Machine learning combined with spatio-temporal analysis technics for battery sizing and allocation in electrical distribution systems

Abstract

The evolution of energy storage system (ESS) technologies has increased the interest in their integration into the electric distribution system (EDS). Despite the benefits they can provide to the EDS, ESSs are still expensive, and their allocation and sizing decisions are complex due to uncertainties and long-term EDS aspects, which require the representation of many scenarios in the decision process increasing the computational effort; therefore, the ESS planning should be carefully evaluated. In addition, the high penetration of renewable distributed generation (DG) units and the expanding integration of electric vehicles intensifies the complexity of the ESS planning problem. Several methods have been applied to address this problem, although socio-economic and geographic factors have not been properly assessed and their inclusion would lead to more adequate decisions for EDS planning; therefore, it is proposed to develop a machine learning algorithm combined with spatial-temporal analysis techniques to properly address those factors to the EDS planning. The machine learning approach for spatial-temporal analysis should estimate the input data of the problem (renewable DG integration and energy consumption through time) using geographic data and socio-economic factors to create a suitable representation of the real behaviors of an EDS; also, an optimization method should be implemented to solve the problem considering the complexity, the number of scenarios and variables of the problem. The research will be conducted at the Delft University of Technology in Netherlands; it should result in a decision support method useful for EDS planning. (AU)