Adaptive Learning Control for Smart Local Energy Community

Abstract

This paper presents a novel Adaptive Learning Control (ALC) approach for optimizing energy management in Smart Local Energy Communities (LECs). The framework combines advanced Reinforcement Learning (RL) with Distributed Model Predictive Control (DMPC), enabling real-time adaptation to fluctuations in energy demand, generation, and pricing. A key feature of this method is the use of data-driven algorithms to handle uncertainties in renewable energy sources, such as generation. While optimizing energy dispatch among prosumers. The energy management system employs an on-line trained neural network with stochastic weights that are continuously trained and updated in real-time using incoming data such as load, renewable output, and electricity prices, which are then used to optimize control actions. Additionally, a Hyperparameter Optimization (HPO) was integrated to ensure a balance between the rapid learning rate and the model stability. Decision making across LECs enhances overall system efficiency and resilience. Simulations and case studies demonstrate that this approach improves energy demand predictions and robust self learning, outperforming the conventional RL method. It maintains stable grid performance under dynamic, stochastic conditions and significantly advances self learning. A comparison between the robust prediction capabilities of ALC-EMS versus the conventional RL-EMS approach highlights the benefits of this innovative method.