Combining Machine Learning Analysis and Incentive-Based Genetic Algorithms to Optimise Energy District Renewable Self-Consumption in Demand-Response Programs

Abstract

The recent rise of renewable energy sources connected to the distribution networks and the high peak consumptions requested by electric vehicle-charging bring new challenges for network operators. To operate smart electricity grids, cooperation between grid-owned and third-party assets becomes crucial. In this paper, we propose a methodology that combines machine learning with multi-objective optimization to accurately plan the exploitation of the energy district’s flexibility with the objective of reducing peak consumption and avoiding reverse power flow. Using historical data, acquired by the smart meters deployed on the pilot district, the district’s power profile can be predicted daily and analyzed to identify potentially critical issues on the network. District’s resources, such as electric vehicles, charging stations, photovoltaic panels, buildings energy management systems, and energy storage systems, have been modeled by taking into account their operational constraints and the multi-objective optimization has been adopted to identify the usage pattern that better suits the distribution operator’s (DSO) needs. The district is subject to incentives and penalties based on its ability to respond to the DSO request. Analysis of the results shows that this methodology can lead to a substantial reduction of both the reverse power flow and peak consumption.