Forecasting Short-Term Electricity Prices via Adaptive Noise-Based Complete Set Empirical Mode Decomposition and Multi-Model Combination

Accurate hourly electricity price forecast is very important for power market participants. Aiming at the problem of non-stationarity, nonlinearity and multi-scale characteristics of electricity price series, this paper presents a hybrid forecasting framework incorporating complete ensemble empirical mode decomposition with adaptive noise. Employing adaptive noise-assisted complete ensemble empirical mode decomposition, the initial electricity price sequence is separated into multiple relatively stable intrinsic mode functions and a residual component. Subsequently, component-specific models are developed utilizing long short-term memory networks and seasonal auto regressive integrated moving average, tailored to the distinct features of each decomposed element. Finally, the final hourly electricity price is obtained by fusing the forecast results of each component.