Electricity load forecasting is an important task in ensuring safety and efficiency of power system. Aiming at the pivotal issue of short-term electricity load forecasting in distribution areas, this paper presents a novel approach that combines electrical feature data processing and Informer model optimization. The study employs discrete wavelet transform (DWT) for denoising current data while utilizing Prophet model for extracting temporal features to enhance input data quality. Additionally, the method incorporates ProbSparse self-attention mechanism and self-attention distillation, thereby bolstering feature capture and prediction speed within the model. The model extracted by DWT and Prophet features is superior to the original model under the same indices. Validation with instance data demonstrates that the DWT-Informer model, enriched through both data preprocessing and model optimization, outperforms the baseline model across various performance metrics.