GIS-Integrated Agent-Based Modeling of Residential Solar PV Diffusion Energy Systems transformation Scott A. Robinson, Matt Stringer, Varun Rai, & Abhishek Tondon
Motivation
Agent Based Modeling -> Time Follow decision rules ( functions ) Have memory Perceive their environment Are heterogeneous Are autonomous Agents: From: Deffuant, 2002.
Agent Attribute Example: Wealth PV Adoption by QuartileAverage Income by Quartile
Agent Attribute: Wealth
Environment Example: Tree Cover > 60% Tree cover < 15% Tree cover
Yes No Are there PV owners in my network? RA: select one network connection. Is connection credible? No further activity Agent Initialization: Small World Network of n% Locals, 1-n% Non-locals. Assign initial Attitude Modify SIA. Is SIA >= threshold? ADOPT Financially capable? Wealth + NPV + PP (Control) Behavioral Model Attitude becomes socially informed: SIA From: Watts, 1998.
Implementation Focus Test Site : One zip code in Austin, TX 7692 households 146 PV Adopters (1.9%) as of Q City of Austin had approx PV Adopters Time Period : Q – Q Methods : Multiple runs in each batch to allow for inherent randomness in network initialization and interaction effects Runs in a batch have identical parameters Validation : Batches test different parameters against real test site data.
Temporal Validation Empirical Many strong interactions, radial neighborhoods, 90% local connections. Adopters are EOHs. Few weak interactions, no EOHs Weak interactions More non-local connections Weak interactions, contiguous neighborhoods
Spatial Validation
Current Work Agent Class: Installers -> Time
Summary ABMs are virtual laboratories PV diffusion is a complex process with rich interaction effects : Agent behavior: theory of planned behavior Agent networks: small world networks Agent interaction: relative agreement algorithm Multidimensional validation (space and time) allows the robustness of the ABM to be tested against “ ground truth ” events. Early testing: Strong, monthly interactions 90% geographic locals. 2000ft radial neighborhoods Existing adopters with low uncertainty in attitude. Low RMSE (3.6), and accurate clustering (1 false positive).
Q & A Robinson, S.A., Stringer, M, Rai, V., Tondon, A., "GIS-Integrated Agent- Based Modeling of Residential Solar PV Diffusion,“ USAEE North America Conference Proceedings 2013, Anchorage, AK. Rai, V. and Robinson, S. A. "Effective Information Channels for Reducing Costs of Environmentally-Friendly Technologies: Evidence from Residential PV Markets," Environmental Research Letters 8(1), , 2013 Rai, V. and Sigrin, B. "Diffusion of Environmentally-friendly Energy Technologies: Buy vs. Lease Differences in Residential PV Markets," Environmental Research Letters, 8(1), , Rai, V., and McAndrews, K. “Decision-making and behavior change in residential adopters of solar PV,” World Renewable Energy Forum, 2012, Denver, CO. Selected References:
Appendix: TPB Theory of Reasoned Action Rational Choice Continuous opinions, discrete actions (CODA) Consumat Framework Stages of Change …and many more Other options:
Energy Systems transformation From Deffuant et al Appendix: Relative Agreement Algorithm
AE Program Data + App. Status + Address + Date + System Specs COA Parcel Data + Home value + Address + Land Use + Sq. footage GIS of Parcels + Coordinates + DEM + Geometry + Tree cover Financial Model + Cash flows + Discount Rates Appendix: Data Streams UT Solar Survey + Sources of Info. + Decision-making Agent: Attitude Uncertainty Wealth Home sq. footage Age of home Network PP Discount rate Environment: Tree Cover Shade Electricity Price
Appendix: Model Design
Appendix: Seasonal Effects
Energy Systems transformation BatchmuEOHsLocals Relative Agreement Percent Locals AUC mu 20.5NoRadial1x90% YesContiguous4x90% YesRadial4x90% YesRadial3x90% YesRadial3x90% YesRadial3x80%0.682 Appendix: Key Batch Parameters