Feeder-Level Deep Learning-based Photovoltaic Penetration Estimation Scheme. / Zhang, Xiaoyu; Kuenzel, Stefanie; Watkins, Chris.
2020. Paper presented at 12th IEEE PES Asia-Pacific Power and Energy Engineering Conference, Nanjing, China.Research output: Contribution to conference › Paper
Feeder-Level Deep Learning-based Photovoltaic Penetration Estimation Scheme. / Zhang, Xiaoyu; Kuenzel, Stefanie; Watkins, Chris.
2020. Paper presented at 12th IEEE PES Asia-Pacific Power and Energy Engineering Conference, Nanjing, China.Research output: Contribution to conference › Paper
}
TY - CONF
T1 - Feeder-Level Deep Learning-based Photovoltaic Penetration Estimation Scheme
AU - Zhang, Xiaoyu
AU - Kuenzel, Stefanie
AU - Watkins, Chris
PY - 2020/10/13
Y1 - 2020/10/13
N2 - The increasing penetration of renewable energy to the distribution grids, especially photovoltaic (PV), helps smooth out supply and demand, and reduces greenhouse gas emissions. However, the PV generation is behind-the-meter, and cannot be detected by the smart meter. To address this problem, a hybrid regression multi-layer perceptron (MLP) deep neural network (DNN) model is designed to separate the PV generation from the overall grid measurements. The model utilizes grid measurements, weather-related measurements, satellite-driven irradiance measurements, and temporal information as inputs to evaluate the PV generation in real-time. We also examine the performance of the model with different levels of PV penetration. We show the proposed model reduces the mean square error by 49% compared to single variable input models.
AB - The increasing penetration of renewable energy to the distribution grids, especially photovoltaic (PV), helps smooth out supply and demand, and reduces greenhouse gas emissions. However, the PV generation is behind-the-meter, and cannot be detected by the smart meter. To address this problem, a hybrid regression multi-layer perceptron (MLP) deep neural network (DNN) model is designed to separate the PV generation from the overall grid measurements. The model utilizes grid measurements, weather-related measurements, satellite-driven irradiance measurements, and temporal information as inputs to evaluate the PV generation in real-time. We also examine the performance of the model with different levels of PV penetration. We show the proposed model reduces the mean square error by 49% compared to single variable input models.
KW - deep neural network
KW - photovoltaic penetration
KW - energy disaggregation
U2 - 10.1109/APPEEC48164.2020.9220536
DO - 10.1109/APPEEC48164.2020.9220536
M3 - Paper
T2 - 12th IEEE PES Asia-Pacific Power and Energy Engineering Conference
Y2 - 20 September 2020 through 23 September 2020
ER -