A Machine Learning-Based Hybrid Deep Neural Network Approach for Adaptive Demand Side Management Using Community-Level Battery Storage Systems

This study aims to reduce peak demand within a community of residential and commercial consumers by optimizing the operation of a community-level Battery Storage System (BSS). A machine learning approach has been used to make real-time decisions on charging and discharging the BSS. The proposed method employs a hybrid neural network architecture consisting of a Convolutional Neural Network (CNN) for feature extraction, a Long Short-Term Memory (LSTM) network for capturing temporal patterns and dependencies, and a Deep Neural Network (DNN) along with a hybrid semi-supervised kind of learning approach for making operational decisions. The model is trained using a labeled dataset of power demand data for 1000 houses, along with the corresponding BSS power transactions. The sampling frequency of the dataset is 5-min. The output includes both the charge/discharge decision and the amount of power involved, ensuring real-time optimization of the Demand Side Management process. Simulation results and hardware implementation show significant potential in reducing peak demand and optimizing energy costs, contributing to a more sustainable and reliable energy distribution system.