Demand Side Flexibility Envelope Quantification Under Data Scarcity

Efficient management of distributed energy resources often requires a tremendous amount of data to model component constraints and consumer behaviors. However, in many practical scenarios, collecting high-resolution demand data is significant costly, which jeopardizes the efficient utilization of demand-side resources. This paper proposes a novel learning framework to quantify demand-side flexibility envelope under data scarcity. We construct a long-short-term memory network to capture the relationship between consumers’ load consumption behavior and demand flexibility envelope. The network only uses low-resolution load data to predict consumers’ personalized flexibility envelope and able to extrapolate across various stakeholders and conditions. We provide a flexibility envelope generation algorithm to generate the training label and expedites the training process with limited high-resolution historical load data and environmental data. Simulations using Pecan Street consumer data demonstrate that our method generate effective demand flexibility envelopes for network training, and the cross-validation results shows our learning framework accurately predict the demand flexibility envelope despite data scarcity.