An Online Learning Algorithm for Ultra-Short-Term Load Forecasting in Real-Time Electricity Spot Market Based on Deep Extreme Learning Machine

In the context of the deregulated electricity market, ultra-short-term load forecasting is crucial for market pricing and trading. Accurate forecasting outputs effectively aid market participants in making rational bidding and purchasing decisions. In our paper, an advanced online learning algorithm is introduced for ultra-short-term load forecasting under the background of the real-time electricity spot market, by leveraging an online deep extreme learning machine (DELM) and a heuristic algorithm. Firstly, the abnormal data type in the smart grid is considerably classified into several typical scenarios, and a Fourier residual sequence is deployed to restore incorrect data to the original form. Additionally, DELM is employed as the core online training algorithm to map the relationship between input features and forecasting output by feeding newly generated data. Subsequent optimization of the model hyper parameter is achieved through the lion swarm optimization (LSO) algorithm, which effectively improves the training efficiency and generalization of DELM. The case study shows the superiority of the LSO-DELM over traditional machine learning models on the real-world data in the electricity spot market. The integration of these advanced methodologies significantly enhances the precision and efficiency of the load forecasting task; participants in the electricity spot market are empowered to optimize resource allocation and minimize operational costs.