Multi-Agent AI System for Coordinated Dispatch of Renewable Energy and Storage in Islanded Microgrids

Abstract

The growing implementation of renewable energy resources in power systems has brought with it operational complexity especially in microgrids that are islanded and require energy balancing in real time and autonomously. Solar and wind sources that are renewable are variable and intermittent, which means that a balance will not be maintained between supply and demand unless complex control measures are applied. The conventional centralized method of dispatching is not flexible and adaptable to create efficient management of distributed and dynamic microgrids environments. In response to these issues, the Multi-Agent Artificial Intelligence (AI) system suggested in this research aims at coordinating renewable energy generation and storage using the coordinated dispatch of renewable energy generation and storage in islanded microgrids. The suggested system is based on a decentralized structure, according to which every energy resource, including solar PV, energy storage, and controllable loads, is controlled by an intelligent agent. These agents are independent and transparent to each other, and yet they are all inquisitive to imply and coordinate their efforts at locally distributed goals like energy efficiency, cost savings and load balancing at a system-wide level. Among the machine learning techniques that are incorporated in this paper are Recurrent Neural Networks (RNN) and Gradient Boosting Regression (GBR), which are used to gum long into the future or give short-term energy generation and consumption forecasting using historical and real-time weather information. The forecasting module enables the agents to gain a foresight to system dynamics and make pro-active dispatch decisions so that the take-off on fossil-fuel backup is minimized, and system reliability is maximized. The system validation is performed empirically on two open datasets: Spain Energy Demand and Generation and Liege Microgrid datasets. The data sets used in these are of a high-resolution weather, load, and generation which make real world simulation easy. The suggested multi-Agent AI structure is superior to the traditional control strategies as it is responsive, scalable, and coherent in its functioning. The study shows the promise of integrating AI on prediction and decentralized control to develop more intelligent, more autonomous microgrid systems that will work in more isolated and renewably self-reliant systems.