基于FreTS-iTransformer-DTW的电力系统净负荷短期预测

打开文本图片集

中图分类号：TM715 文献标识码：A

Abstract： For forecasting net load composed of various uncertain variables such as aggregate load， wind power，and solar power，traditional forecasting models can no longer achieve effective accuracy. This paper proposes a short-term multivariate net load forecasting method combining a frequency-domain time series network with iTransformer. The frequency domain sequence network performs multi-objective feature extraction and fusion on input data across temporal and variable dimensions. Leveraging the multi-head self-attention mechanism within the iTransformer encoder，it calculates the self-attention representation of input feature vectors to output predicted determination points. Finally， dynamic time warping （DTW） is applied to compare global DTW distances and optimal alignment paths across models，evaluating the proposed model's capability to address random fluctuations in net load. Experiments demonstrate that this model outperforms others in both stability and prediction accuracy，validating its robust resilience to randomness and volatility in net load.

Keywords：volatility；net load forecasting;FreTS;iTransformer;DTW

随着全球能源向可持续转型，风能及太阳能等可再生能源在新型电力系统中占比提升[1-2]，但因其出力不确定及间歇性强等问题，增加了电网运行调度难度[3]。（剩余11122字）