Today, more efficient energy systems are necessary to address global climate change and the scarcity of fossil fuel supplies. Energy communities and consumer-shared ownership of renewable energy resources are essential cornerstones of the overall success of the energy transition. This article proposes a new management framework for sustainable energy communities to reduce the mismatch between internal power generation and demand, the so-called imbalance. Proposals on this subject have already been made in the literature, primarily based on static optimization models or multi-step procedures that employ machine learning algorithms. Based on contributions from both techniques, an innovative decision-making model with a first forecasting phase and a second optimization phase is provided. It relies on deep neural networks tailored to forecast the generation and consumption profiles. Then, the energy dispatch efficiency is improved by employing a distributed optimization model to address issues in the traditional centralized framework, including high communication requirements, substantial computational burden, and limited scalability. With respect to previous studies by the authors on the same matter, the methodology’s effectiveness is improved here in both the prediction and optimization stages, using more robust and accurate neural networks that enable a more fault-tolerant distributed architecture and increased scalability of the overall management system. To test the accuracy and robustness, the proposed procedure was applied to real measured data from the energy cluster at the University of California, San Diego (USA). The performed tests assess a percentage reduction in the imbalance between the original and optimized management systems, ranging from 40% to 50%, with an increase compared to previous published methodologies of more than 10%. • We propose an innovative and comprehensive decision-making scheme designed to enhance the efficiency of an energy community. The scheme relies on an optimization model that enables optimum coordination among multiple distributed energy resources for the energy dispatch within the energy community. The model obtains the required data from a group of neural networks (specifically, LSTM predictors) to predict the future behavior of grid actors (e.g., solar power generation and power consumption). • The decision-maker performs an overall two-step prediction-decision process for the activities of the days ahead, providing optimized scheduling that will serve as the basis for the energy dispatch for the next day. The coordination among multiple distributed energy resources and the selection of appropriate demand-side management programs are optimal in the sense that they ensure the supply–demand balance, reducing the imbalance to a prescribed power dispatch profile. • The main contributions and novelties of the presented research are: 1. The energy dispatch performance is improved by a decision (optimization) model that considers the requirements of all the agents in the local grid. 2. The optimization model relies on a distributed framework to overcome the issues of the traditional centralized framework, which is subject to performance limitations such as a single point of failure, high communication requirements, substantial computation burden, and limited flexibility and scalability. 3. We propose a network of improved local predictors based on LSTM deep neural networks to forecast the generation and load profiles locally. 4. To test accuracy and robustness, we apply the proposed approach to the real energy cluster of the University of California in San Diego (USA) with real measured data over two years (2018–2019).
Rosato et al. (Sun,) studied this question.