Current short-term load forecasting models exhibit deficiencies in the dynamic fusion mechanism for electricity prices and load

as well as in load data denoising and temporal feature extraction. This paper proposes a forecasting framework integrating multi-dimensional electricity prices and meteorological features. Firstly

a hybrid denoising decomposition module is constructed by combining stacked denoising autoencoders (SDAE) and ensemble empirical mode decomposition (EEMD) to effectively suppress noise interference and mode mixing problems in the load sequence. EEMD decomposes the denoised load sequence into intrinsic mode functions (IMFs) components. Secondly

based on maximum information coefficient (MIC) analysis

electricity price and meteorological features are fused into high-frequency and low-frequency IMFs

achieving differentiated feature integration. On this basis

a crossover prediction strategy is proposed. For high-frequency components

an improved informer model with global timestamp coding and sparse attention mechanism is introduced to capture the characteristics of short-term violent fluctuations. For the low-frequency components

the bi-directional long short-term memory (BiLSTM) network is used to capture the long-term trend and periodicity. Finally

results using the Australian NEM dataset demonstrate that the model significantly outperforms comparative models lacking electricity price features or the frequency-divided strategy on MAPE and MSE metrics. The framework enhances accuracy and stability

providing reliable technical support for forecasting under dynamic pricing.