engleski

Application of ANN for management of energy storage systems with high share of renewable energy sources

The aim of this paper is to preset the possibilities of using Artificial Neural Network (ANN) method for smart management of energy storages in a energy system with high share of renewable energy sources and integrated electricity, heat and transport. Electricity and heat are integrated through ground source heat pumps, and transport is integrated with battery electric vehicles (BEV's) in smart charging mode. The overall method consists from the two separate methods: (1) finding optimal energy flows for training and testing of the ANN in EnergyPLAN, where geothermal reservoir is optimized as the weekly or seasonal heat storage and BEV's smart charge provides the demand response and (2) training, validating and testing the ANN in TensorFlow module. Input variables are related to intermittent and stochastic, but predictable wind and solar insolation, heat and electricity demand, as well as demand for BEV's. Output variables are charge and discharge of both thermal and BEV storage. BEV storage is not constant over time, but depends on time of day. System also has a biomass cogeneration unit. Exchange of energy with the centralized system is only through the DSO and minimum exchange with the DSO is a goal function for all scenarios. Few configurations of ANN's have been tested, with different additional signals, number of layers, time series window size, activation function etc. Preliminary results show moderate-to-good success with few combinations of ANN features and R2 values of 0.8. Further investigation should lead to better performance in terms of testing and validation against the input data from EnergyPLAN.

Artificial Neural Network, EnergyPLAN, Smart energy, TensorFlow

SDEWES2021.0982