Dynamic Chromosome Interpretation in Evolutionary Algorithms for Distributed Energy Resources Scheduling

Integrating renewable generation into the existing electricity grid to reduce Greenhouse Gas (GHG) emissions involves several challenges. These include, e.g., volatile generation and demand, and can be overcome by increasing flexibility in the grid. One possibility to provide this flexibility is the optimized scheduling of Distributed Energy Resources (DERs). Such a scheduling task requires a powerful optimization algorithm, such as Evolutionary Algorithms (EAs). However, EAs can produce poor solution quality w.r.t. solution time when solving complex and large scale scheduling tasks of DERs. Hence, in our work, a concept for improving the EA optimization process for scheduling DERs is presented and evaluated. In this concept, Machine Learning (ML) algorithms learn from already found solutions to predict the optimization quality in advance. By this, the computational effort of the EA is directed to particularly difficult areas of the search space. This is achieved by dynamic interpretation and consequent interval length assignment of the solutions proposed by the EA. We evaluate our approach by comparing two experiments and show that our novel concept leads to a significant increase of the evaluated fitness by up to 9.4%.