Demand response potential evaluation based on feature fusion with expert knowledge and multi-image

Abstract

Potential evaluation to assist demand response decisions has garnered significant attention with the development of new power systems. However, existing data-driven methods are challenging to properly exploit multivariate features and the process of response potential evaluation is unclear. Therefore, the authors propose an evaluation method that fuses expert features with multi-image inputs and analyses the model evaluation process based on gradient. First, typical load profiles are extracted by the proposed procedure. Next, features derived from expert knowledge are calculated from the perspectives of adjustability, regularity, and sensitivity of electricity usage. Additionally, the typical load profile's recurrence plot, Markov leapfrog field, and Gramian angle field are created and incorporated into the colourful image as inputs. Then, the evaluation results are obtained by a two-stream neural network fusing multivariate features. In the experiments, the proposed method is validated and discussed by comparing with many existing methods using London household users' data under the time-of-use price, providing new insights for demand response potential evaluation.