1. To manage the growing energy demand, there is a need for energy system optimization, especially on the demand side. We use a first principles approach to build a high-resolution energy demand model. This framework generates activity-based, building-level, time-dependent energy demand profiles. The model associates appliance usage with each household activity and calculates energy consumption based on the appliance energy rating, the duration of the energy-consuming activity, and the type of activity performed by each household member. This model was then enhanced to account for the effects of higher rooftop solar penetration on the temporal changes of households’ energy demand profiles. This information is expected to be useful to electric power utilities in efficiently managing the sudden changes in demand in high penetration regions.
2. We apply highly imbalanced survey data with only 2% solar adopters to build a prediction model. Given that our focus is on predicting adopters, we use an alternative approach that allows us to optimize the model under a specific user-defined objective. It is called the decision-adjusted approach. The logistic model maximizes the likelihood of the distribution. Instead of only maximizing the likelihood, the decision-adjusted model allows trade-offs between predicting adopters and maximizing the overall prediction accuracy of the model. In the decision-adjusted model, we want to maximize the precision, or positive predictive value (PPV), given a certain number of predicted positives. The precision is defined as Precision = TP/(TP + FP) i.e. true positives divided by true plus false positives. It measures the fraction of true adopters among all adopters identified by the model as adopters. Our results show that the decision-adjusted model performs better than the traditional logistic model in general.
3. We show that peer effects have an important role in the spread of solar adoption. This leads to a natural problem of how to design incentives to maximize adoption in such a model. While this is an instance of an “influence maximization” problem, prior results from the influence maximization literature cannot be used directly. In our work, we extend prior results from the literature on the use of submodularity to obtain a greedy approximation. We use this new result to do optimal “seed set” selection for a highly detailed, data-driven, agent-based model of household rooftop solar adoption.

Figures

A prototype analytical tool has been made available to NRECA to estimate the probability of rooftop solar adoption.

• UVA SEEDS2 Static Prediction Model 
• UVA SEEDS2 Diffusion Model

Conference Publications

S Thorve, Z Hu, K Lakkaraju, J Letchford, A Vullikanti, A Marathe, S Swarup. An Active Learning Method for the Comparison of Agent-based Models. Proceedings of the International Conference on Autonomous Agents and Multiagent Systems (AAMAS). Auckland, New Zealand, May 9-13, 2020. 

Z. Hu, X. Deng, A. Marathe, S. Swarup and A. Vullikanti. Decision-Adjusted Modeling for Imbalanced Classification: Predicting Rooftop Solar Panel Adoption in Rural Virginia. Computational Social Science Annual Conference, Santa Fe, NM, July 2019. 

A Gupta, Z Hu, A Marathe, S Swarup and A Vullikanti. Predictors of Rooftop Solar Adoption in Rural Virginia. Computational Social Science Conference 2018, October 25-28, Santa Fe, NM. 

S Thorve, S Swarup, A Marathe, Y Chung Baek, E Nordberg, M Marathe. Simulating Residential Energy Demand in Urban and Rural Areas. Winter Simulation Conference, December 9-12, 2018, Gothenburg, Sweden. 

M Padhee and A Pal. Effect of solar PV penetration on residential energy consumption pattern. IEEE 50th North American Power Symposium. Fargo, North Dakota, September 17-19, 2018 

M. Padhee, A. Pal, and K. A. Vance. Analyzing Effects of Seasonal Variations in Wind Generation and Load on Voltage Profiles. IEEE 49th North American Power Symposium, Morgantown, WV, September 17-19, 2017 

M. Padhee and A. Pal, “Fast DTW and fuzzy clustering for scenario generation in power system planning problems,” submitted to 52nd North American Power Symposium, Tempe, AZ, Oct. 11–13, 2020. 

A Gupta, R Graham, S Swarup, A Marathe, K Lakkaraju, A Vullikanti. Designing Incentives to Maximize the Adoption of Rooftop Solar Technology. Proceedings of the International Conference on Autonomous Agents and Multiagent Systems (AAMAS). Stockholm, Sweden, July 10-15, 2018.

Journal Articles

S Thorve, Z Hu, K Lakkaraju, J Letchford, A Vullikanti, A Marathe, S Swarup. An Active Learning Method for the Comparison of Agent-based Models. Invited to Journal of Autonomous Agents and Multiagent Systems (JAAMAS), submitted July 2020. 

R Subbiah, A Pal, E Nordberg, A Marathe, M Marathe. Energy Demand Model for Residential Sector: A First Principles Approach. IEEE Transactions on Sustainable Energy, vol. 8, no. 3, July, pages 1215-1224, 2017. 

M. Padhee, A. Pal, and B. Jafarpisheh, “A decentralized BESS allocation scheme for T&D networks considering systemic uncertainties,” submitted to IEEE Trans. Sustain. Energy, Apr. 2020.

Book Chapters

R Meyers, P Miller, T Schenk, WM Ford, RF Hirsh, S Klopfer, A Marathe, A Seth, MJ Stern. A framework for sustainable siting of wind energy facilities: Economic, social and environmental factors. Energy Impacts: A Multidisciplinary Exploration of North American Energy Development, Eds. by Jeffrey B. Jacquet, Julia H. Haggerty, and Gene L. Theodori. 2019. 

S. Swarup, A. Marathe, M. Marathe and C. Barrett. Simulation Analytics for Social and Behavioral Modeling. Chapter 26, Social-Behavioral Modeling for Complex Systems, edited by P. Davis, A. O'Mahony and J. Pfautz, John Wiley & Sons, pages 617-632, April 2019.