A long-sequence photovoltaic power generation prediction model based on improved Informer

With the promotion and application of photovoltaic power generation technology, it is necessary to accurately predict the photovoltaic power generation. Considering the inherent characteristics of photovoltaic power generation and the comprehensive demands of power market operation, long-term sequence prediction is required. However, when participating in a distributed photovoltaic power generation project with wide distribution in a domestic oilfield, the prediction results were not satisfactory when applying some mainstream prediction models. This is because these prediction models lack the ability to capture the rich frequency information in the data set. To address the challenges faced by long-term sequence photovoltaic power generation prediction, such as the complexity of long-term trends and periodic changes . In the data, long-range dependencies, and model generalization ability, this paper proposes an improved self-attention mechanism model based on temporal-frequency enhanced channels (Temporal-Frequency Enhanced Channel Informer, TFECInformer), and proposes a novel relative position encoding algorithm for time series data to optimize the calculation method of the Informer attention mechanism. The TFECInformer model is used for prediction and compared with the traditional Lstm model and Transformer model. The experimental results show that this model fits the prediction task well.