An ensemble learning driven improved transformer method for time series forecasting

An intelligent forecasting model based on ensemble learning， named iTransformer-XGBoost， is proposed in this paper to address the issue of precision limitations of traditional time series prediction models. Firstly， the Pearson correlation coefficient is employed to select the key features affecting time series data and construct an optimized input dataset. Then， the iTransformer model is utilized to capture long-term dependencies within the time series and generate preliminary prediction results. Meanwhile， the XGBoost algorithm is introduced to achieve nonlinear modeling of the time series data. Finally， a threshold-based combination strategy is applied to the prediction results fusion of iTransformer and XGBoost， thereby determining the integrated output and improving overall forecasting performance. The model is validated by using photovoltaic power-related data， and the experimental results demonstrate that the proposed approach achieves higher accuracy and stability in time series forecasting compared with traditional methods. Furthermore， it shows great potential for application in aerospace photovoltaic scenarios.