Fusion of deep learning and machine learning methods for hourly locational marginal price forecast in power systems

In this paper, we propose STPLF, which stands for the short-term forecasting of locational marginal price components, including the forecasting of non-conforming hourly net loads. The volatility of transmission-level hourly locational marginal prices (LMPs) is caused by several factors, including weather data, hourly gas prices, historical hourly loads, and market prices. In addition, variations of non-conforming net loads, which are affected by behind-the-meter distributed energy resources (DERs) and retail customer loads, could have a major impact on the volatility of hourly LMPs, as bulk grid operators have limited visibility of such retail-level resources. We propose a fusion forecasting model for the STPLF, which uses machine learning and deep learning methods to forecast non-conforming loads and respective hourly prices. Additionally, data preprocessing and feature extraction are used to increase the accuracy of the STPLF. The proposed STPLF model also includes a post-processing stage for calculating the probability of hourly LMP spikes. We use a practical set of data to analyze the STPLF results and validate the proposed probabilistic method for calculating the LMP spikes.