Short-term Prediction Method of Distributed Photovoltaic Power Generation Based on LSTM Network

High-precision prediction of distributed photovoltaic power is crucial for the secure and stable operation of new power systems. The conventional prediction models based on forward neural networks or regression analyses lack historical memory, which result in low robustness and weak adaptive ability. Aiming at this issue, this paper proposes a short-term prediction method of distributed PV power generation based on the long short-term memory (LSTM) network. Firstly, this paper employs the quartile method to identify abnormal values in the dataset, and utilizes the Lagrangian interpolation method to supplement and modify them. Subsequently, a PV power prediction model based on LSTM network is developed and compared with the prediction results of random forest (RF) and multi-layer perceptron (MLP) regression models. The simulation results demonstrate that the LSTM network-based distributed PV power short-term power prediction model could effectively reflect the dynamic characteristics of the time-series data with a high level of accuracy. The prediction results provide promising data support to the power dispatch department.