EMD-LSTM-AM PV power prediction model

This paper proposes a hybrid EMD - LSTM - AM model based on digital twins for photovoltaic power prediction.First, high - precision photovoltaic power and weather data are collected and preprocessed via a digital twin system. Then, EMD decomposes the complex power time series into multiple IMFs, with representative ones selected for model input. Next, a two - layer LSTM network processes each IMF to extract time - series features, and AM weights the hidden states of the LSTM to highlight key information. Finally, a fully connected layer maps the weighted features to sub - power predictions for each IMF, and the results are post - processed to yield the final power prediction. Model parameters are set using the TensorFlow framework. Experiments compare the proposed model with RNN, LSTM, EMD - LSTM, and LSTM - AM models, demonstrating the EMD - LSTM - AM model’s superior predictive accuracy.