본문 바로가기

CNN-LSTM Neural Network-Based Short-Term PV Power Generation Forecaster

문서 유형
연구출판물 · 학술대회 발표논문
등재
Scopus
피인용
Scopus 3
FWCI (Scopus)
0.99

저자

  1. 시티 누르파딜라 빈티 자이니제1저자교신저자Universiti Malaysia Sabah, Mechanical Engineering, Kota Kinabalu, Sabah, Malaysia
  2. 이득우Pusan National University, Department of Energy Convergence Technology, Busan, Rep. of Korea
  3. Heng, Choong WaiUniversiti Malaysia Sabah, Mechanical Engineering, Kota Kinabalu, Sabah, Malaysia

초록

Solar renewable energy technology is vastly applied across the globe to tackle consumer's energy demand. Power generation through solar energy is heavily influenced by the uncertainty of meteorological conditions and surrounding facilities. Establishing model predictive control (MPC) through hybrid deep learning (DL) forecaster is crucial to closely monitor and provide systematic planning for the solar power generation station. A hybrid deep learning through convolutional long-short-term memory (CNN-LSTM) model is proposed to forecast the hourly power generation. The proposed CNN-LSTM model excluded the usage of max pooling layer in its structure which is typically utilized in other studies. A comparison of the forecasting performance between the proposed CNN-LSTM and CNN-LSTM max pooling model is conducted throughout the study for forecasting training, validation, and testing datasets. Transfer learning by feeding new datasets from other stations is conducted to testify the ability of each model in adapting new datasets. Developing an accurate DL forecaster for MPC in solar power generation stations will substantially enhance the productivity of energy generation.

키워드

data pipeline generationhybrid deep learningregressionsolar power generationsupervised learning

식별자

Scopus EID
2-s2.0-85209652783