A Day‐Ahead Aggregated Demand Response Carbon Reduction Strategy Using a Bidirectional LSTM Autoencoder for Residential Load Resources

Residential electricity consumption has become an important source of carbon emissions and a key concern in the low‐carbon energy transition. To address the limitations of existing methods in temporal modeling and large‐scale residential demand scheduling, this paper proposes a day‐ahead aggregated demand‐response carbon reduction strategy based on a Bidirectional LSTM Autoencoder (BiLSTM‐AE). The proposed strategy captures bidirectional temporal dependencies and nonlinear dynamic patterns in residential load sequences. The learned features are combined with clustering and class‐level dispatch to characterize user consumption behavior and assess carbon‐reduction potential. Simulation results based on residential load data from an anonymized urban distribution‐network area in Southwest China show that the proposed BiLSTM‐AE method achieves superior clustering performance compared with the comparative models and, on this basis, more accurately identifies high‐response classes within the response windows, improving the average unit response capacity by 11.35% relative to the traditional grid‐based model. These results verify the effectiveness of the proposed strategy for residential demand‐response aggregation and high‐response class identification.