Physical model and long short‑term memory‑based combined prediction of photovoltaic power generation

Yaoyu Wu; Jing Liu; Suhuan Li; Mingyue Jin

Physical model and long short‑term memory‑based combined prediction of photovoltaic power generation

Vol. 24, No. 7, pp. 1118-1128, Jul. 2024

10.1007/s43236-024-00782-9

PV power prediction

Solar radiation mechanism modeling

Oblique irradiance conversion

Long short-term memory networks

Abstract

Solar energy is clean and pollution free. However, the evident intermittency and volatility of illumination make power systems uncertain. Therefore, establishing a photovoltaic prediction model to enhance prediction precision is conducive to lessening the uncertainty of photovoltaic (PV) power generation and to ensuring the safe and stable operation of power grid scheduling. The radiation from the sun to the Earth has a certain regularity, which can be estimated under ideal weather conditions. However, the radiation is affected by climate, cloud cover, and other reasons. Therefore, this paper puts forward a PV prediction model combining a physical model and a neural network that can modify solar radiation in complex weather through the neural network to enhance the accuracy of PV power prediction. First, a solar radiation model (SRM) is established by using the solar radiation mechanism to estimate the sum radiation value on the horizontal plane. Then the slope radiation value received by the PV panel is calculated by the slope irradiance conversion method. Second, the main factors that greatly influence PV power are screened out by the Pearson method. The calculated slope radiation and the main influencing factors are taken as inputs. The long short-term memory network (LSTM) is selected to set up the SRM-LSTM PV power prediction method. The significance of the suggested method is verified by the true data from Alice Springs, Australia. The results show that when compared with the backpropagation (BP) prediction method, the MAE and RMSE were reduced by 22.18% and 33.89% under complex weather conditions, respectively. When compared with the LSTM prediction method, the MAE and RMSE were reduced by 15.99% and 21.73%, respectively. These results demonstrate high accuracy.

Statistics

Show / Hide Statistics

Cumulative Counts from September 30th, 2019
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.

Cite this article

[IEEE Style]

Y. Wu, J. Liu, S. Li, M. Jin, "Physical model and long short‑term memory‑based combined prediction of photovoltaic power generation," Journal of Power Electronics, vol. 24, no. 7, pp. 1118-1128, 2024. DOI: 10.1007/s43236-024-00782-9.

[ACM Style]

Yaoyu Wu, Jing Liu, Suhuan Li, and Mingyue Jin. 2024. Physical model and long short‑term memory‑based combined prediction of photovoltaic power generation. Journal of Power Electronics, 24, 7, (2024), 1118-1128. DOI: 10.1007/s43236-024-00782-9.