LSTM-RNN 기반 태양광 발전량 추정을 통한 고속도로 주변부 태양광발전 시설의 적지 선별 기술

초록

The selection of suitable PV sites is critical for high electricity generation. However, available solar energy rather than power generation has often been estimated to evaluate potential PV sites due to a lack of available power data. Thus, this study proposes the use of PV power data collected from existing plants for training and in turn for predicting the power outputs in other sites, which were not used for training. Particularly, this study investigates slopes of national highway network by using GIS; in this way, available land for PV systems can further be secured. The data include the power outputs from 172 existing plants as well as weather conditions, collected monthly. The temporal patterns in the time series data are learned using the long short-term memory(LSTM) recurrent neural network(RNN) model and then applied to highway slopes extracted from GIS layers. As a result, the proposed model shows a MAPE of 10.873(%) and R² of 0.726. Then, the top ten sites on slopes are identified and evaluated as an example. In this regard, this study may explore a computational approach that can predict the amount of potential power generation using existing data and search for suitable locations of PV facilities using GIS data.

키워드