Short-term load forecasting method of IES based on RLA neural network with dual signal processing

Abstract

Significant fluctuations and strong nonlinearity in integrated energy system data pose challenges to the precision of multi-energy load forecasting. To address this issue, the present study proposes a residual network (ResNet) - long short-term memory (LSTM) with an attention neural network model founded on a dual signal processing architecture. Initially, the dual signal processing framework was introduced to overcome the limitations of classical signal decomposition, particularly its inability to effectively mitigate high-frequency noise. This approach moderates the nonstationary attributes of the initial data and furnishes comprehensive information for the subsequent predictive model. Subsequently, the study employs the ResNet model for spatial feature extraction and the LSTM model for temporal feature extraction to augment the efficiency and precision in identifying multi-energy load characteristics. An attention mechanism is further integrated to allocate weights to diverse load features, thereby optimizing data transfer processing. The experimental results confirm that the proposed model offers superior adaptability to the integrated energy systems environment and substantially minimizes prediction errors in comparison to alternative models.