, GV and EV mileage, solar system size, battery capacity and peak solar hours are not significant parameters that contribute to TTEC prediction.