Comparison of electricity savings in community units through ESS and PV generation using ANN-based prediction model under Korean climatic conditions

Abstract

Electrical energy saving was evaluated by taking advantage of PV and ESS in a community unit. An artificial neural network (ANN) and long short-term memory (LSTM) were employed to create a predictive model for PV generation. Annual demand data for residential buildings were estimated using EnergyPlus, while data for other buildings were collected from measurements in J Energy Town, Republic of Korea. Pearson correlation coefficients identified six crucial variables for the model. Comparative analysis of 310 cases revealed that the best-performing model was an ANN with three hidden layers and nodes of 14, 13 and 11. The model satisfied ASHRAE guidelines with a CV(RMSE) of 29.1 % and NMBE of −7.14 %. Evaluating electricity consumption in the community, case B (PV generation) showed a significant 46.3 % reduction compared to case A, while case D achieved a 5 % energy savings relative to case E over the year.