Ultra-short-term Photovoltaic Power Generation Prediction Based on Transformer Encoder
Aiming at the problem that existing photovoltaic power generation power prediction methods are insufficient in exploring the intrinsic coupling and correlation characteristics among meteorological data features,the ultra-short-term photovoltaic power generation prediction method is proposed based on the Transformer encoder.Firstly,the input matrix is reconstructed using historical and numerical weather forecast data.Then the input embedding is generated with multi-layer perceptron or one-dimensional convolution filter.And temporal information embedding and temporal feature embedding are added to the input.Next,the multi-head self-attention mechanism is applied to automatically mine the intrinsic coupling relationship among data features.Finally,the power prediction sequence is produced through the decoding layer.The numerical results demonstrate that the proposed method has higher prediction accuracy and better robustness for ultra-short-term photovoltaic power generation.
photovoltaic power generation predictionself-attention mechanismfeature embedding
NETL
NSTL
万方数据
