Dependency-based FlexOffers

Abstract

Smart grid actors such as aggregators need scalable yet simple and powerful ways to aggregate, optimize, and disaggregate large collections of flexible loads (e.g., from heat-pumps and electric vehicles) based on models of flexible loads, e.g., state-space models. Based on system- and user-specific variables and constraints, e.g., power or temperature bounds, such models specify dependencies between system inputs, states, and energy amounts consumed/produced at discrete time intervals. Traditional approaches, using exact or simple approximate models, do not scale well, introduce errors, or unacceptably reduce the flexibility (solution space) when total energy needs to be optimized for many time intervals while respecting a large number of model constraints. To mitigate these problems, we propose the so-called dependency-based flexoffer (DFO) -- a low-complexity generalized model that allows efficiently approximating various exact models of both individual and aggregated loads while retaining most of the flexibility. We propose algorithms for generating DFOs as inner and outer approximations of the exact models. Additionally, we provide efficient algorithms for aggregating DFO instances and disaggregating energy series while respecting all DFO constraints and ensuring energy balance. An extensive experimental evaluation with thermostatic (heat-pump) and storage-like (battery) loads shows that DFOs offer a good trade-off between performance and flexibility when a large number of flexible loads need to be aggregated and/or optimized.