Skip to content Skip to navigation

Energy Behavior Diffusion Modeling

Avg energy of agents over simulated social distance vs. percent of population clowned
Even if we cannot predict the exact amount of energy that may be saved due to various manipulations to change behavior, the possible interventions might at least be rank ordered. Such a ranking, crossed with cost, would be used in choosing which manipulations to employ.
Avg energy of agents over simulated social distance vs. percent of population clowned
Investigator(s): 

The goal of this project was to use computational simulation to estimate the differential efficacy of behavioral manipulations on energy usage applied in various environments. Specifically, the investigator aimed to develop a general methodology to enable analysts to propose a novel intervention and then rapidly and inexpensively explore its potential efficacy under a variety of conditions, and, if desired, to rank these predictions against other proposed manipulations. The method would also enable the choosing of different real-world settings in which to deploy different manipulations to maximize outcomes.

Two models were developed. The "A" model was based upon existing modeling technologies and was used to pilot the basic modeling theory and test the hypothesis that very simple manipulations would produce observable differentials in energy utilization. Based upon results from the "A" model experiment, the researcher developed a "B" model, called ESim, which was more general. ESim was built on a cloud-based modeling infrastructure that would, in principle, enable other modelers to experiment with this and other similar models.

Specific statistical results were obtained from the ESim model, demonstrating that under different conditions of communication graph structure between agents (i.e., approximate households) order of 10 percent energy savings could be expected.

Future Work

The tool developed here could be used to make predictions, which could then be tested in field trials. Parameters such as effect sizes for a wide variety of behavioral approaches, derived from existing research and evaluation studies, could be incorporated to improve the model. As more studies of behavioral program pilots are run, the true empirical responses to the corresponding programs can be incorporated into the model. A description of the work and a user’s guide is online, and the code (both for the model, and for the graphical user interface so that non-programmers can utilize the model to make predictions) is available for others to use. Please contact the investigator: jshrager@stanford.edu.