Dynamic Incentive Allocation for City-scale Deep Decarbonization

TL;DR

This paper tackles the inefficiency and inequity of residential decarbonization incentives by formulating a city-scale, emissions-based optimization that dynamically allocates a fixed budget to households to maximize carbon reductions. It combines an offline contextual bandit survey phase with a subsequent city-wide offering stage (Contextual Lower Confidence Bound) to learn and apply household responsiveness to incentives, while accommodating equity constraints. The approach yields up to 32.23% higher carbon reductions than status-quo incentives and attains an average of 78.84% of the optimal carbon reduction under equity considerations, robustness to changing gas/electric prices, and grid carbon intensity. This data-driven framework demonstrates practical potential for targeted, equitable, and scalable decarbonization policy design, with broad applicability to varied urban contexts and price scenarios.

Abstract

Greenhouse gas emissions from the residential sector represent a significant fraction of global emissions. Governments and utilities have designed incentives to stimulate the adoption of decarbonization technologies such as rooftop PV and heat pumps. However, studies have shown that many of these incentives are inefficient since a substantial fraction of spending does not actually promote adoption, and incentives are not equitably distributed across socioeconomic groups. We present a novel data-driven approach that adopts a holistic, emissions-based and city-scale perspective on decarbonization. We propose an optimization model that dynamically allocates a total incentive budget to households to directly maximize city-wide carbon reduction. We leverage techniques for the multi-armed bandits problem to estimate human factors, such as a household's willingness to adopt new technologies given a certain incentive. We apply our proposed framework to a city in the Northeast U.S., using real household energy data, grid carbon intensity data, and future price scenarios. We show that our learning-based technique significantly outperforms an example status quo incentive scheme, achieving up to 32.23% higher carbon reductions. We show that our framework can accommodate equity-aware constraints to equitably allocate incentives across socioeconomic groups, achieving 78.84% of the carbon reductions of the optimal solution on average.

Figures (9)

• Figure 1: A representative break-even point illustration under the NetBenefit model with a 5, 10 and 15 year payback period; 96% of houses fail to break even with the installation costs of decarbonization within 5 year payback period, 70% of houses fail to break even within 10 year payback period, and 33% of houses fail to break even within a 15 year payback period.
• Figure 2: Considering the NetBenefit cost model with a modest discount rate of 5% for payback periods of (a) 5 years, (b) 10 years and (c) 15 years. Our learning-based incentive model achieves an average of 17.71% higher carbon reduction than the status quo across all budgets and payback periods.
• Figure 3: Considering the NetBenefit Cost Model with a discount rate of 2% for payback periods of (a) 5 years, (b) 10 years and (c) 15 years. Our learning-based incentive model average of 15.22% higher carbon reduction than the status quo across all budgets and payback periods
• Figure 4: The effect of changing environments characterized by (a) gas price, (b) electricity price, and (c) carbon intensity.
• Figure 5: The allocation of incentive by the Bandits Incentive Allocation model to high, medium, and low income homes for a few total budget values.