PV Power Prediction Based on LSTM in a Federated Learning Architecture

In this paper, we propose a Photovoltaic(PV) power prediction method based on LSTM in a federated learning architecture. The local PV prediction model uses a long short-term memory (LSTM) neural network. Multi-region PV data are jointly involved in the construction of PV prediction models, and the sliding window prediction mechanism is used to generate the prediction results under the federal architecture, which completes the continuous output of the predicted value of PV power generation. The proposed approach can significantly improves the prediction accuracy.